diff --git a/__init__.py b/__init__.py
index 43198c8..1e2aadb 100644
--- a/__init__.py
+++ b/__init__.py
@@ -1,4 +1,7 @@
-from .json_loader import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
+from .project_loader import PROJECT_NODE_CLASS_MAPPINGS, PROJECT_NODE_DISPLAY_NAME_MAPPINGS
+
+NODE_CLASS_MAPPINGS = PROJECT_NODE_CLASS_MAPPINGS
+NODE_DISPLAY_NAME_MAPPINGS = PROJECT_NODE_DISPLAY_NAME_MAPPINGS
 
 WEB_DIRECTORY = "./web"
 
diff --git a/api_routes.py b/api_routes.py
new file mode 100644
index 0000000..62f8512
--- /dev/null
+++ b/api_routes.py
@@ -0,0 +1,80 @@
+"""REST API endpoints for ComfyUI to query project data from SQLite.
+
+All endpoints are read-only. Mounted on the NiceGUI/FastAPI server.
+"""
+
+import logging
+from typing import Any
+
+from fastapi import HTTPException, Query
+from nicegui import app
+
+from db import ProjectDB
+
+logger = logging.getLogger(__name__)
+
+# The DB instance is set by register_api_routes()
+_db: ProjectDB | None = None
+
+
+def register_api_routes(db: ProjectDB) -> None:
+    """Register all REST API routes with the NiceGUI/FastAPI app."""
+    global _db
+    _db = db
+
+    app.add_api_route("/api/projects", _list_projects, methods=["GET"])
+    app.add_api_route("/api/projects/{name}/files", _list_files, methods=["GET"])
+    app.add_api_route("/api/projects/{name}/files/{file_name}/sequences", _list_sequences, methods=["GET"])
+    app.add_api_route("/api/projects/{name}/files/{file_name}/data", _get_data, methods=["GET"])
+    app.add_api_route("/api/projects/{name}/files/{file_name}/keys", _get_keys, methods=["GET"])
+
+
+def _get_db() -> ProjectDB:
+    if _db is None:
+        raise HTTPException(status_code=503, detail="Database not initialized")
+    return _db
+
+
+def _list_projects() -> dict[str, Any]:
+    db = _get_db()
+    projects = db.list_projects()
+    return {"projects": [p["name"] for p in projects]}
+
+
+def _list_files(name: str) -> dict[str, Any]:
+    db = _get_db()
+    files = db.list_project_files(name)
+    return {"files": [{"name": f["name"], "data_type": f["data_type"]} for f in files]}
+
+
+def _list_sequences(name: str, file_name: str) -> dict[str, Any]:
+    db = _get_db()
+    seqs = db.list_project_sequences(name, file_name)
+    return {"sequences": seqs}
+
+
+def _get_data(name: str, file_name: str, seq: int = Query(default=1)) -> dict[str, Any]:
+    db = _get_db()
+    proj = db.get_project(name)
+    if not proj:
+        raise HTTPException(status_code=404, detail=f"Project '{name}' not found")
+    df = db.get_data_file_by_names(name, file_name)
+    if not df:
+        raise HTTPException(status_code=404, detail=f"File '{file_name}' not found in project '{name}'")
+    data = db.get_sequence(df["id"], seq)
+    if data is None:
+        raise HTTPException(status_code=404, detail=f"Sequence {seq} not found")
+    return data
+
+
+def _get_keys(name: str, file_name: str, seq: int = Query(default=1)) -> dict[str, Any]:
+    db = _get_db()
+    proj = db.get_project(name)
+    if not proj:
+        raise HTTPException(status_code=404, detail=f"Project '{name}' not found")
+    df = db.get_data_file_by_names(name, file_name)
+    if not df:
+        raise HTTPException(status_code=404, detail=f"File '{file_name}' not found in project '{name}'")
+    keys, types = db.get_sequence_keys(df["id"], seq)
+    total = db.count_sequences(df["id"])
+    return {"keys": keys, "types": types, "total_sequences": total}
diff --git a/db.py b/db.py
new file mode 100644
index 0000000..e9088f9
--- /dev/null
+++ b/db.py
@@ -0,0 +1,349 @@
+import json
+import logging
+import sqlite3
+import time
+from pathlib import Path
+from typing import Any
+
+from utils import load_json, KEY_BATCH_DATA, KEY_HISTORY_TREE
+
+logger = logging.getLogger(__name__)
+
+DEFAULT_DB_PATH = Path.home() / ".comfyui_json_manager" / "projects.db"
+
+SCHEMA_SQL = """
+CREATE TABLE IF NOT EXISTS projects (
+    id          INTEGER PRIMARY KEY AUTOINCREMENT,
+    name        TEXT NOT NULL UNIQUE,
+    folder_path TEXT NOT NULL,
+    description TEXT NOT NULL DEFAULT '',
+    created_at  REAL NOT NULL,
+    updated_at  REAL NOT NULL
+);
+
+CREATE TABLE IF NOT EXISTS data_files (
+    id          INTEGER PRIMARY KEY AUTOINCREMENT,
+    project_id  INTEGER NOT NULL REFERENCES projects(id) ON DELETE CASCADE,
+    name        TEXT NOT NULL,
+    data_type   TEXT NOT NULL DEFAULT 'generic',
+    top_level   TEXT NOT NULL DEFAULT '{}',
+    created_at  REAL NOT NULL,
+    updated_at  REAL NOT NULL,
+    UNIQUE(project_id, name)
+);
+
+CREATE TABLE IF NOT EXISTS sequences (
+    id              INTEGER PRIMARY KEY AUTOINCREMENT,
+    data_file_id    INTEGER NOT NULL REFERENCES data_files(id) ON DELETE CASCADE,
+    sequence_number INTEGER NOT NULL,
+    data            TEXT NOT NULL DEFAULT '{}',
+    updated_at      REAL NOT NULL,
+    UNIQUE(data_file_id, sequence_number)
+);
+
+CREATE TABLE IF NOT EXISTS history_trees (
+    id            INTEGER PRIMARY KEY AUTOINCREMENT,
+    data_file_id  INTEGER NOT NULL UNIQUE REFERENCES data_files(id) ON DELETE CASCADE,
+    tree_data     TEXT NOT NULL DEFAULT '{}',
+    updated_at    REAL NOT NULL
+);
+"""
+
+
+class ProjectDB:
+    """SQLite database for project-based data management."""
+
+    def __init__(self, db_path: str | Path | None = None):
+        self.db_path = Path(db_path) if db_path else DEFAULT_DB_PATH
+        self.db_path.parent.mkdir(parents=True, exist_ok=True)
+        self.conn = sqlite3.connect(
+            str(self.db_path),
+            check_same_thread=False,
+            isolation_level=None,  # autocommit — explicit BEGIN/COMMIT only
+        )
+        self.conn.row_factory = sqlite3.Row
+        self.conn.execute("PRAGMA journal_mode=WAL")
+        self.conn.execute("PRAGMA foreign_keys=ON")
+        self.conn.executescript(SCHEMA_SQL)
+
+    def close(self):
+        self.conn.close()
+
+    # ------------------------------------------------------------------
+    # Projects CRUD
+    # ------------------------------------------------------------------
+
+    def create_project(self, name: str, folder_path: str, description: str = "") -> int:
+        now = time.time()
+        cur = self.conn.execute(
+            "INSERT INTO projects (name, folder_path, description, created_at, updated_at) "
+            "VALUES (?, ?, ?, ?, ?)",
+            (name, folder_path, description, now, now),
+        )
+        self.conn.commit()
+        return cur.lastrowid
+
+    def list_projects(self) -> list[dict]:
+        rows = self.conn.execute(
+            "SELECT id, name, folder_path, description, created_at, updated_at "
+            "FROM projects ORDER BY name"
+        ).fetchall()
+        return [dict(r) for r in rows]
+
+    def get_project(self, name: str) -> dict | None:
+        row = self.conn.execute(
+            "SELECT id, name, folder_path, description, created_at, updated_at "
+            "FROM projects WHERE name = ?",
+            (name,),
+        ).fetchone()
+        return dict(row) if row else None
+
+    def delete_project(self, name: str) -> bool:
+        cur = self.conn.execute("DELETE FROM projects WHERE name = ?", (name,))
+        self.conn.commit()
+        return cur.rowcount > 0
+
+    # ------------------------------------------------------------------
+    # Data files
+    # ------------------------------------------------------------------
+
+    def create_data_file(
+        self, project_id: int, name: str, data_type: str = "generic", top_level: dict | None = None
+    ) -> int:
+        now = time.time()
+        tl = json.dumps(top_level or {})
+        cur = self.conn.execute(
+            "INSERT INTO data_files (project_id, name, data_type, top_level, created_at, updated_at) "
+            "VALUES (?, ?, ?, ?, ?, ?)",
+            (project_id, name, data_type, tl, now, now),
+        )
+        self.conn.commit()
+        return cur.lastrowid
+
+    def list_data_files(self, project_id: int) -> list[dict]:
+        rows = self.conn.execute(
+            "SELECT id, project_id, name, data_type, created_at, updated_at "
+            "FROM data_files WHERE project_id = ? ORDER BY name",
+            (project_id,),
+        ).fetchall()
+        return [dict(r) for r in rows]
+
+    def get_data_file(self, project_id: int, name: str) -> dict | None:
+        row = self.conn.execute(
+            "SELECT id, project_id, name, data_type, top_level, created_at, updated_at "
+            "FROM data_files WHERE project_id = ? AND name = ?",
+            (project_id, name),
+        ).fetchone()
+        if row is None:
+            return None
+        d = dict(row)
+        d["top_level"] = json.loads(d["top_level"])
+        return d
+
+    def get_data_file_by_names(self, project_name: str, file_name: str) -> dict | None:
+        row = self.conn.execute(
+            "SELECT df.id, df.project_id, df.name, df.data_type, df.top_level, "
+            "df.created_at, df.updated_at "
+            "FROM data_files df JOIN projects p ON df.project_id = p.id "
+            "WHERE p.name = ? AND df.name = ?",
+            (project_name, file_name),
+        ).fetchone()
+        if row is None:
+            return None
+        d = dict(row)
+        d["top_level"] = json.loads(d["top_level"])
+        return d
+
+    # ------------------------------------------------------------------
+    # Sequences
+    # ------------------------------------------------------------------
+
+    def upsert_sequence(self, data_file_id: int, sequence_number: int, data: dict) -> None:
+        now = time.time()
+        self.conn.execute(
+            "INSERT INTO sequences (data_file_id, sequence_number, data, updated_at) "
+            "VALUES (?, ?, ?, ?) "
+            "ON CONFLICT(data_file_id, sequence_number) DO UPDATE SET data=excluded.data, updated_at=excluded.updated_at",
+            (data_file_id, sequence_number, json.dumps(data), now),
+        )
+        self.conn.commit()
+
+    def get_sequence(self, data_file_id: int, sequence_number: int) -> dict | None:
+        row = self.conn.execute(
+            "SELECT data FROM sequences WHERE data_file_id = ? AND sequence_number = ?",
+            (data_file_id, sequence_number),
+        ).fetchone()
+        return json.loads(row["data"]) if row else None
+
+    def list_sequences(self, data_file_id: int) -> list[int]:
+        rows = self.conn.execute(
+            "SELECT sequence_number FROM sequences WHERE data_file_id = ? ORDER BY sequence_number",
+            (data_file_id,),
+        ).fetchall()
+        return [r["sequence_number"] for r in rows]
+
+    def count_sequences(self, data_file_id: int) -> int:
+        """Return the number of sequences for a data file."""
+        row = self.conn.execute(
+            "SELECT COUNT(*) AS cnt FROM sequences WHERE data_file_id = ?",
+            (data_file_id,),
+        ).fetchone()
+        return row["cnt"]
+
+    def query_total_sequences(self, project_name: str, file_name: str) -> int:
+        """Return total sequence count by project and file names."""
+        df = self.get_data_file_by_names(project_name, file_name)
+        if not df:
+            return 0
+        return self.count_sequences(df["id"])
+
+    def get_sequence_keys(self, data_file_id: int, sequence_number: int) -> tuple[list[str], list[str]]:
+        """Returns (keys, types) for a sequence's data dict."""
+        data = self.get_sequence(data_file_id, sequence_number)
+        if not data:
+            return [], []
+        keys = []
+        types = []
+        for k, v in data.items():
+            keys.append(k)
+            if isinstance(v, bool):
+                types.append("STRING")
+            elif isinstance(v, int):
+                types.append("INT")
+            elif isinstance(v, float):
+                types.append("FLOAT")
+            else:
+                types.append("STRING")
+        return keys, types
+
+    def delete_sequences_for_file(self, data_file_id: int) -> None:
+        self.conn.execute("DELETE FROM sequences WHERE data_file_id = ?", (data_file_id,))
+        self.conn.commit()
+
+    # ------------------------------------------------------------------
+    # History trees
+    # ------------------------------------------------------------------
+
+    def save_history_tree(self, data_file_id: int, tree_data: dict) -> None:
+        now = time.time()
+        self.conn.execute(
+            "INSERT INTO history_trees (data_file_id, tree_data, updated_at) "
+            "VALUES (?, ?, ?) "
+            "ON CONFLICT(data_file_id) DO UPDATE SET tree_data=excluded.tree_data, updated_at=excluded.updated_at",
+            (data_file_id, json.dumps(tree_data), now),
+        )
+        self.conn.commit()
+
+    def get_history_tree(self, data_file_id: int) -> dict | None:
+        row = self.conn.execute(
+            "SELECT tree_data FROM history_trees WHERE data_file_id = ?",
+            (data_file_id,),
+        ).fetchone()
+        return json.loads(row["tree_data"]) if row else None
+
+    # ------------------------------------------------------------------
+    # Import
+    # ------------------------------------------------------------------
+
+    def import_json_file(self, project_id: int, json_path: str | Path, data_type: str = "generic") -> int:
+        """Import a JSON file into the database, splitting batch_data into sequences.
+
+        Safe to call repeatedly — existing data_file is updated, sequences are
+        replaced, and history_tree is upserted. Atomic: all-or-nothing.
+        """
+        json_path = Path(json_path)
+        data, _ = load_json(json_path)
+        file_name = json_path.stem
+
+        top_level = {k: v for k, v in data.items() if k not in (KEY_BATCH_DATA, KEY_HISTORY_TREE)}
+
+        self.conn.execute("BEGIN IMMEDIATE")
+        try:
+            existing = self.conn.execute(
+                "SELECT id FROM data_files WHERE project_id = ? AND name = ?",
+                (project_id, file_name),
+            ).fetchone()
+
+            if existing:
+                df_id = existing["id"]
+                now = time.time()
+                self.conn.execute(
+                    "UPDATE data_files SET data_type = ?, top_level = ?, updated_at = ? WHERE id = ?",
+                    (data_type, json.dumps(top_level), now, df_id),
+                )
+                self.conn.execute("DELETE FROM sequences WHERE data_file_id = ?", (df_id,))
+            else:
+                now = time.time()
+                cur = self.conn.execute(
+                    "INSERT INTO data_files (project_id, name, data_type, top_level, created_at, updated_at) "
+                    "VALUES (?, ?, ?, ?, ?, ?)",
+                    (project_id, file_name, data_type, json.dumps(top_level), now, now),
+                )
+                df_id = cur.lastrowid
+
+            # Import sequences from batch_data
+            batch_data = data.get(KEY_BATCH_DATA, [])
+            if isinstance(batch_data, list):
+                for item in batch_data:
+                    if not isinstance(item, dict):
+                        continue
+                    seq_num = int(item.get("sequence_number", 0))
+                    now = time.time()
+                    self.conn.execute(
+                        "INSERT INTO sequences (data_file_id, sequence_number, data, updated_at) "
+                        "VALUES (?, ?, ?, ?) "
+                        "ON CONFLICT(data_file_id, sequence_number) DO UPDATE SET data=excluded.data, updated_at=excluded.updated_at",
+                        (df_id, seq_num, json.dumps(item), now),
+                    )
+
+            # Import history tree
+            history_tree = data.get(KEY_HISTORY_TREE)
+            if history_tree and isinstance(history_tree, dict):
+                now = time.time()
+                self.conn.execute(
+                    "INSERT INTO history_trees (data_file_id, tree_data, updated_at) "
+                    "VALUES (?, ?, ?) "
+                    "ON CONFLICT(data_file_id) DO UPDATE SET tree_data=excluded.tree_data, updated_at=excluded.updated_at",
+                    (df_id, json.dumps(history_tree), now),
+                )
+
+            self.conn.execute("COMMIT")
+            return df_id
+        except Exception:
+            try:
+                self.conn.execute("ROLLBACK")
+            except Exception:
+                pass
+            raise
+
+    # ------------------------------------------------------------------
+    # Query helpers (for REST API)
+    # ------------------------------------------------------------------
+
+    def query_sequence_data(self, project_name: str, file_name: str, sequence_number: int) -> dict | None:
+        """Query a single sequence by project name, file name, and sequence number."""
+        df = self.get_data_file_by_names(project_name, file_name)
+        if not df:
+            return None
+        return self.get_sequence(df["id"], sequence_number)
+
+    def query_sequence_keys(self, project_name: str, file_name: str, sequence_number: int) -> tuple[list[str], list[str]]:
+        """Query keys and types for a sequence."""
+        df = self.get_data_file_by_names(project_name, file_name)
+        if not df:
+            return [], []
+        return self.get_sequence_keys(df["id"], sequence_number)
