Merge remote-tracking branch 'forgejo/master'

New 'mosaic ingest' CLI command bypasses HTTP collection to import
raw documents (PDF, HTML, text) or pre-extracted JSONL data directly
into the pipeline.

Modes:
- Single/multiple files: -f file1.pdf -f file2.html -s source_name
- Directory bulk import: -d /path/to/dump/ -s source_name -r
- JSONL extracted data: -f data.jsonl -s name -c tools
- Real-time watch: -d /incoming/ -s name -w (polls every 2s)
- Auto-parse: --parse flag runs parser on newly ingested docs

Files are hashed, deduplicated, and stored in the cache using the
same hash-sharded scheme as HTTP-collected documents.

DevTrack: mosaic #498

mosaic.py

@@ -654,6 +654,259 @@ def export_cmd(ctx, category, fmt):
         db.close()
 
 
+# ---------------------------------------------------------------------------
+# ingest
+# ---------------------------------------------------------------------------
+
+MIME_TO_DOCTYPE = {
+    'application/pdf': 'pdf',
+    'text/html': 'html',
+    'text/plain': 'text',
+    'message/rfc822': 'email',
+}
+
+
+def _detect_doc_type(file_path: str) -> str:
+    """Detect document type from file content."""
+    from utils.sanitize import validate_content_type
+    mime = validate_content_type(file_path)
+    return MIME_TO_DOCTYPE.get(mime, 'text')
+
+
+def _hash_file(file_path: str) -> str:
+    """Compute SHA256 of a file."""
+    sha = hashlib.sha256()
+    with open(file_path, 'rb') as f:
+        for chunk in iter(lambda: f.read(8192), b''):
+            sha.update(chunk)
+    return sha.hexdigest()
+
+
+def _ingest_raw_file(db, cache, file_path: str, source_name: str) -> bool:
+    """Ingest a single raw document file into the DB and cache.
+
+    Returns True if the document was new, False if already exists.
+    """
+    from storage.models import Document
+
+    fpath = Path(file_path)
+    content_hash = _hash_file(str(fpath))
+
+    existing = db.get_document(content_hash)
+    if existing:
+        return False
+
+    doc_type = _detect_doc_type(str(fpath))
+    ext = fpath.suffix.lstrip('.')
+
+    # Store in cache
+    cached_path = cache.get_shard_path(content_hash, source_name, ext=ext)
+    cached_path.parent.mkdir(parents=True, exist_ok=True)
+    import shutil
+    shutil.copy2(str(fpath), str(cached_path))
+
+    doc = Document(
+        id=content_hash,
+        source=source_name,
+        source_url=str(fpath.resolve()),
+        doc_type=doc_type,
+        title=fpath.stem,
+        raw_path=str(cached_path),
+        content_hash=content_hash,
+        fetch_date=datetime.now().isoformat(),
+        char_count=fpath.stat().st_size,
+        status='CACHED',
+        parse_status='PENDING',
+    )
+    db.insert_document(doc)
+    return True
+
+
+@cli.command()
+@click.option("--file", "-f", "files", multiple=True, type=click.Path(exists=True),
+              help="File(s) to ingest. Repeat for multiple files.")
+@click.option("--dir", "-d", "directory", type=click.Path(exists=True, file_okay=False),
+              help="Directory of files to ingest.")
+@click.option("--source", "-s", type=str, required=True, help="Source name to tag ingested docs.")
+@click.option("--category", "-c", type=click.Choice(EXPORT_CATEGORIES, case_sensitive=False),
+              default=None, help="JSONL category (for importing pre-extracted data).")
+@click.option("--watch", "-w", is_flag=True, default=False,
+              help="Watch directory for new files (real-time mode).")
+@click.option("--recursive", "-r", is_flag=True, default=False,
+              help="Recurse into subdirectories.")
+@click.option("--parse", "auto_parse", is_flag=True, default=False,
+              help="Automatically parse ingested documents.")
+@click.pass_context
+def ingest(ctx, files, directory, source, category, watch, recursive, auto_parse):
+    """Ingest local files or data dumps into the Mosaic pipeline.
+
+    Import raw documents (PDF, HTML, text) or pre-extracted JSONL data
+    without HTTP collection. Supports single files, directories, and
+    real-time directory watching.
+
+    \b
+    Examples:
+      mosaic ingest -f dump.pdf -s wikileaks_dump
+      mosaic ingest -d /path/to/cables/ -s cablegate -r --parse
+      mosaic ingest -f extracted.jsonl -s vault7 -c tools
+      mosaic ingest -d /incoming/ -s live_feed -w
+    """
+    db_path = ctx.obj["db_path"]
+    output_dir = ctx.obj["output_dir"]
+    db = open_db(db_path)
+
+    try:
+        # JSONL import mode
+        if category:
+            store = open_jsonl_store(db, output_dir)
+            total = 0
+            targets = list(files)
+            if directory:
+                p = Path(directory)
+                targets.extend(str(f) for f in p.glob("*.jsonl"))
+
+            if not targets:
+                console.print("[yellow]No JSONL files to import.[/yellow]")
+                return
+
+            for fpath in targets:
+                with console.status(f"[bold green]Importing {Path(fpath).name}..."):
+                    count = store.import_jsonl(fpath, category)
+                console.print(f"  [green]+{count}[/green] {category} from {Path(fpath).name}")
+                total += count
+
+            console.print(f"\n[bold green]Imported {total} {category} records total.[/bold green]")
+            return
+
+        # Raw document ingest mode
+        cache = open_cache(output_dir)
+        file_list = list(files)
+
+        if directory:
+            p = Path(directory)
+            pattern = "**/*" if recursive else "*"
+            file_list.extend(
+                str(f) for f in p.glob(pattern)