+
+    def list_project_files(self, project_name: str) -> list[dict]:
+        """List data files for a project by name."""
+        proj = self.get_project(project_name)
+        if not proj:
+            return []
+        return self.list_data_files(proj["id"])
+
+    def list_project_sequences(self, project_name: str, file_name: str) -> list[int]:
+        """List sequence numbers for a file in a project."""
+        df = self.get_data_file_by_names(project_name, file_name)
+        if not df:
+            return []
+        return self.list_sequences(df["id"])
diff --git a/json_loader.py b/json_loader.py
deleted file mode 100644
index eed69fb..0000000
--- a/json_loader.py
+++ /dev/null
@@ -1,384 +0,0 @@
-import json
-import os
-import logging
-from typing import Any
-
-logger = logging.getLogger(__name__)
-
-KEY_BATCH_DATA = "batch_data"
-MAX_DYNAMIC_OUTPUTS = 32
-
-
-class AnyType(str):
-    """Universal connector type that matches any ComfyUI type."""
-    def __ne__(self, __value: object) -> bool:
-        return False
-
-any_type = AnyType("*")
-
-
-try:
-    from server import PromptServer
-    from aiohttp import web
-except ImportError:
-    PromptServer = None
-
-
-def to_float(val: Any) -> float:
-    try:
-        return float(val)
-    except (ValueError, TypeError):
-        return 0.0
-
-def to_int(val: Any) -> int:
-    try:
-        return int(float(val))
-    except (ValueError, TypeError):
-        return 0
-
-def get_batch_item(data: dict[str, Any], sequence_number: int) -> dict[str, Any]:
-    """Resolve batch item by sequence_number field, falling back to array index."""
-    if KEY_BATCH_DATA in data and isinstance(data[KEY_BATCH_DATA], list) and len(data[KEY_BATCH_DATA]) > 0:
-        # Search by sequence_number field first
-        for item in data[KEY_BATCH_DATA]:
-            if int(item.get("sequence_number", 0)) == sequence_number:
-                return item
-        # Fallback to array index
-        idx = max(0, min(sequence_number - 1, len(data[KEY_BATCH_DATA]) - 1))
-        logger.warning(f"No item with sequence_number={sequence_number}, falling back to index {idx}")
-        return data[KEY_BATCH_DATA][idx]
-    return data
-
-# --- Shared Helper ---
-def read_json_data(json_path: str) -> dict[str, Any]:
-    if not os.path.exists(json_path):
-        logger.warning(f"File not found at {json_path}")
-        return {}
-    try:
-        with open(json_path, 'r') as f:
-            data = json.load(f)
-    except (json.JSONDecodeError, IOError) as e:
-        logger.warning(f"Error reading {json_path}: {e}")
-        return {}
-    if not isinstance(data, dict):
-        logger.warning(f"Expected dict from {json_path}, got {type(data).__name__}")
-        return {}
-    return data
-
-# --- API Route ---
-if PromptServer is not None:
-    @PromptServer.instance.routes.get("/json_manager/get_keys")
-    async def get_keys_route(request):
-        json_path = request.query.get("path", "")
-        try:
-            seq = int(request.query.get("sequence_number", "1"))
-        except (ValueError, TypeError):
-            seq = 1
-        data = read_json_data(json_path)
-        target = get_batch_item(data, seq)
-        keys = []
-        types = []
-        if isinstance(target, dict):
-            for k, v in target.items():
-                keys.append(k)
-                if isinstance(v, bool):
-                    types.append("STRING")
-                elif isinstance(v, int):
-                    types.append("INT")
-                elif isinstance(v, float):
-                    types.append("FLOAT")
-                else:
-                    types.append("STRING")
-        return web.json_response({"keys": keys, "types": types})
-
-
-# ==========================================
-# 0. DYNAMIC NODE
-# ==========================================
-
-class JSONLoaderDynamic:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {
-            "required": {
-                "json_path": ("STRING", {"default": "", "multiline": False}),
-                "sequence_number": ("INT", {"default": 1, "min": 1, "max": 9999}),
-            },
-            "optional": {
-                "output_keys": ("STRING", {"default": ""}),
-                "output_types": ("STRING", {"default": ""}),
-            },
-        }
-
-    RETURN_TYPES = tuple(any_type for _ in range(MAX_DYNAMIC_OUTPUTS))
-    RETURN_NAMES = tuple(f"output_{i}" for i in range(MAX_DYNAMIC_OUTPUTS))
-    FUNCTION = "load_dynamic"
-    CATEGORY = "utils/json"
-    OUTPUT_NODE = False
-
-    def load_dynamic(self, json_path, sequence_number, output_keys="", output_types=""):
-        data = read_json_data(json_path)
-        target = get_batch_item(data, sequence_number)
-
-        keys = [k.strip() for k in output_keys.split(",") if k.strip()] if output_keys else []
-
-        results = []
-        for key in keys:
-            val = target.get(key, "")
-            if isinstance(val, bool):
-                results.append(str(val).lower())
-            elif isinstance(val, int):
-                results.append(val)
-            elif isinstance(val, float):
-                results.append(val)
-            else:
-                results.append(str(val))
-
-        # Pad to MAX_DYNAMIC_OUTPUTS
-        while len(results) < MAX_DYNAMIC_OUTPUTS:
-            results.append("")
-
-        return tuple(results)
-
-
-# ==========================================
-# 1. STANDARD NODES (Single File)
-# ==========================================
-
-class JSONLoaderLoRA:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {"required": {"json_path": ("STRING", {"default": "", "multiline": False})}}
-
-    RETURN_TYPES = ("STRING", "STRING", "STRING", "STRING", "STRING", "STRING")
-    RETURN_NAMES = ("lora_1_high", "lora_1_low", "lora_2_high", "lora_2_low", "lora_3_high", "lora_3_low")
-    FUNCTION = "load_loras"
-    CATEGORY = "utils/json"
-
-    def load_loras(self, json_path):
-        data = read_json_data(json_path)
-        return (
-            str(data.get("lora 1 high", "")), str(data.get("lora 1 low", "")),
-            str(data.get("lora 2 high", "")), str(data.get("lora 2 low", "")),
-            str(data.get("lora 3 high", "")), str(data.get("lora 3 low", ""))
-        )
-
-class JSONLoaderStandard:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {"required": {"json_path": ("STRING", {"default": "", "multiline": False})}}
-
-    RETURN_TYPES = ("STRING", "STRING", "STRING", "STRING", "STRING", "FLOAT", "INT", "STRING", "STRING", "STRING")
-    RETURN_NAMES = ("general_prompt", "general_negative", "current_prompt", "negative", "camera", "flf", "seed", "video_file_path", "reference_image_path", "flf_image_path")
-    FUNCTION = "load_standard"
-    CATEGORY = "utils/json"
-
-    def load_standard(self, json_path):
-        data = read_json_data(json_path)
-        return (
-            str(data.get("general_prompt", "")), str(data.get("general_negative", "")),
-            str(data.get("current_prompt", "")), str(data.get("negative", "")),
-            str(data.get("camera", "")), to_float(data.get("flf", 0.0)),
-            to_int(data.get("seed", 0)), str(data.get("video file path", "")),
-            str(data.get("reference image path", "")), str(data.get("flf image path", ""))
-        )
-
-class JSONLoaderVACE:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {"required": {"json_path": ("STRING", {"default": "", "multiline": False})}}
-
-    RETURN_TYPES = ("STRING", "STRING", "STRING", "STRING", "STRING", "FLOAT", "INT", "INT", "INT", "INT", "STRING", "INT", "INT", "STRING", "STRING")
-    RETURN_NAMES = ("general_prompt", "general_negative", "current_prompt", "negative", "camera", "flf", "seed", "frame_to_skip", "input_a_frames", "input_b_frames", "reference_path", "reference_switch", "vace_schedule", "video_file_path", "reference_image_path")
-    FUNCTION = "load_vace"
-    CATEGORY = "utils/json"
-
-    def load_vace(self, json_path):
-        data = read_json_data(json_path)
-        return (
-            str(data.get("general_prompt", "")), str(data.get("general_negative", "")),
-            str(data.get("current_prompt", "")), str(data.get("negative", "")),
-            str(data.get("camera", "")), to_float(data.get("flf", 0.0)),
-            to_int(data.get("seed", 0)),
-            to_int(data.get("frame_to_skip", 81)), to_int(data.get("input_a_frames", 16)),
-            to_int(data.get("input_b_frames", 16)), str(data.get("reference path", "")),
-            to_int(data.get("reference switch", 1)), to_int(data.get("vace schedule", 1)),
-            str(data.get("video file path", "")), str(data.get("reference image path", ""))
-        )
-
-# ==========================================
-# 2. BATCH NODES
-# ==========================================
-
-class JSONLoaderBatchLoRA:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {"required": {"json_path": ("STRING", {"default": "", "multiline": False}), "sequence_number": ("INT", {"default": 1, "min": 1, "max": 9999})}}
-    RETURN_TYPES = ("STRING", "STRING", "STRING", "STRING", "STRING", "STRING")
-    RETURN_NAMES = ("lora_1_high", "lora_1_low", "lora_2_high", "lora_2_low", "lora_3_high", "lora_3_low")
-    FUNCTION = "load_batch_loras"
-    CATEGORY = "utils/json"
-
-    def load_batch_loras(self, json_path, sequence_number):
-        data = read_json_data(json_path)
-        target_data = get_batch_item(data, sequence_number)
-        return (
-            str(target_data.get("lora 1 high", "")), str(target_data.get("lora 1 low", "")),
-            str(target_data.get("lora 2 high", "")), str(target_data.get("lora 2 low", "")),
-            str(target_data.get("lora 3 high", "")), str(target_data.get("lora 3 low", ""))
-        )
-
-class JSONLoaderBatchI2V:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {"required": {"json_path": ("STRING", {"default": "", "multiline": False}), "sequence_number": ("INT", {"default": 1, "min": 1, "max": 9999})}}
-    RETURN_TYPES = ("STRING", "STRING", "STRING", "STRING", "STRING", "FLOAT", "INT", "STRING", "STRING", "STRING")
-    RETURN_NAMES = ("general_prompt", "general_negative", "current_prompt", "negative", "camera", "flf", "seed", "video_file_path", "reference_image_path", "flf_image_path")
-    FUNCTION = "load_batch_i2v"
-    CATEGORY = "utils/json"
-
-    def load_batch_i2v(self, json_path, sequence_number):
-        data = read_json_data(json_path)
-        target_data = get_batch_item(data, sequence_number)
-
-        return (
-            str(target_data.get("general_prompt", "")), str(target_data.get("general_negative", "")),
-            str(target_data.get("current_prompt", "")), str(target_data.get("negative", "")),
-            str(target_data.get("camera", "")), to_float(target_data.get("flf", 0.0)),
-            to_int(target_data.get("seed", 0)), str(target_data.get("video file path", "")),
-            str(target_data.get("reference image path", "")), str(target_data.get("flf image path", ""))
-        )
-
-class JSONLoaderBatchVACE:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {"required": {"json_path": ("STRING", {"default": "", "multiline": False}), "sequence_number": ("INT", {"default": 1, "min": 1, "max": 9999})}}
-    RETURN_TYPES = ("STRING", "STRING", "STRING", "STRING", "STRING", "FLOAT", "INT", "INT", "INT", "INT", "STRING", "INT", "INT", "STRING", "STRING")
-    RETURN_NAMES = ("general_prompt", "general_negative", "current_prompt", "negative", "camera", "flf", "seed", "frame_to_skip", "input_a_frames", "input_b_frames", "reference_path", "reference_switch", "vace_schedule", "video_file_path", "reference_image_path")
-    FUNCTION = "load_batch_vace"
-    CATEGORY = "utils/json"
-
-    def load_batch_vace(self, json_path, sequence_number):
-        data = read_json_data(json_path)
-        target_data = get_batch_item(data, sequence_number)
-
-        return (
-            str(target_data.get("general_prompt", "")), str(target_data.get("general_negative", "")),
-            str(target_data.get("current_prompt", "")), str(target_data.get("negative", "")),
-            str(target_data.get("camera", "")), to_float(target_data.get("flf", 0.0)),
-            to_int(target_data.get("seed", 0)), to_int(target_data.get("frame_to_skip", 81)),
-            to_int(target_data.get("input_a_frames", 16)), to_int(target_data.get("input_b_frames", 16)),
-            str(target_data.get("reference path", "")), to_int(target_data.get("reference switch", 1)),
-            to_int(target_data.get("vace schedule", 1)), str(target_data.get("video file path", "")),
-            str(target_data.get("reference image path", ""))
-        )
-
-# ==========================================
-# 3. UNIVERSAL CUSTOM NODES (1, 3, 6 Slots)
-# ==========================================
-
-class JSONLoaderCustom1:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {
-            "required": {
-                "json_path": ("STRING", {"default": "", "multiline": False}),
-                "sequence_number": ("INT", {"default": 1, "min": 1, "max": 9999}),
-            },
-            "optional": { "key_1": ("STRING", {"default": "", "multiline": False}) }
-        }
-    RETURN_TYPES = ("STRING",)
-    RETURN_NAMES = ("val_1",)
-    FUNCTION = "load_custom"
-    CATEGORY = "utils/json"
-
-    def load_custom(self, json_path, sequence_number, key_1=""):
-        data = read_json_data(json_path)
-        target_data = get_batch_item(data, sequence_number)
-        return (str(target_data.get(key_1, "")),)
-
-class JSONLoaderCustom3:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {
-            "required": {
-                "json_path": ("STRING", {"default": "", "multiline": False}),
-                "sequence_number": ("INT", {"default": 1, "min": 1, "max": 9999}),
-            },
-            "optional": {
-                "key_1": ("STRING", {"default": "", "multiline": False}),
-                "key_2": ("STRING", {"default": "", "multiline": False}),
-                "key_3": ("STRING", {"default": "", "multiline": False})
-            }
-        }
-    RETURN_TYPES = ("STRING", "STRING", "STRING")
-    RETURN_NAMES = ("val_1", "val_2", "val_3")
-    FUNCTION = "load_custom"
-    CATEGORY = "utils/json"
-
-    def load_custom(self, json_path, sequence_number, key_1="", key_2="", key_3=""):
-        data = read_json_data(json_path)
-        target_data = get_batch_item(data, sequence_number)
-        return (
-            str(target_data.get(key_1, "")),
-            str(target_data.get(key_2, "")),
-            str(target_data.get(key_3, ""))
-        )
-
-class JSONLoaderCustom6:
-    @classmethod
-    def INPUT_TYPES(s):
-        return {
-            "required": {
-                "json_path": ("STRING", {"default": "", "multiline": False}),
-                "sequence_number": ("INT", {"default": 1, "min": 1, "max": 9999}),
-            },
-            "optional": {
-                "key_1": ("STRING", {"default": "", "multiline": False}),
-                "key_2": ("STRING", {"default": "", "multiline": False}),
-                "key_3": ("STRING", {"default": "", "multiline": False}),
-                "key_4": ("STRING", {"default": "", "multiline": False}),
-                "key_5": ("STRING", {"default": "", "multiline": False}),
-                "key_6": ("STRING", {"default": "", "multiline": False})
-            }
-        }
-    RETURN_TYPES = ("STRING", "STRING", "STRING", "STRING", "STRING", "STRING")
-    RETURN_NAMES = ("val_1", "val_2", "val_3", "val_4", "val_5", "val_6")
-    FUNCTION = "load_custom"
-    CATEGORY = "utils/json"
-
-    def load_custom(self, json_path, sequence_number, key_1="", key_2="", key_3="", key_4="", key_5="", key_6=""):
-        data = read_json_data(json_path)
-        target_data = get_batch_item(data, sequence_number)
-        return (
-            str(target_data.get(key_1, "")), str(target_data.get(key_2, "")),
-            str(target_data.get(key_3, "")), str(target_data.get(key_4, "")),
-            str(target_data.get(key_5, "")), str(target_data.get(key_6, ""))
-        )
-
-# --- Mappings ---
-NODE_CLASS_MAPPINGS = {
-    "JSONLoaderDynamic": JSONLoaderDynamic,
-    "JSONLoaderLoRA": JSONLoaderLoRA,
-    "JSONLoaderStandard": JSONLoaderStandard,
-    "JSONLoaderVACE": JSONLoaderVACE,
-    "JSONLoaderBatchLoRA": JSONLoaderBatchLoRA,
-    "JSONLoaderBatchI2V": JSONLoaderBatchI2V,
-    "JSONLoaderBatchVACE": JSONLoaderBatchVACE,
-    "JSONLoaderCustom1": JSONLoaderCustom1,
-    "JSONLoaderCustom3": JSONLoaderCustom3,
-    "JSONLoaderCustom6": JSONLoaderCustom6
-}
-
-NODE_DISPLAY_NAME_MAPPINGS = {
-    "JSONLoaderDynamic": "JSON Loader (Dynamic)",
-    "JSONLoaderLoRA": "JSON Loader (LoRAs Only)",
-    "JSONLoaderStandard": "JSON Loader (Standard/I2V)",
-    "JSONLoaderVACE": "JSON Loader (VACE Full)",
-    "JSONLoaderBatchLoRA": "JSON Batch Loader (LoRAs)",
-    "JSONLoaderBatchI2V": "JSON Batch Loader (I2V)",
-    "JSONLoaderBatchVACE": "JSON Batch Loader (VACE)",
-    "JSONLoaderCustom1": "JSON Loader (Custom 1)",
-    "JSONLoaderCustom3": "JSON Loader (Custom 3)",
-    "JSONLoaderCustom6": "JSON Loader (Custom 6)"
-}
diff --git a/main.py b/main.py
index bf34d82..072fe1e 100644
--- a/main.py
+++ b/main.py
@@ -1,4 +1,5 @@
 import json
+import logging
 from pathlib import Path
 
 from nicegui import ui
@@ -14,6 +15,18 @@ from tab_batch_ng import render_batch_processor
 from tab_timeline_ng import render_timeline_tab
 from tab_raw_ng import render_raw_editor
 from tab_comfy_ng import render_comfy_monitor
+from tab_projects_ng import render_projects_tab
+from db import ProjectDB
+from api_routes import register_api_routes
+
+logger = logging.getLogger(__name__)
+
+# Single shared DB instance for both the UI and API routes
+_shared_db: ProjectDB | None = None
+try:
+    _shared_db = ProjectDB()
+except Exception as e:
+    logger.warning(f"Failed to initialize ProjectDB: {e}")
 
 
 @ui.page('/')
@@ -165,7 +178,13 @@ def index():
         config=config,
         current_dir=Path(config.get('last_dir', str(Path.cwd()))),
         snippets=load_snippets(),
+        db_enabled=config.get('db_enabled', False),
+        current_project=config.get('current_project', ''),
     )
+
+    # Use the shared DB instance
+    state.db = _shared_db
+
     dual_pane = {'active': False, 'state': None}
 
     # ------------------------------------------------------------------
@@ -187,6 +206,7 @@ def index():
                 ui.tab('batch', label='Batch Processor')
                 ui.tab('timeline', label='Timeline')
                 ui.tab('raw', label='Raw Editor')
+                ui.tab('projects', label='Projects')
 
             with ui.tab_panels(tabs, value='batch').classes('w-full'):
                 with ui.tab_panel('batch'):
@@ -195,6 +215,8 @@ def index():
                     render_timeline_tab(state)
                 with ui.tab_panel('raw'):
                     render_raw_editor(state)
+                with ui.tab_panel('projects'):
+                    render_projects_tab(state)
 
             if state.show_comfy_monitor:
                 ui.separator()
@@ -490,4 +512,8 @@ def render_sidebar(state: AppState, dual_pane: dict):
     ui.checkbox('Show Comfy Monitor', value=True, on_change=on_monitor_toggle)
 
 
+# Register REST API routes for ComfyUI connectivity (uses the shared DB instance)
+if _shared_db is not None:
+    register_api_routes(_shared_db)
+
 ui.run(title='AI Settings Manager', port=8080, reload=True)
diff --git a/project_loader.py b/project_loader.py
new file mode 100644
index 0000000..6420517
--- /dev/null
+++ b/project_loader.py
@@ -0,0 +1,215 @@
+import json
+import logging
+import urllib.parse
+import urllib.request
+import urllib.error
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+MAX_DYNAMIC_OUTPUTS = 32
+
+
+class AnyType(str):
+    """Universal connector type that matches any ComfyUI type."""
+    def __ne__(self, __value: object) -> bool:
+        return False
+
+any_type = AnyType("*")
+
+
+try:
+    from server import PromptServer
+    from aiohttp import web
+except ImportError:
+    PromptServer = None
+
+
+def to_float(val: Any) -> float:
+    try:
+        return float(val)
+    except (ValueError, TypeError):
+        return 0.0
+
+def to_int(val: Any) -> int:
+    try:
+        return int(float(val))
+    except (ValueError, TypeError):
+        return 0
+
+
+def _fetch_json(url: str) -> dict:
+    """Fetch JSON from a URL using stdlib urllib.
+
+    On error, returns a dict with an "error" key describing the failure.
+    """
+    try:
+        with urllib.request.urlopen(url, timeout=5) as resp:
+            return json.loads(resp.read())
+    except urllib.error.HTTPError as e:
+        # HTTPError is a subclass of URLError — must be caught first
+        body = ""
+        try:
+            raw = e.read()
+            detail = json.loads(raw)
+            body = detail.get("detail", str(raw, "utf-8", errors="replace"))
+        except Exception:
+            body = str(e)
+        logger.warning(f"HTTP {e.code} from {url}: {body}")
+        return {"error": "http_error", "status": e.code, "message": body}
+    except (urllib.error.URLError, OSError) as e:
+        reason = str(e.reason) if hasattr(e, "reason") else str(e)
+        logger.warning(f"Network error fetching {url}: {reason}")
+        return {"error": "network_error", "message": reason}
+    except json.JSONDecodeError as e:
+        logger.warning(f"Invalid JSON from {url}: {e}")
+        return {"error": "parse_error", "message": str(e)}
+
+
+def _fetch_data(manager_url: str, project: str, file: str, seq: int) -> dict:
+    """Fetch sequence data from the NiceGUI REST API."""
+    p = urllib.parse.quote(project, safe='')
+    f = urllib.parse.quote(file, safe='')
+    url = f"{manager_url.rstrip('/')}/api/projects/{p}/files/{f}/data?seq={seq}"
+    return _fetch_json(url)
+
+
+def _fetch_keys(manager_url: str, project: str, file: str, seq: int) -> dict:
+    """Fetch keys/types from the NiceGUI REST API."""
+    p = urllib.parse.quote(project, safe='')
+    f = urllib.parse.quote(file, safe='')
+    url = f"{manager_url.rstrip('/')}/api/projects/{p}/files/{f}/keys?seq={seq}"
+    return _fetch_json(url)
+
+
+# --- ComfyUI-side proxy endpoints (for frontend JS) ---
+if PromptServer is not None:
+    @PromptServer.instance.routes.get("/json_manager/list_projects")
+    async def list_projects_proxy(request):
+        manager_url = request.query.get("url", "http://localhost:8080")
+        url = f"{manager_url.rstrip('/')}/api/projects"
+        data = _fetch_json(url)
+        return web.json_response(data)
+
+    @PromptServer.instance.routes.get("/json_manager/list_project_files")
+    async def list_project_files_proxy(request):
+        manager_url = request.query.get("url", "http://localhost:8080")
+        project = urllib.parse.quote(request.query.get("project", ""), safe='')
+        url = f"{manager_url.rstrip('/')}/api/projects/{project}/files"
+        data = _fetch_json(url)
+        return web.json_response(data)
+
+    @PromptServer.instance.routes.get("/json_manager/list_project_sequences")
+    async def list_project_sequences_proxy(request):
+        manager_url = request.query.get("url", "http://localhost:8080")
+        project = urllib.parse.quote(request.query.get("project", ""), safe='')
+        file_name = urllib.parse.quote(request.query.get("file", ""), safe='')
+        url = f"{manager_url.rstrip('/')}/api/projects/{project}/files/{file_name}/sequences"
+        data = _fetch_json(url)
+        return web.json_response(data)
+
+    @PromptServer.instance.routes.get("/json_manager/get_project_keys")
+    async def get_project_keys_proxy(request):
+        manager_url = request.query.get("url", "http://localhost:8080")
+        project = request.query.get("project", "")
+        file_name = request.query.get("file", "")
+        try:
+            seq = int(request.query.get("seq", "1"))
+        except (ValueError, TypeError):
+            seq = 1
+        data = _fetch_keys(manager_url, project, file_name, seq)
+        if data.get("error") in ("http_error", "network_error", "parse_error"):
+            status = data.get("status", 502)
+            return web.json_response(data, status=status)
+        return web.json_response(data)
+
+
+
+# ==========================================
+# 0. DYNAMIC NODE (Project-based)
+# ==========================================
+
+class ProjectLoaderDynamic:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {
+            "required": {
+                "manager_url": ("STRING", {"default": "http://localhost:8080", "multiline": False}),
+                "project_name": ("STRING", {"default": "", "multiline": False}),
+                "file_name": ("STRING", {"default": "", "multiline": False}),
+                "sequence_number": ("INT", {"default": 1, "min": 1, "max": 9999}),
+            },
+            "optional": {
+                "output_keys": ("STRING", {"default": ""}),
+                "output_types": ("STRING", {"default": ""}),
+            },
+        }
+
+    RETURN_TYPES = ("INT",) + tuple(any_type for _ in range(MAX_DYNAMIC_OUTPUTS))
+    RETURN_NAMES = ("total_sequences",) + tuple(f"output_{i}" for i in range(MAX_DYNAMIC_OUTPUTS))
+    FUNCTION = "load_dynamic"
+    CATEGORY = "utils/json/project"
+    OUTPUT_NODE = False
+
+    def load_dynamic(self, manager_url, project_name, file_name, sequence_number,
+                     output_keys="", output_types=""):
+        # Fetch keys metadata (includes total_sequences count)
+        keys_meta = _fetch_keys(manager_url, project_name, file_name, sequence_number)
+        if keys_meta.get("error") in ("http_error", "network_error", "parse_error"):
+            msg = keys_meta.get("message", "Unknown error")
+            raise RuntimeError(f"Failed to fetch project keys: {msg}")
+        total_sequences = keys_meta.get("total_sequences", 0)
+
+        data = _fetch_data(manager_url, project_name, file_name, sequence_number)
+        if data.get("error") in ("http_error", "network_error", "parse_error"):
+            msg = data.get("message", "Unknown error")
+            raise RuntimeError(f"Failed to fetch sequence data: {msg}")
+
+        # Parse keys — try JSON array first, fall back to comma-split for compat
+        keys = []
+        if output_keys:
+            try:
+                keys = json.loads(output_keys)
+            except (json.JSONDecodeError, TypeError):
+                keys = [k.strip() for k in output_keys.split(",") if k.strip()]
+
+        # Parse types for coercion
+        types = []
+        if output_types:
+            try:
+                types = json.loads(output_types)
+            except (json.JSONDecodeError, TypeError):
+                types = [t.strip() for t in output_types.split(",")]
+
+        results = []
+        for i, key in enumerate(keys):
+            val = data.get(key, "")
+            declared_type = types[i] if i < len(types) else ""
+            # Coerce based on declared output type when possible
+            if declared_type == "INT":
+                results.append(to_int(val))
+            elif declared_type == "FLOAT":
+                results.append(to_float(val))
+            elif isinstance(val, bool):
+                results.append(str(val).lower())
+            elif isinstance(val, int):
+                results.append(val)
+            elif isinstance(val, float):
+                results.append(val)
+            else:
+                results.append(str(val))
+
+        while len(results) < MAX_DYNAMIC_OUTPUTS:
+            results.append("")
+
+        return (total_sequences,) + tuple(results)
+
+
+# --- Mappings ---
+PROJECT_NODE_CLASS_MAPPINGS = {
+    "ProjectLoaderDynamic": ProjectLoaderDynamic,
+}
+
+PROJECT_NODE_DISPLAY_NAME_MAPPINGS = {
+    "ProjectLoaderDynamic": "Project Loader (Dynamic)",
+}
diff --git a/state.py b/state.py
index e4aeab4..bef8818 100644
--- a/state.py
+++ b/state.py
@@ -17,6 +17,11 @@ class AppState:
     live_toggles: dict = field(default_factory=dict)
     show_comfy_monitor: bool = True
 
+    # Project DB fields
+    db: Any = None
+    current_project: str = ""
+    db_enabled: bool = False
+
     # Set at runtime by main.py / tab_comfy_ng.py
     _render_main: Any = None
     _load_file: Callable | None = None
@@ -29,4 +34,7 @@ class AppState:
             config=self.config,
             current_dir=self.current_dir,
             snippets=self.snippets,
+            db=self.db,
+            current_project=self.current_project,
+            db_enabled=self.db_enabled,
         )
diff --git a/tab_batch_ng.py b/tab_batch_ng.py
index da47601..3324169 100644
--- a/tab_batch_ng.py
+++ b/tab_batch_ng.py
@@ -6,7 +6,7 @@ from nicegui import ui
 
 from state import AppState
 from utils import (
-    DEFAULTS, save_json, load_json,
+    DEFAULTS, save_json, load_json, sync_to_db,
     KEY_BATCH_DATA, KEY_HISTORY_TREE, KEY_PROMPT_HISTORY, KEY_SEQUENCE_NUMBER,
 )
 from history_tree import HistoryTree
@@ -161,6 +161,8 @@ def render_batch_processor(state: AppState):
             new_data = {KEY_BATCH_DATA: [first_item], KEY_HISTORY_TREE: {},
                         KEY_PROMPT_HISTORY: []}
             save_json(new_path, new_data)
+            if state.db_enabled and state.current_project and state.db:
+                sync_to_db(state.db, state.current_project, new_path, new_data)
             ui.notify(f'Created {new_name}', type='positive')
 
         ui.button('Create Batch Copy', icon='content_copy', on_click=create_batch)
@@ -215,6 +217,8 @@ def render_batch_processor(state: AppState):
                 batch_list.append(new_item)
                 data[KEY_BATCH_DATA] = batch_list
                 save_json(file_path, data)
+                if state.db_enabled and state.current_project and state.db:
+                    sync_to_db(state.db, state.current_project, file_path, data)
                 render_sequence_list.refresh()
 
             with ui.row().classes('q-mt-sm'):
@@ -250,6 +254,8 @@ def render_batch_processor(state: AppState):
         batch_list.sort(key=lambda s: int(s.get(KEY_SEQUENCE_NUMBER, 0)))
         data[KEY_BATCH_DATA] = batch_list
         save_json(file_path, data)
+        if state.db_enabled and state.current_project and state.db:
+            sync_to_db(state.db, state.current_project, file_path, data)
         ui.notify('Sorted by sequence number!', type='positive')
         render_sequence_list.refresh()
 
@@ -289,6 +295,8 @@ def render_batch_processor(state: AppState):
                 htree.commit(snapshot_payload, note=note)
                 data[KEY_HISTORY_TREE] = htree.to_dict()
                 save_json(file_path, data)
+                if state.db_enabled and state.current_project and state.db:
+                    sync_to_db(state.db, state.current_project, file_path, data)
                 state.restored_indicator = None
                 commit_input.set_value('')
                 ui.notify('Batch Saved & Snapshot Created!', type='positive')
@@ -306,6 +314,8 @@ def _render_sequence_card(i, seq, batch_list, data, file_path, state,
     def commit(message=None):
         data[KEY_BATCH_DATA] = batch_list
         save_json(file_path, data)
+        if state.db_enabled and state.current_project and state.db:
+            sync_to_db(state.db, state.current_project, file_path, data)
         if message:
             ui.notify(message, type='positive')
         refresh_list.refresh()
@@ -448,7 +458,7 @@ def _render_sequence_card(i, seq, batch_list, data, file_path, state,
 
         # --- VACE Settings (full width) ---
         with ui.expansion('VACE Settings', icon='settings').classes('w-full'):
-            _render_vace_settings(i, seq, batch_list, data, file_path, refresh_list)
+            _render_vace_settings(i, seq, batch_list, data, file_path, state, refresh_list)
 
         # --- LoRA Settings ---
         with ui.expansion('LoRA Settings', icon='style').classes('w-full'):
@@ -530,7 +540,7 @@ def _render_sequence_card(i, seq, batch_list, data, file_path, state,
 # VACE Settings sub-section
 # ======================================================================
 
-def _render_vace_settings(i, seq, batch_list, data, file_path, refresh_list):
+def _render_vace_settings(i, seq, batch_list, data, file_path, state, refresh_list):
     # VACE Schedule (needed early for both columns)
     sched_val = max(0, min(int(seq.get('vace schedule', 1)), len(VACE_MODES) - 1))
 
@@ -568,6 +578,8 @@ def _render_vace_settings(i, seq, batch_list, data, file_path, refresh_list):
                         shifted += 1
                     data[KEY_BATCH_DATA] = batch_list
                     save_json(file_path, data)
+                    if state.db_enabled and state.current_project and state.db:
+                        sync_to_db(state.db, state.current_project, file_path, data)
                     ui.notify(f'Shifted {shifted} sequences by {delta:+d}', type='positive')
                     refresh_list.refresh()
 
@@ -713,6 +725,8 @@ def _render_mass_update(batch_list, data, file_path, state: AppState, refresh_li
             htree.commit(snapshot, f"Mass update: {', '.join(selected_keys)}")
             data[KEY_HISTORY_TREE] = htree.to_dict()
             save_json(file_path, data)
+            if state.db_enabled and state.current_project and state.db:
+                sync_to_db(state.db, state.current_project, file_path, data)
             ui.notify(f'Updated {len(targets)} sequences', type='positive')
             if refresh_list:
                 refresh_list.refresh()
diff --git a/tab_projects_ng.py b/tab_projects_ng.py
new file mode 100644
index 0000000..32494ac
--- /dev/null
+++ b/tab_projects_ng.py
@@ -0,0 +1,165 @@
+import logging
+from pathlib import Path
+
+from nicegui import ui
+
+from state import AppState
+from db import ProjectDB
+from utils import save_config, sync_to_db, KEY_BATCH_DATA
+
+logger = logging.getLogger(__name__)
+
+
+def render_projects_tab(state: AppState):
+    """Render the Projects management tab."""
+
+    # --- DB toggle ---
+    def on_db_toggle(e):
+        state.db_enabled = e.value
+        state.config['db_enabled'] = e.value
+        save_config(state.current_dir, state.config.get('favorites', []), state.config)
+        render_project_content.refresh()
+
+    ui.switch('Enable Project Database', value=state.db_enabled,
+              on_change=on_db_toggle).classes('q-mb-md')
+
+    @ui.refreshable
+    def render_project_content():
+        if not state.db_enabled:
+            ui.label('Project database is disabled. Enable it above to manage projects.').classes(
+                'text-caption q-pa-md')
+            return
+
+        if not state.db:
+            ui.label('Database not initialized.').classes('text-warning q-pa-md')
+            return
+
+        # --- Create project form ---
+        with ui.card().classes('w-full q-pa-md q-mb-md'):
+            ui.label('Create New Project').classes('section-header')
+            name_input = ui.input('Project Name', placeholder='my_project').classes('w-full')
+            desc_input = ui.input('Description (optional)', placeholder='A short description').classes('w-full')
+
+            def create_project():
+                name = name_input.value.strip()
+                if not name:
+                    ui.notify('Please enter a project name', type='warning')
+                    return
+                try:
+                    state.db.create_project(name, str(state.current_dir), desc_input.value.strip())
+                    name_input.set_value('')
+                    desc_input.set_value('')
+                    ui.notify(f'Created project "{name}"', type='positive')
+                    render_project_list.refresh()
+                except Exception as e:
+                    ui.notify(f'Error: {e}', type='negative')
+
+            ui.button('Create Project', icon='add', on_click=create_project).classes('w-full')
+
+        # --- Active project indicator ---
+        if state.current_project:
+            ui.label(f'Active Project: {state.current_project}').classes(
+                'text-bold text-primary q-pa-sm')
+
+        # --- Project list ---
+        @ui.refreshable
+        def render_project_list():
+            projects = state.db.list_projects()
+            if not projects:
+                ui.label('No projects yet. Create one above.').classes('text-caption q-pa-md')
+                return
+
+            for proj in projects:
+                is_active = proj['name'] == state.current_project
+                card_style = 'border-left: 3px solid var(--accent);' if is_active else ''
+
+                with ui.card().classes('w-full q-pa-sm q-mb-sm').style(card_style):
+                    with ui.row().classes('w-full items-center'):
+                        with ui.column().classes('col'):
+                            ui.label(proj['name']).classes('text-bold')
+                            if proj['description']:
+                                ui.label(proj['description']).classes('text-caption')
+                            ui.label(f'Path: {proj["folder_path"]}').classes('text-caption')
+                            files = state.db.list_data_files(proj['id'])
+                            ui.label(f'{len(files)} data file(s)').classes('text-caption')
+
+                        with ui.row().classes('q-gutter-xs'):
+                            if not is_active:
+                                def activate(name=proj['name']):
+                                    state.current_project = name
+                                    state.config['current_project'] = name
+                                    save_config(state.current_dir,
+                                                state.config.get('favorites', []),
+                                                state.config)
+                                    ui.notify(f'Activated project "{name}"', type='positive')
+                                    render_project_list.refresh()
+
+                                ui.button('Activate', icon='check_circle',
+                                          on_click=activate).props('flat dense color=primary')
+                            else:
+                                def deactivate():
+                                    state.current_project = ''
+                                    state.config['current_project'] = ''
+                                    save_config(state.current_dir,
+                                                state.config.get('favorites', []),
+                                                state.config)
+                                    ui.notify('Deactivated project', type='info')
+                                    render_project_list.refresh()
+
+                                ui.button('Deactivate', icon='cancel',
+                                          on_click=deactivate).props('flat dense')
+
+                            def import_folder(pid=proj['id'], pname=proj['name']):
+                                _import_folder(state, pid, pname, render_project_list)
+
+                            ui.button('Import Folder', icon='folder_open',
+                                      on_click=import_folder).props('flat dense')
+
+                            def delete_proj(name=proj['name']):
+                                state.db.delete_project(name)
+                                if state.current_project == name:
+                                    state.current_project = ''
+                                    state.config['current_project'] = ''
+                                    save_config(state.current_dir,
+                                                state.config.get('favorites', []),
+                                                state.config)
+                                ui.notify(f'Deleted project "{name}"', type='positive')
+                                render_project_list.refresh()
+
+                            ui.button(icon='delete',
+                                      on_click=delete_proj).props('flat dense color=negative')
+
+        render_project_list()
+
+    render_project_content()
+
+
+def _import_folder(state: AppState, project_id: int, project_name: str, refresh_fn):
+    """Bulk import all .json files from current directory into a project."""
+    json_files = sorted(state.current_dir.glob('*.json'))
+    json_files = [f for f in json_files if f.name not in (
+        '.editor_config.json', '.editor_snippets.json')]
+
+    if not json_files:
+        ui.notify('No JSON files in current directory', type='warning')
+        return
+
+    imported = 0
+    skipped = 0
+    for jf in json_files:
+        file_name = jf.stem
+        existing = state.db.get_data_file(project_id, file_name)
+        if existing:
+            skipped += 1
+            continue
+        try:
+            state.db.import_json_file(project_id, jf)
+            imported += 1
+        except Exception as e:
+            logger.warning(f"Failed to import {jf}: {e}")
+
+    msg = f'Imported {imported} file(s)'
+    if skipped:
+        msg += f', skipped {skipped} existing'
+    ui.notify(msg, type='positive')
+    refresh_fn.refresh()
diff --git a/tab_raw_ng.py b/tab_raw_ng.py
index 39ec6f3..bdc4933 100644
--- a/tab_raw_ng.py
+++ b/tab_raw_ng.py
@@ -4,7 +4,7 @@ import json
 from nicegui import ui
 
 from state import AppState
-from utils import save_json, get_file_mtime, KEY_HISTORY_TREE, KEY_PROMPT_HISTORY
+from utils import save_json, sync_to_db, get_file_mtime, KEY_HISTORY_TREE, KEY_PROMPT_HISTORY
 
 
 def render_raw_editor(state: AppState):
@@ -52,6 +52,8 @@ def render_raw_editor(state: AppState):
                             input_data[KEY_PROMPT_HISTORY] = data[KEY_PROMPT_HISTORY]
 
                     save_json(file_path, input_data)
+                    if state.db_enabled and state.current_project and state.db:
+                        sync_to_db(state.db, state.current_project, file_path, input_data)
 
                     data.clear()
                     data.update(input_data)
diff --git a/tab_timeline_ng.py b/tab_timeline_ng.py
index daaf460..d0467f4 100644
--- a/tab_timeline_ng.py
+++ b/tab_timeline_ng.py
@@ -5,7 +5,7 @@ from nicegui import ui
 
 from state import AppState
 from history_tree import HistoryTree
-from utils import save_json, KEY_BATCH_DATA, KEY_HISTORY_TREE
+from utils import save_json, sync_to_db, KEY_BATCH_DATA, KEY_HISTORY_TREE
 
 
 def _delete_nodes(htree, data, file_path, node_ids):
@@ -134,6 +134,8 @@ def _render_batch_delete(htree, data, file_path, state, refresh_fn):
     def do_batch_delete():
         current_valid = state.timeline_selected_nodes & set(htree.nodes.keys())
         _delete_nodes(htree, data, file_path, current_valid)
+        if state.db_enabled and state.current_project and state.db:
+            sync_to_db(state.db, state.current_project, file_path, data)
         state.timeline_selected_nodes = set()
         ui.notify(
             f'Deleted {len(current_valid)} node{"s" if len(current_valid) != 1 else ""}!',
@@ -179,7 +181,7 @@ def _find_branch_for_node(htree, node_id):
 
 
 def _render_node_manager(all_nodes, htree, data, file_path, restore_fn, refresh_fn,
-                         selected):
+                         selected, state=None):
     """Render branch-grouped node manager with restore, rename, delete, and preview."""
     ui.label('Manage Version').classes('section-header')
 
@@ -291,6 +293,8 @@ def _render_node_manager(all_nodes, htree, data, file_path, restore_fn, refresh_
                     htree.nodes[sel_id]['note'] = rename_input.value
                     data[KEY_HISTORY_TREE] = htree.to_dict()
                     save_json(file_path, data)
+                    if state and state.db_enabled and state.current_project and state.db:
+                        sync_to_db(state.db, state.current_project, file_path, data)
                     ui.notify('Label updated', type='positive')
                     refresh_fn()
 
@@ -304,6 +308,8 @@ def _render_node_manager(all_nodes, htree, data, file_path, restore_fn, refresh_
             def delete_selected():
                 if sel_id in htree.nodes:
                     _delete_nodes(htree, data, file_path, {sel_id})
+                    if state and state.db_enabled and state.current_project and state.db:
+                        sync_to_db(state.db, state.current_project, file_path, data)
                     ui.notify('Node Deleted', type='positive')
                     refresh_fn()
 
@@ -377,7 +383,7 @@ def render_timeline_tab(state: AppState):
             _render_node_manager(
                 all_nodes, htree, data, file_path,
                 _restore_and_refresh, render_timeline.refresh,
-                selected)
+                selected, state=state)
 
     def _toggle_select(nid, checked):
         if checked:
@@ -492,6 +498,8 @@ def _restore_node(data, node, htree, file_path, state: AppState):
     htree.head_id = node['id']
     data[KEY_HISTORY_TREE] = htree.to_dict()
     save_json(file_path, data)
+    if state.db_enabled and state.current_project and state.db:
+        sync_to_db(state.db, state.current_project, file_path, data)
     label = f"{node.get('note', 'Step')} ({node['id'][:4]})"
     state.restored_indicator = label
     ui.notify('Restored!', type='positive')
diff --git a/tests/test_db.py b/tests/test_db.py
new file mode 100644
index 0000000..bea102f
--- /dev/null
+++ b/tests/test_db.py
@@ -0,0 +1,369 @@
+import json
+from pathlib import Path
+
+import pytest
+
+from db import ProjectDB
+from utils import KEY_BATCH_DATA, KEY_HISTORY_TREE
+
+
+@pytest.fixture
+def db(tmp_path):
+    """Create a fresh ProjectDB in a temp directory."""
+    db_path = tmp_path / "test.db"
+    pdb = ProjectDB(db_path)
+    yield pdb
+    pdb.close()
+
+
+# ------------------------------------------------------------------
+# Projects CRUD
+# ------------------------------------------------------------------
+
+class TestProjects:
+    def test_create_and_get(self, db):
+        pid = db.create_project("proj1", "/some/path", "A test project")
+        assert pid > 0
+        proj = db.get_project("proj1")
+        assert proj is not None
+        assert proj["name"] == "proj1"
+        assert proj["folder_path"] == "/some/path"
+        assert proj["description"] == "A test project"
+
+    def test_list_projects(self, db):
+        db.create_project("beta", "/b")
+        db.create_project("alpha", "/a")
+        projects = db.list_projects()
+        assert len(projects) == 2
+        assert projects[0]["name"] == "alpha"
+        assert projects[1]["name"] == "beta"
+
+    def test_get_nonexistent(self, db):
+        assert db.get_project("nope") is None
+
+    def test_delete_project(self, db):
+        db.create_project("to_delete", "/x")
+        assert db.delete_project("to_delete") is True
+        assert db.get_project("to_delete") is None
+
+    def test_delete_nonexistent(self, db):
+        assert db.delete_project("nope") is False
+
+    def test_unique_name_constraint(self, db):
+        db.create_project("dup", "/a")
+        with pytest.raises(Exception):
+            db.create_project("dup", "/b")
+
+
+# ------------------------------------------------------------------
+# Data files
+# ------------------------------------------------------------------
+
+class TestDataFiles:
+    def test_create_and_list(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch_i2v", "i2v", {"extra": "meta"})
+        assert df_id > 0
+        files = db.list_data_files(pid)
+        assert len(files) == 1
+        assert files[0]["name"] == "batch_i2v"
+        assert files[0]["data_type"] == "i2v"
+
+    def test_get_data_file(self, db):
+        pid = db.create_project("p1", "/p1")
+        db.create_data_file(pid, "batch_i2v", "i2v", {"key": "value"})
+        df = db.get_data_file(pid, "batch_i2v")
+        assert df is not None
+        assert df["top_level"] == {"key": "value"}
+
+    def test_get_data_file_by_names(self, db):
+        pid = db.create_project("p1", "/p1")
+        db.create_data_file(pid, "batch_i2v", "i2v")
+        df = db.get_data_file_by_names("p1", "batch_i2v")
+        assert df is not None
+        assert df["name"] == "batch_i2v"
+
+    def test_get_nonexistent_data_file(self, db):
+        pid = db.create_project("p1", "/p1")
+        assert db.get_data_file(pid, "nope") is None
+
+    def test_unique_constraint(self, db):
+        pid = db.create_project("p1", "/p1")
+        db.create_data_file(pid, "batch_i2v", "i2v")
+        with pytest.raises(Exception):
+            db.create_data_file(pid, "batch_i2v", "vace")
+
+    def test_cascade_delete(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch_i2v", "i2v")
+        db.upsert_sequence(df_id, 1, {"prompt": "hello"})
+        db.save_history_tree(df_id, {"nodes": {}})
+        db.delete_project("p1")
+        assert db.get_data_file(pid, "batch_i2v") is None
+
+
+# ------------------------------------------------------------------
+# Sequences
+# ------------------------------------------------------------------
+
+class TestSequences:
+    def test_upsert_and_get(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        db.upsert_sequence(df_id, 1, {"prompt": "hello", "seed": 42})
+        data = db.get_sequence(df_id, 1)
+        assert data == {"prompt": "hello", "seed": 42}
+
+    def test_upsert_updates_existing(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        db.upsert_sequence(df_id, 1, {"prompt": "v1"})
+        db.upsert_sequence(df_id, 1, {"prompt": "v2"})
+        data = db.get_sequence(df_id, 1)
+        assert data["prompt"] == "v2"
+
+    def test_list_sequences(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        db.upsert_sequence(df_id, 3, {"a": 1})
+        db.upsert_sequence(df_id, 1, {"b": 2})
+        db.upsert_sequence(df_id, 2, {"c": 3})
+        seqs = db.list_sequences(df_id)
+        assert seqs == [1, 2, 3]
+
+    def test_get_nonexistent_sequence(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        assert db.get_sequence(df_id, 99) is None
+
+    def test_get_sequence_keys(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        db.upsert_sequence(df_id, 1, {
+            "prompt": "hello",
+            "seed": 42,
+            "cfg": 1.5,
+            "flag": True,
+        })
+        keys, types = db.get_sequence_keys(df_id, 1)
+        assert "prompt" in keys
+        assert "seed" in keys
+        idx_prompt = keys.index("prompt")
+        idx_seed = keys.index("seed")
+        idx_cfg = keys.index("cfg")
+        idx_flag = keys.index("flag")
+        assert types[idx_prompt] == "STRING"
+        assert types[idx_seed] == "INT"
+        assert types[idx_cfg] == "FLOAT"
+        assert types[idx_flag] == "STRING"  # bools -> STRING
+
+    def test_get_sequence_keys_nonexistent(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        keys, types = db.get_sequence_keys(df_id, 99)
+        assert keys == []
+        assert types == []
+
+    def test_delete_sequences_for_file(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        db.upsert_sequence(df_id, 1, {"a": 1})
+        db.upsert_sequence(df_id, 2, {"b": 2})
+        db.delete_sequences_for_file(df_id)
+        assert db.list_sequences(df_id) == []
+
+    def test_count_sequences(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        assert db.count_sequences(df_id) == 0
+        db.upsert_sequence(df_id, 1, {"a": 1})
+        db.upsert_sequence(df_id, 2, {"b": 2})
+        db.upsert_sequence(df_id, 3, {"c": 3})
+        assert db.count_sequences(df_id) == 3
+
+    def test_query_total_sequences(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        db.upsert_sequence(df_id, 1, {"a": 1})
+        db.upsert_sequence(df_id, 2, {"b": 2})
+        assert db.query_total_sequences("p1", "batch") == 2
+
+    def test_query_total_sequences_nonexistent(self, db):
+        assert db.query_total_sequences("nope", "nope") == 0
+
+
+# ------------------------------------------------------------------
+# History trees
+# ------------------------------------------------------------------
+
+class TestHistoryTrees:
+    def test_save_and_get(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        tree = {"nodes": {"abc": {"id": "abc"}}, "head_id": "abc"}
+        db.save_history_tree(df_id, tree)
+        result = db.get_history_tree(df_id)
+        assert result == tree
+
+    def test_upsert_updates(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        db.save_history_tree(df_id, {"v": 1})
+        db.save_history_tree(df_id, {"v": 2})
+        result = db.get_history_tree(df_id)
+        assert result == {"v": 2}
+
+    def test_get_nonexistent(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        assert db.get_history_tree(df_id) is None
+
+
+# ------------------------------------------------------------------
+# Import
+# ------------------------------------------------------------------
+
+class TestImport:
+    def test_import_json_file(self, db, tmp_path):
+        pid = db.create_project("p1", "/p1")
+        json_path = tmp_path / "batch_prompt_i2v.json"
+        data = {
+            KEY_BATCH_DATA: [
+                {"sequence_number": 1, "prompt": "hello", "seed": 42},
+                {"sequence_number": 2, "prompt": "world", "seed": 99},
+            ],
+            KEY_HISTORY_TREE: {"nodes": {}, "head_id": None},
+        }
+        json_path.write_text(json.dumps(data))
+
+        df_id = db.import_json_file(pid, json_path, "i2v")
+        assert df_id > 0
+
+        seqs = db.list_sequences(df_id)
+        assert seqs == [1, 2]
+
+        s1 = db.get_sequence(df_id, 1)
+        assert s1["prompt"] == "hello"
+        assert s1["seed"] == 42
+
+        tree = db.get_history_tree(df_id)
+        assert tree == {"nodes": {}, "head_id": None}
+
+    def test_import_file_name_from_stem(self, db, tmp_path):
+        pid = db.create_project("p1", "/p1")
+        json_path = tmp_path / "my_batch.json"
+        json_path.write_text(json.dumps({KEY_BATCH_DATA: [{"sequence_number": 1}]}))
+        db.import_json_file(pid, json_path)
+        df = db.get_data_file(pid, "my_batch")
+        assert df is not None
+
+    def test_import_no_batch_data(self, db, tmp_path):
+        pid = db.create_project("p1", "/p1")
+        json_path = tmp_path / "simple.json"
+        json_path.write_text(json.dumps({"prompt": "flat file"}))
+        df_id = db.import_json_file(pid, json_path)
+        seqs = db.list_sequences(df_id)
+        assert seqs == []
+
+    def test_reimport_updates_existing(self, db, tmp_path):
+        """Re-importing the same file should update data, not crash."""
+        pid = db.create_project("p1", "/p1")
+        json_path = tmp_path / "batch.json"
+
+        # First import
+        data_v1 = {KEY_BATCH_DATA: [{"sequence_number": 1, "prompt": "v1"}]}
+        json_path.write_text(json.dumps(data_v1))
+        df_id_1 = db.import_json_file(pid, json_path, "i2v")
+
+        # Second import (same file, updated data)
+        data_v2 = {KEY_BATCH_DATA: [{"sequence_number": 1, "prompt": "v2"}, {"sequence_number": 2, "prompt": "new"}]}
+        json_path.write_text(json.dumps(data_v2))
+        df_id_2 = db.import_json_file(pid, json_path, "vace")
+
+        # Should reuse the same data_file row
+        assert df_id_1 == df_id_2
+        # Data type should be updated
+        df = db.get_data_file(pid, "batch")
+        assert df["data_type"] == "vace"
+        # Sequences should reflect v2
+        seqs = db.list_sequences(df_id_2)
+        assert seqs == [1, 2]
+        s1 = db.get_sequence(df_id_2, 1)
+        assert s1["prompt"] == "v2"
+
+    def test_import_skips_non_dict_batch_items(self, db, tmp_path):
+        """Non-dict elements in batch_data should be silently skipped, not crash."""
+        pid = db.create_project("p1", "/p1")
+        json_path = tmp_path / "mixed.json"
+        data = {KEY_BATCH_DATA: [
+            {"sequence_number": 1, "prompt": "valid"},
+            "not a dict",
+            42,
+            None,
+            {"sequence_number": 3, "prompt": "also valid"},
+        ]}
+        json_path.write_text(json.dumps(data))
+        df_id = db.import_json_file(pid, json_path)
+
+        seqs = db.list_sequences(df_id)
+        assert seqs == [1, 3]
+
+    def test_import_atomic_on_error(self, db, tmp_path):
+        """If import fails partway, no partial data should be committed."""
+        pid = db.create_project("p1", "/p1")
+        json_path = tmp_path / "batch.json"
+        data = {KEY_BATCH_DATA: [{"sequence_number": 1, "prompt": "hello"}]}
+        json_path.write_text(json.dumps(data))
+        db.import_json_file(pid, json_path)
+
+        # Now try to import with bad data that will cause an error
+        # (overwrite the file with invalid sequence_number that causes int() to fail)
+        bad_data = {KEY_BATCH_DATA: [{"sequence_number": "not_a_number", "prompt": "bad"}]}
+        json_path.write_text(json.dumps(bad_data))
+        with pytest.raises(ValueError):
+            db.import_json_file(pid, json_path)
+
+        # Original data should still be intact (rollback worked)
+        df = db.get_data_file(pid, "batch")
+        assert df is not None
+        s1 = db.get_sequence(df["id"], 1)
+        assert s1["prompt"] == "hello"
+
+
+# ------------------------------------------------------------------
+# Query helpers
+# ------------------------------------------------------------------
+
+class TestQueryHelpers:
+    def test_query_sequence_data(self, db):
+        pid = db.create_project("myproject", "/mp")
+        df_id = db.create_data_file(pid, "batch_i2v", "i2v")
+        db.upsert_sequence(df_id, 1, {"prompt": "test", "seed": 7})
+        result = db.query_sequence_data("myproject", "batch_i2v", 1)
+        assert result == {"prompt": "test", "seed": 7}
+
+    def test_query_sequence_data_not_found(self, db):
+        assert db.query_sequence_data("nope", "nope", 1) is None
+
+    def test_query_sequence_keys(self, db):
+        pid = db.create_project("myproject", "/mp")
+        df_id = db.create_data_file(pid, "batch_i2v", "i2v")
+        db.upsert_sequence(df_id, 1, {"prompt": "test", "seed": 7})
+        keys, types = db.query_sequence_keys("myproject", "batch_i2v", 1)
+        assert "prompt" in keys
+        assert "seed" in keys
+
+    def test_list_project_files(self, db):
+        pid = db.create_project("p1", "/p1")
+        db.create_data_file(pid, "file_a", "i2v")
+        db.create_data_file(pid, "file_b", "vace")
+        files = db.list_project_files("p1")
+        assert len(files) == 2
+
+    def test_list_project_sequences(self, db):
+        pid = db.create_project("p1", "/p1")
+        df_id = db.create_data_file(pid, "batch", "generic")
+        db.upsert_sequence(df_id, 1, {})
+        db.upsert_sequence(df_id, 2, {})
+        seqs = db.list_project_sequences("p1", "batch")
+        assert seqs == [1, 2]
diff --git a/tests/test_json_loader.py b/tests/test_json_loader.py
deleted file mode 100644
index 2729e4d..0000000
--- a/tests/test_json_loader.py
+++ /dev/null
@@ -1,165 +0,0 @@
-import json
-import os
-
-import pytest
-
-from json_loader import (
-    to_float, to_int, get_batch_item, read_json_data,
-    JSONLoaderDynamic, MAX_DYNAMIC_OUTPUTS,
-)
-
-
-class TestToFloat:
-    def test_valid(self):
-        assert to_float("3.14") == 3.14
-        assert to_float(5) == 5.0
-
-    def test_invalid(self):
-        assert to_float("abc") == 0.0
-
-    def test_none(self):
-        assert to_float(None) == 0.0
-
-
-class TestToInt:
-    def test_valid(self):
-        assert to_int("7") == 7
-        assert to_int(3.9) == 3
-
-    def test_invalid(self):
-        assert to_int("xyz") == 0
-
-    def test_none(self):
-        assert to_int(None) == 0
-
-
-class TestGetBatchItem:
-    def test_lookup_by_sequence_number_field(self):
-        data = {"batch_data": [
-            {"sequence_number": 1, "a": "first"},
-            {"sequence_number": 5, "a": "fifth"},
-            {"sequence_number": 3, "a": "third"},
-        ]}
-        assert get_batch_item(data, 5) == {"sequence_number": 5, "a": "fifth"}
-        assert get_batch_item(data, 3) == {"sequence_number": 3, "a": "third"}
-
-    def test_fallback_to_index(self):
-        data = {"batch_data": [{"a": 1}, {"a": 2}, {"a": 3}]}
-        assert get_batch_item(data, 2) == {"a": 2}
-
-    def test_clamp_high(self):
-        data = {"batch_data": [{"a": 1}, {"a": 2}]}
-        assert get_batch_item(data, 99) == {"a": 2}
-
-    def test_clamp_low(self):
-        data = {"batch_data": [{"a": 1}, {"a": 2}]}
-        assert get_batch_item(data, 0) == {"a": 1}
-
-    def test_no_batch_data(self):
-        data = {"key": "val"}
-        assert get_batch_item(data, 1) == data
-
-
-class TestReadJsonData:
-    def test_missing_file(self, tmp_path):
-        assert read_json_data(str(tmp_path / "nope.json")) == {}
-
-    def test_invalid_json(self, tmp_path):
-        p = tmp_path / "bad.json"
-        p.write_text("{broken")
-        assert read_json_data(str(p)) == {}
-
-    def test_non_dict_json(self, tmp_path):
-        p = tmp_path / "list.json"
-        p.write_text(json.dumps([1, 2, 3]))
-        assert read_json_data(str(p)) == {}
-
-    def test_valid(self, tmp_path):
-        p = tmp_path / "ok.json"
-        p.write_text(json.dumps({"key": "val"}))
-        assert read_json_data(str(p)) == {"key": "val"}
-
-
-class TestJSONLoaderDynamic:
-    def _make_json(self, tmp_path, data):
-        p = tmp_path / "test.json"
-        p.write_text(json.dumps(data))
-        return str(p)
-
-    def test_known_keys(self, tmp_path):
-        path = self._make_json(tmp_path, {"name": "alice", "age": 30, "score": 9.5})
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(path, 1, output_keys="name,age,score")
-        assert result[0] == "alice"
-        assert result[1] == 30
-        assert result[2] == 9.5
-
-    def test_empty_output_keys(self, tmp_path):
-        path = self._make_json(tmp_path, {"name": "alice"})
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(path, 1, output_keys="")
-        assert len(result) == MAX_DYNAMIC_OUTPUTS
-        assert all(v == "" for v in result)
-
-    def test_pads_to_max(self, tmp_path):
-        path = self._make_json(tmp_path, {"a": "1", "b": "2"})
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(path, 1, output_keys="a,b")
-        assert len(result) == MAX_DYNAMIC_OUTPUTS
-        assert result[0] == "1"
-        assert result[1] == "2"
-        assert all(v == "" for v in result[2:])
-
-    def test_type_preservation_int(self, tmp_path):
-        path = self._make_json(tmp_path, {"count": 42})
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(path, 1, output_keys="count")
-        assert result[0] == 42
-        assert isinstance(result[0], int)
-
-    def test_type_preservation_float(self, tmp_path):
-        path = self._make_json(tmp_path, {"rate": 3.14})
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(path, 1, output_keys="rate")
-        assert result[0] == 3.14
-        assert isinstance(result[0], float)
-
-    def test_type_preservation_str(self, tmp_path):
-        path = self._make_json(tmp_path, {"label": "hello"})
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(path, 1, output_keys="label")
-        assert result[0] == "hello"
-        assert isinstance(result[0], str)
-
-    def test_bool_becomes_string(self, tmp_path):
-        path = self._make_json(tmp_path, {"flag": True, "off": False})
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(path, 1, output_keys="flag,off")
-        assert result[0] == "true"
-        assert result[1] == "false"
-        assert isinstance(result[0], str)
-
-    def test_missing_key_returns_empty_string(self, tmp_path):
-        path = self._make_json(tmp_path, {"a": "1"})
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(path, 1, output_keys="a,nonexistent")
-        assert result[0] == "1"
-        assert result[1] == ""
-
-    def test_missing_file_returns_all_empty(self, tmp_path):
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(str(tmp_path / "nope.json"), 1, output_keys="a,b")
-        assert len(result) == MAX_DYNAMIC_OUTPUTS
-        assert result[0] == ""
-        assert result[1] == ""
-
-    def test_batch_data(self, tmp_path):
-        path = self._make_json(tmp_path, {
-            "batch_data": [
-                {"sequence_number": 1, "x": "first"},
-                {"sequence_number": 2, "x": "second"},
-            ]
-        })
-        loader = JSONLoaderDynamic()
-        result = loader.load_dynamic(path, 2, output_keys="x")
-        assert result[0] == "second"
diff --git a/tests/test_project_loader.py b/tests/test_project_loader.py
new file mode 100644
index 0000000..dab80b2
--- /dev/null
+++ b/tests/test_project_loader.py
@@ -0,0 +1,211 @@
+import json
+from unittest.mock import patch, MagicMock
+from io import BytesIO
+
+import pytest
+
+from project_loader import (
+    ProjectLoaderDynamic,
+    _fetch_json,
+    _fetch_data,
+    _fetch_keys,
+    MAX_DYNAMIC_OUTPUTS,
+)
+
+
+def _mock_urlopen(data: dict):
+    """Create a mock context manager for urllib.request.urlopen."""
+    response = MagicMock()
+    response.read.return_value = json.dumps(data).encode()
+    response.__enter__ = lambda s: s
+    response.__exit__ = MagicMock(return_value=False)
+    return response
+
+
+class TestFetchHelpers:
+    def test_fetch_json_success(self):
+        data = {"key": "value"}
+        with patch("project_loader.urllib.request.urlopen", return_value=_mock_urlopen(data)):
+            result = _fetch_json("http://example.com/api")
+        assert result == data
+
+    def test_fetch_json_network_error(self):
+        with patch("project_loader.urllib.request.urlopen", side_effect=OSError("connection refused")):
+            result = _fetch_json("http://example.com/api")
+        assert result["error"] == "network_error"
+        assert "connection refused" in result["message"]
+
+    def test_fetch_json_http_error(self):
+        import urllib.error
+        err = urllib.error.HTTPError(
+            "http://example.com/api", 404, "Not Found", {},
+            BytesIO(json.dumps({"detail": "Project 'x' not found"}).encode())
+        )
+        with patch("project_loader.urllib.request.urlopen", side_effect=err):
+            result = _fetch_json("http://example.com/api")
+        assert result["error"] == "http_error"
+        assert result["status"] == 404
+        assert "not found" in result["message"].lower()
+
+    def test_fetch_data_builds_url(self):
+        data = {"prompt": "hello"}
+        with patch("project_loader.urllib.request.urlopen", return_value=_mock_urlopen(data)) as mock:
+            result = _fetch_data("http://localhost:8080", "proj1", "batch_i2v", 1)
+        assert result == data
+        called_url = mock.call_args[0][0]
+        assert "/api/projects/proj1/files/batch_i2v/data?seq=1" in called_url
+
+    def test_fetch_keys_builds_url(self):
+        data = {"keys": ["prompt"], "types": ["STRING"]}
+        with patch("project_loader.urllib.request.urlopen", return_value=_mock_urlopen(data)) as mock:
+            result = _fetch_keys("http://localhost:8080", "proj1", "batch_i2v", 1)
+        assert result == data
+        called_url = mock.call_args[0][0]
+        assert "/api/projects/proj1/files/batch_i2v/keys?seq=1" in called_url
+
+    def test_fetch_data_strips_trailing_slash(self):
+        data = {"prompt": "hello"}
+        with patch("project_loader.urllib.request.urlopen", return_value=_mock_urlopen(data)) as mock:
+            _fetch_data("http://localhost:8080/", "proj1", "file1", 1)
+        called_url = mock.call_args[0][0]
+        assert "//api" not in called_url
+
+    def test_fetch_data_encodes_special_chars(self):
+        """Project/file names with spaces or special chars should be percent-encoded."""
+        data = {"prompt": "hello"}
+        with patch("project_loader.urllib.request.urlopen", return_value=_mock_urlopen(data)) as mock:
+            _fetch_data("http://localhost:8080", "my project", "batch file", 1)
+        called_url = mock.call_args[0][0]
+        assert "my%20project" in called_url
+        assert "batch%20file" in called_url
+        assert " " not in called_url.split("?")[0]  # no raw spaces in path
+
+
+class TestProjectLoaderDynamic:
+    def _keys_meta(self, total=5):
+        return {"keys": [], "types": [], "total_sequences": total}
+
+    def test_load_dynamic_with_keys(self):
+        data = {"prompt": "hello", "seed": 42, "cfg": 1.5}
+        node = ProjectLoaderDynamic()
+        with patch("project_loader._fetch_keys", return_value=self._keys_meta()):
+            with patch("project_loader._fetch_data", return_value=data):
+                result = node.load_dynamic(
+                    "http://localhost:8080", "proj1", "batch_i2v", 1,
+                    output_keys="prompt,seed,cfg"
+                )
+        assert result[0] == 5  # total_sequences
+        assert result[1] == "hello"
+        assert result[2] == 42
+        assert result[3] == 1.5
+        assert len(result) == MAX_DYNAMIC_OUTPUTS + 1
+
+    def test_load_dynamic_with_json_encoded_keys(self):
+        """JSON-encoded output_keys should be parsed correctly."""
+        import json as _json
+        data = {"my,key": "comma_val", "normal": "ok"}
+        node = ProjectLoaderDynamic()
+        keys_json = _json.dumps(["my,key", "normal"])
+        with patch("project_loader._fetch_keys", return_value=self._keys_meta()):
+            with patch("project_loader._fetch_data", return_value=data):
+                result = node.load_dynamic(
+                    "http://localhost:8080", "proj1", "batch_i2v", 1,
+                    output_keys=keys_json
+                )
+        assert result[1] == "comma_val"
+        assert result[2] == "ok"
+
+    def test_load_dynamic_type_coercion(self):
+        """output_types should coerce values to declared types."""
+        import json as _json
+        data = {"seed": "42", "cfg": "1.5", "prompt": "hello"}
+        node = ProjectLoaderDynamic()
+        keys_json = _json.dumps(["seed", "cfg", "prompt"])
+        types_json = _json.dumps(["INT", "FLOAT", "STRING"])
+        with patch("project_loader._fetch_keys", return_value=self._keys_meta()):
+            with patch("project_loader._fetch_data", return_value=data):
+                result = node.load_dynamic(
+                    "http://localhost:8080", "proj1", "batch_i2v", 1,
+                    output_keys=keys_json, output_types=types_json
+                )
+        assert result[1] == 42       # string "42" coerced to int
+        assert result[2] == 1.5      # string "1.5" coerced to float
+        assert result[3] == "hello"  # string stays string
+
+    def test_load_dynamic_empty_keys(self):
+        node = ProjectLoaderDynamic()
+        with patch("project_loader._fetch_keys", return_value=self._keys_meta()):
+            with patch("project_loader._fetch_data", return_value={"prompt": "hello"}):
+                result = node.load_dynamic(
+                    "http://localhost:8080", "proj1", "batch_i2v", 1,
+                    output_keys=""
+                )
+        # Slot 0 is total_sequences (INT), rest are empty strings
+        assert result[0] == 5
+        assert all(v == "" for v in result[1:])
+
+    def test_load_dynamic_missing_key(self):
+        node = ProjectLoaderDynamic()
+        with patch("project_loader._fetch_keys", return_value=self._keys_meta()):
+            with patch("project_loader._fetch_data", return_value={"prompt": "hello"}):
+                result = node.load_dynamic(
+                    "http://localhost:8080", "proj1", "batch_i2v", 1,
+                    output_keys="nonexistent"
+                )
+        assert result[1] == ""
+
+    def test_load_dynamic_bool_becomes_string(self):
+        node = ProjectLoaderDynamic()
+        with patch("project_loader._fetch_keys", return_value=self._keys_meta()):
+            with patch("project_loader._fetch_data", return_value={"flag": True}):
+                result = node.load_dynamic(
+                    "http://localhost:8080", "proj1", "batch_i2v", 1,
+                    output_keys="flag"
+                )
+        assert result[1] == "true"
+
+    def test_load_dynamic_returns_total_sequences(self):
+        """total_sequences should be the first output from keys metadata."""
+        node = ProjectLoaderDynamic()
+        with patch("project_loader._fetch_keys", return_value={"keys": [], "types": [], "total_sequences": 42}):
+            with patch("project_loader._fetch_data", return_value={}):
+                result = node.load_dynamic(
+                    "http://localhost:8080", "proj1", "batch_i2v", 1,
+                    output_keys=""
+                )
+        assert result[0] == 42
+
+    def test_load_dynamic_raises_on_network_error(self):
+        """Network errors from _fetch_keys should raise RuntimeError."""
+        node = ProjectLoaderDynamic()
+        error_resp = {"error": "network_error", "message": "Connection refused"}
+        with patch("project_loader._fetch_keys", return_value=error_resp):
+            with pytest.raises(RuntimeError, match="Failed to fetch project keys"):
+                node.load_dynamic("http://localhost:8080", "proj1", "batch", 1)
+
+    def test_load_dynamic_raises_on_data_fetch_error(self):
+        """Network errors from _fetch_data should raise RuntimeError."""
+        node = ProjectLoaderDynamic()
+        error_resp = {"error": "http_error", "status": 404, "message": "Sequence not found"}
+        with patch("project_loader._fetch_keys", return_value=self._keys_meta()):
+            with patch("project_loader._fetch_data", return_value=error_resp):
+                with pytest.raises(RuntimeError, match="Failed to fetch sequence data"):
+                    node.load_dynamic("http://localhost:8080", "proj1", "batch", 1)
+
+    def test_input_types_has_manager_url(self):
+        inputs = ProjectLoaderDynamic.INPUT_TYPES()
+        assert "manager_url" in inputs["required"]
+        assert "project_name" in inputs["required"]
+        assert "file_name" in inputs["required"]
+        assert "sequence_number" in inputs["required"]
+
+    def test_category(self):
+        assert ProjectLoaderDynamic.CATEGORY == "utils/json/project"
+
+
+class TestNodeMappings:
+    def test_mappings_exist(self):
+        from project_loader import PROJECT_NODE_CLASS_MAPPINGS, PROJECT_NODE_DISPLAY_NAME_MAPPINGS
+        assert "ProjectLoaderDynamic" in PROJECT_NODE_CLASS_MAPPINGS
+        assert len(PROJECT_NODE_CLASS_MAPPINGS) == 1
+        assert len(PROJECT_NODE_DISPLAY_NAME_MAPPINGS) == 1
diff --git a/utils.py b/utils.py
index 2e49007..af80c60 100644
--- a/utils.py
+++ b/utils.py
@@ -160,6 +160,80 @@ def get_file_mtime(path: str | Path) -> float:
         return path.stat().st_mtime
     return 0
 
+def sync_to_db(db, project_name: str, file_path: Path, data: dict) -> None:
+    """Dual-write helper: sync JSON data to the project database.
+
+    Resolves (or creates) the data_file, upserts all sequences from batch_data,
+    and saves the history_tree. All writes happen in a single transaction.
+    """
+    if not db or not project_name:
+        return
+    try:
+        proj = db.get_project(project_name)
+        if not proj:
+            return
+        file_name = Path(file_path).stem
+
+        # Use a single transaction for atomicity
+        db.conn.execute("BEGIN IMMEDIATE")
+        try:
+            df = db.get_data_file(proj["id"], file_name)
+            top_level = {k: v for k, v in data.items()
+                         if k not in (KEY_BATCH_DATA, KEY_HISTORY_TREE)}
+            if not df:
+                now = __import__('time').time()
+                cur = db.conn.execute(
+                    "INSERT INTO data_files (project_id, name, data_type, top_level, created_at, updated_at) "
+                    "VALUES (?, ?, ?, ?, ?, ?)",
+                    (proj["id"], file_name, "generic", json.dumps(top_level), now, now),
+                )
+                df_id = cur.lastrowid
+            else:
+                df_id = df["id"]
+                # Update top_level metadata
+                now = __import__('time').time()
+                db.conn.execute(
+                    "UPDATE data_files SET top_level = ?, updated_at = ? WHERE id = ?",
+                    (json.dumps(top_level), now, df_id),
+                )
+
+            # Sync sequences
+            batch_data = data.get(KEY_BATCH_DATA, [])
+            if isinstance(batch_data, list):
+                db.conn.execute("DELETE FROM sequences WHERE data_file_id = ?", (df_id,))
+                for item in batch_data:
+                    if not isinstance(item, dict):
+                        continue
+                    seq_num = int(item.get(KEY_SEQUENCE_NUMBER, 0))
+                    now = __import__('time').time()
+                    db.conn.execute(
+                        "INSERT INTO sequences (data_file_id, sequence_number, data, updated_at) "
+                        "VALUES (?, ?, ?, ?)",
+                        (df_id, seq_num, json.dumps(item), now),
+                    )
+
+            # Sync history tree
+            history_tree = data.get(KEY_HISTORY_TREE)
+            if history_tree and isinstance(history_tree, dict):
+                now = __import__('time').time()
+                db.conn.execute(
+                    "INSERT INTO history_trees (data_file_id, tree_data, updated_at) "
+                    "VALUES (?, ?, ?) "
+                    "ON CONFLICT(data_file_id) DO UPDATE SET tree_data=excluded.tree_data, updated_at=excluded.updated_at",
+                    (df_id, json.dumps(history_tree), now),
+                )
+
+            db.conn.execute("COMMIT")
+        except Exception:
+            try:
+                db.conn.execute("ROLLBACK")
+            except Exception:
+                pass
+            raise
+    except Exception as e:
+        logger.warning(f"sync_to_db failed: {e}")
+
+
 def generate_templates(current_dir: Path) -> None:
     """Creates batch template files if folder is empty."""
     first = DEFAULTS.copy()
diff --git a/web/json_dynamic.js b/web/json_dynamic.js
deleted file mode 100644
index 81e11f1..0000000
--- a/web/json_dynamic.js
+++ /dev/null
@@ -1,140 +0,0 @@
-import { app } from "../../scripts/app.js";
-import { api } from "../../scripts/api.js";
-
-app.registerExtension({
-    name: "json.manager.dynamic",
-
-    async beforeRegisterNodeDef(nodeType, nodeData, app) {
-        if (nodeData.name !== "JSONLoaderDynamic") return;
-
-        const origOnNodeCreated = nodeType.prototype.onNodeCreated;
-        nodeType.prototype.onNodeCreated = function () {
-            origOnNodeCreated?.apply(this, arguments);
-
-            // Hide internal widgets (managed by JS)
-            for (const name of ["output_keys", "output_types"]) {
-                const w = this.widgets?.find(w => w.name === name);
-                if (w) { w.type = "hidden"; w.computeSize = () => [0, -4]; }
-            }
-
-            // Remove all 32 default outputs from Python RETURN_TYPES
-            while (this.outputs.length > 0) {
-                this.removeOutput(0);
-            }
-
-            // Add Refresh button
-            this.addWidget("button", "Refresh Outputs", null, () => {
-                this.refreshDynamicOutputs();
-            });
-
-            this.setSize(this.computeSize());
-        };
-
-        nodeType.prototype.refreshDynamicOutputs = async function () {
-            const pathWidget = this.widgets?.find(w => w.name === "json_path");
-            const seqWidget = this.widgets?.find(w => w.name === "sequence_number");
-            if (!pathWidget?.value) return;
-
-            try {
-                const resp = await api.fetchApi(
-                    `/json_manager/get_keys?path=${encodeURIComponent(pathWidget.value)}&sequence_number=${seqWidget?.value || 1}`
-                );
-                const { keys, types } = await resp.json();
-
-                // Store keys and types in hidden widgets for persistence
-                const okWidget = this.widgets?.find(w => w.name === "output_keys");
-                if (okWidget) okWidget.value = keys.join(",");
-                const otWidget = this.widgets?.find(w => w.name === "output_types");
-                if (otWidget) otWidget.value = types.join(",");
-
-                // Build a map of current output names to slot indices
-                const oldSlots = {};
-                for (let i = 0; i < this.outputs.length; i++) {
-                    oldSlots[this.outputs[i].name] = i;
-                }
-
-                // Build new outputs, reusing existing slots to preserve links
-                const newOutputs = [];
-                for (let k = 0; k < keys.length; k++) {
-                    const key = keys[k];
-                    const type = types[k] || "*";
-                    if (key in oldSlots) {
-                        // Reuse existing slot object (keeps links intact)
-                        const slot = this.outputs[oldSlots[key]];
-                        slot.type = type;
-                        newOutputs.push(slot);
-                        delete oldSlots[key];
-                    } else {
-                        // New key — create a fresh slot
-                        newOutputs.push({ name: key, type: type, links: null });
-                    }
-                }
-
-                // Disconnect links on slots that are being removed
-                for (const name in oldSlots) {
-                    const idx = oldSlots[name];
-                    if (this.outputs[idx]?.links?.length) {
-                        for (const linkId of [...this.outputs[idx].links]) {
-                            this.graph?.removeLink(linkId);
-                        }
-                    }
-                }
-
-                // Reassign the outputs array and fix link slot indices
-                this.outputs = newOutputs;
-                // Update link origin_slot to match new positions
-                if (this.graph) {
-                    for (let i = 0; i < this.outputs.length; i++) {
-                        const links = this.outputs[i].links;
-                        if (!links) continue;
-                        for (const linkId of links) {
-                            const link = this.graph.links[linkId];
-                            if (link) link.origin_slot = i;
-                        }
-                    }
-                }
-
-                this.setSize(this.computeSize());
-                app.graph.setDirtyCanvas(true, true);
-            } catch (e) {
-                console.error("[JSONLoaderDynamic] Refresh failed:", e);
-            }
-        };
-
-        // Restore state on workflow load
-        const origOnConfigure = nodeType.prototype.onConfigure;
-        nodeType.prototype.onConfigure = function (info) {
-            origOnConfigure?.apply(this, arguments);
-
-            // Hide internal widgets
-            for (const name of ["output_keys", "output_types"]) {
-                const w = this.widgets?.find(w => w.name === name);
-                if (w) { w.type = "hidden"; w.computeSize = () => [0, -4]; }
-            }
-
-            const okWidget = this.widgets?.find(w => w.name === "output_keys");
-            const otWidget = this.widgets?.find(w => w.name === "output_types");
-
-            const keys = okWidget?.value
-                ? okWidget.value.split(",").filter(k => k.trim())
-                : [];
-            const types = otWidget?.value
-                ? otWidget.value.split(",")
-                : [];
-
-            // On load, LiteGraph already restored serialized outputs with links.
-            // Rename and set types to match stored state (preserves links).
-            for (let i = 0; i < this.outputs.length && i < keys.length; i++) {
-                this.outputs[i].name = keys[i].trim();
-                if (types[i]) this.outputs[i].type = types[i];
-            }
-
-            // Remove any extra outputs beyond the key count
-            while (this.outputs.length > keys.length) {
-                this.removeOutput(this.outputs.length - 1);
-            }
-
-            this.setSize(this.computeSize());
-        };
-    },
-});
diff --git a/web/project_dynamic.js b/web/project_dynamic.js
new file mode 100644
index 0000000..7f2de6e
--- /dev/null
+++ b/web/project_dynamic.js
@@ -0,0 +1,255 @@
+import { app } from "../../scripts/app.js";
+import { api } from "../../scripts/api.js";
+
+app.registerExtension({
+    name: "json.manager.project.dynamic",
+
+    async beforeRegisterNodeDef(nodeType, nodeData, app) {
+        if (nodeData.name !== "ProjectLoaderDynamic") return;
+
+        const origOnNodeCreated = nodeType.prototype.onNodeCreated;
+        nodeType.prototype.onNodeCreated = function () {
+            origOnNodeCreated?.apply(this, arguments);
+
+            // Hide internal widgets (managed by JS)
+            for (const name of ["output_keys", "output_types"]) {
+                const w = this.widgets?.find(w => w.name === name);
+                if (w) { w.type = "hidden"; w.computeSize = () => [0, -4]; }
+            }
+
+            // Do NOT remove default outputs synchronously here.
+            // During graph loading, ComfyUI creates all nodes (firing onNodeCreated)
+            // before configuring them. Other nodes (e.g. Kijai Set/Get) may resolve
+            // links to our outputs during their configure step. If we remove outputs
+            // here, those nodes find no output slot and error out.
+            //
+            // Instead, defer cleanup: for loaded workflows onConfigure sets _configured
+            // before this runs; for new nodes the defaults are cleaned up.
+            this._configured = false;
+
+            // Add Refresh button
+            this.addWidget("button", "Refresh Outputs", null, () => {
+                this.refreshDynamicOutputs();
+            });
+
+            // Auto-refresh with 500ms debounce on widget changes
+            this._refreshTimer = null;
+            const autoRefreshWidgets = ["project_name", "file_name", "sequence_number"];
+            for (const widgetName of autoRefreshWidgets) {
+                const w = this.widgets?.find(w => w.name === widgetName);
+                if (w) {
+                    const origCallback = w.callback;
+                    const node = this;
+                    w.callback = function (...args) {
+                        origCallback?.apply(this, args);
+                        clearTimeout(node._refreshTimer);
+                        node._refreshTimer = setTimeout(() => {
+                            node.refreshDynamicOutputs();
+                        }, 500);
+                    };
+                }
+            }
+
+            queueMicrotask(() => {
+                if (!this._configured) {
+                    // New node (not loading) — remove the Python default outputs
+                    // and add only the fixed total_sequences slot
+                    while (this.outputs.length > 0) {
+                        this.removeOutput(0);
+                    }
+                    this.addOutput("total_sequences", "INT");
+                    this.setSize(this.computeSize());
+                    app.graph?.setDirtyCanvas(true, true);
+                }
+            });
+        };
+
+        nodeType.prototype._setStatus = function (status, message) {
+            const baseTitle = "Project Loader (Dynamic)";
+            if (status === "ok") {
+                this.title = baseTitle;
+                this.color = undefined;
+                this.bgcolor = undefined;
+            } else if (status === "error") {
+                this.title = baseTitle + " - ERROR";
+                this.color = "#ff4444";
+                this.bgcolor = "#331111";
+                if (message) this.title = baseTitle + ": " + message;
+            } else if (status === "loading") {
+                this.title = baseTitle + " - Loading...";
+            }
+            app.graph?.setDirtyCanvas(true, true);
+        };
+
+        nodeType.prototype.refreshDynamicOutputs = async function () {
+            const urlWidget = this.widgets?.find(w => w.name === "manager_url");
+            const projectWidget = this.widgets?.find(w => w.name === "project_name");
+            const fileWidget = this.widgets?.find(w => w.name === "file_name");
+            const seqWidget = this.widgets?.find(w => w.name === "sequence_number");
+
+            if (!urlWidget?.value || !projectWidget?.value || !fileWidget?.value) return;
+
+            this._setStatus("loading");
+
+            try {
+                const resp = await api.fetchApi(
+                    `/json_manager/get_project_keys?url=${encodeURIComponent(urlWidget.value)}&project=${encodeURIComponent(projectWidget.value)}&file=${encodeURIComponent(fileWidget.value)}&seq=${seqWidget?.value || 1}`
+                );
+
+                if (!resp.ok) {
+                    let errorMsg = `HTTP ${resp.status}`;
+                    try {
+                        const errData = await resp.json();
+                        if (errData.message) errorMsg = errData.message;
+                    } catch (_) {}
+                    this._setStatus("error", errorMsg);
+                    return;
+                }
+
+                const data = await resp.json();
+                const keys = data.keys;
+                const types = data.types;
+
+                // If the API returned an error or missing data, keep existing outputs and links intact
+                if (data.error || !Array.isArray(keys) || !Array.isArray(types)) {
+                    const errMsg = data.error ? data.message || data.error : "Missing keys/types";
+                    this._setStatus("error", errMsg);
+                    return;
+                }
+
+                // Store keys and types in hidden widgets for persistence (comma-separated)
+                const okWidget = this.widgets?.find(w => w.name === "output_keys");
+                if (okWidget) okWidget.value = keys.join(",");
+                const otWidget = this.widgets?.find(w => w.name === "output_types");
+                if (otWidget) otWidget.value = types.join(",");
+
+                // Slot 0 is always total_sequences (INT) — ensure it exists
+                if (this.outputs.length === 0 || this.outputs[0].name !== "total_sequences") {
+                    this.outputs.unshift({ name: "total_sequences", type: "INT", links: null });
+                }
+                this.outputs[0].type = "INT";
+
+                // Build a map of current dynamic output names to slot indices (skip slot 0)
+                const oldSlots = {};
+                for (let i = 1; i < this.outputs.length; i++) {
+                    oldSlots[this.outputs[i].name] = i;
+                }
+
+                // Build new dynamic outputs, reusing existing slots to preserve links
+                const newOutputs = [this.outputs[0]]; // Keep total_sequences at slot 0
+                for (let k = 0; k < keys.length; k++) {
+                    const key = keys[k];
+                    const type = types[k] || "*";
+                    if (key in oldSlots) {
+                        const slot = this.outputs[oldSlots[key]];
+                        slot.type = type;
+                        slot.label = key;
+                        newOutputs.push(slot);
+                        delete oldSlots[key];
+                    } else {
+                        newOutputs.push({ name: key, label: key, type: type, links: null });
+                    }
+                }
+
+                // Disconnect links on slots that are being removed
+                for (const name in oldSlots) {
+                    const idx = oldSlots[name];
+                    if (this.outputs[idx]?.links?.length) {
+                        for (const linkId of [...this.outputs[idx].links]) {
+                            this.graph?.removeLink(linkId);
+                        }
+                    }
+                }
+
+                // Reassign the outputs array and fix link slot indices
+                this.outputs = newOutputs;
+                if (this.graph) {
+                    for (let i = 0; i < this.outputs.length; i++) {
+                        const links = this.outputs[i].links;
+                        if (!links) continue;
+                        for (const linkId of links) {
+                            const link = this.graph.links[linkId];
+                            if (link) link.origin_slot = i;
+                        }
+                    }
+                }
+
+                this._setStatus("ok");
+                this.setSize(this.computeSize());
+                app.graph?.setDirtyCanvas(true, true);
+            } catch (e) {
+                console.error("[ProjectLoaderDynamic] Refresh failed:", e);
+                this._setStatus("error", "Server unreachable");
+            }
+        };
+
+        // Restore state on workflow load
+        const origOnConfigure = nodeType.prototype.onConfigure;
+        nodeType.prototype.onConfigure = function (info) {
+            origOnConfigure?.apply(this, arguments);
+            this._configured = true;
+
+            // Hide internal widgets
+            for (const name of ["output_keys", "output_types"]) {
+                const w = this.widgets?.find(w => w.name === name);
+                if (w) { w.type = "hidden"; w.computeSize = () => [0, -4]; }
+            }
+
+            const okWidget = this.widgets?.find(w => w.name === "output_keys");
+            const otWidget = this.widgets?.find(w => w.name === "output_types");
+
+            const keys = okWidget?.value
+                ? okWidget.value.split(",").filter(k => k.trim())
+                : [];
+            const types = otWidget?.value
+                ? otWidget.value.split(",")
+                : [];
+
+            // Ensure slot 0 is total_sequences (INT)
+            if (this.outputs.length === 0 || this.outputs[0].name !== "total_sequences") {
+                this.outputs.unshift({ name: "total_sequences", type: "INT", links: null });
+                const node = this;
+                queueMicrotask(() => {
+                    if (!node.graph) return;
+                    for (const output of node.outputs) {
+                        output.links = null;
+                    }
+                    for (const linkId in node.graph.links) {
+                        const link = node.graph.links[linkId];
+                        if (!link || link.origin_id !== node.id) continue;
+                        link.origin_slot += 1;
+                        const output = node.outputs[link.origin_slot];
+                        if (output) {
+                            if (!output.links) output.links = [];
+                            output.links.push(link.id);
+                        }
+                    }
+                    app.graph?.setDirtyCanvas(true, true);
+                });
+            }
+            this.outputs[0].type = "INT";
+            this.outputs[0].name = "total_sequences";
+
+            if (keys.length > 0) {
+                for (let i = 0; i < keys.length; i++) {
+                    const slotIdx = i + 1;
+                    if (slotIdx < this.outputs.length) {
+                        this.outputs[slotIdx].name = keys[i].trim();
+                        this.outputs[slotIdx].label = keys[i].trim();
+                        if (types[i]) this.outputs[slotIdx].type = types[i];
+                    }
+                }
+                while (this.outputs.length > keys.length + 1) {
+                    this.removeOutput(this.outputs.length - 1);
+                }
+            } else if (this.outputs.length > 1) {
+                // Widget values empty but serialized dynamic outputs exist — sync widgets
+                const dynamicOutputs = this.outputs.slice(1);
+                if (okWidget) okWidget.value = dynamicOutputs.map(o => o.name).join(",");
+                if (otWidget) otWidget.value = dynamicOutputs.map(o => o.type).join(",");
+            }
+
+            this.setSize(this.computeSize());
+        };
+    },
+});