+                if f.is_file() and not f.name.startswith('.')
+            )
+
+        if watch and directory:
+            _watch_directory(db, cache, directory, source, recursive, auto_parse, ctx)
+            return
+
+        if not file_list:
+            console.print("[yellow]No files to ingest.[/yellow] Use --file, --dir, or --watch.")
+            return
+
+        console.print(Panel(
+            f"[bold green]Ingest[/bold green]\n\n"
+            f"  Source : {source}\n"
+            f"  Files  : {len(file_list)}\n"
+            f"  Parse  : {'yes' if auto_parse else 'no'}",
+            title="[bold]mosaic ingest[/bold]",
+            border_style="green",
+        ))
+
+        new_count = 0
+        skip_count = 0
+
+        with Progress(
+            SpinnerColumn(), TextColumn("{task.description}"),
+            BarColumn(), TextColumn("{task.completed}/{task.total}"),
+        ) as progress:
+            task = progress.add_task("Ingesting files", total=len(file_list))
+            for fpath in file_list:
+                is_new = _ingest_raw_file(db, cache, fpath, source)
+                if is_new:
+                    new_count += 1
+                else:
+                    skip_count += 1
+                progress.advance(task)
+
+        table = Table(title="Ingest Results", box=box.ROUNDED)
+        table.add_column("Metric", style="bold")
+        table.add_column("Count", justify="right")
+        table.add_row("[green]New documents[/green]", str(new_count))
+        table.add_row("[dim]Already cached[/dim]", str(skip_count))
+        table.add_row("[bold]Total processed[/bold]", str(new_count + skip_count))
+        console.print(table)
+
+        # Auto-parse if requested
+        if auto_parse and new_count > 0:
+            console.print("\n[bold]Running parser on new documents...[/bold]")
+            ctx.invoke(parse, source=source)
+
+    finally:
+        db.close()
+
+
+def _watch_directory(db, cache, directory: str, source: str,
+                     recursive: bool, auto_parse: bool, ctx):
+    """Watch a directory for new files and ingest them in real-time."""
+    import time as _time
+
+    watched = Path(directory)
+    seen = set()
+
+    # Seed with existing files
+    pattern = "**/*" if recursive else "*"
+    for f in watched.glob(pattern):
+        if f.is_file() and not f.name.startswith('.'):
+            seen.add(str(f.resolve()))
+
+    console.print(Panel(
+        f"[bold cyan]Watch Mode[/bold cyan]\n\n"
+        f"  Directory : {directory}\n"
+        f"  Source    : {source}\n"
+        f"  Recursive : {'yes' if recursive else 'no'}\n"
+        f"  Auto-parse: {'yes' if auto_parse else 'no'}\n\n"
+        f"  [dim]Press Ctrl+C to stop watching.[/dim]",
+        title="[bold]mosaic ingest --watch[/bold]",
+        border_style="cyan",
+    ))
+    console.print(f"  [dim]Watching {len(seen)} existing files (skipped)...[/dim]")
+
+    try:
+        while True:
+            new_files = []
+            for f in watched.glob(pattern):
+                resolved = str(f.resolve())
+                if f.is_file() and not f.name.startswith('.') and resolved not in seen:
+                    new_files.append(f)
+                    seen.add(resolved)
+
+            for fpath in new_files:
+                is_new = _ingest_raw_file(db, cache, str(fpath), source)
+                if is_new:
+                    console.print(f"  [green]+[/green] {fpath.name}")
+                    if auto_parse:
+                        from parsers.base import ParserManager
+                        from storage.models import Document
+                        cfg = ctx.obj["cfg"]
+                        parser = ParserManager(
+                            memory_limit_mb=int(cfg.get('parser', 'subprocess_memory_limit_mb', fallback='512')),
+                            timeout_seconds=int(cfg.get('parser', 'subprocess_timeout_seconds', fallback='60')),
+                        )
+                        doc = db.get_document(_hash_file(str(fpath)))
+                        if doc and doc.parse_status == 'PENDING':
+                            result = parser.parse(doc.raw_path, doc.doc_type, doc.content_hash)
+                            if result.success:
+                                doc.text = result.text
+                                doc.title = result.title or doc.title
+                                doc.char_count = result.char_count
+                                ps = 'NEEDS_OCR' if result.needs_ocr else 'PARSED'
+                                db.insert_document(doc)
+                                db.update_document_status(doc.id, parse_status=ps)
+                                console.print(f"    [dim]parsed → {ps}[/dim]")
+                            else:
+                                db.update_document_status(doc.id, parse_status='PARSE_FAILED')
+                                console.print(f"    [red]parse failed[/red]")
+
+            _time.sleep(2)
+    except KeyboardInterrupt:
+        console.print("\n[dim]Watch stopped.[/dim]")
+
+
 # ---------------------------------------------------------------------------
 # status
 # ---------------------------------------------------------------------------
@@ -901,9 +1154,10 @@ MENU_ITEMS = [
     ("4", "Analyze", "Full pipeline", "analyze"),
     ("5", "Query", "Interactive Q&A", "query"),
     ("6", "Export", "Export to JSONL", "export"),
-    ("7", "Status", "Progress dashboard", "status"),
-    ("8", "Verify", "Integrity check", "verify"),
-    ("9", "Sources", "List profiles", "sources"),
+    ("7", "Ingest", "Import local files/dumps", "ingest"),
+    ("8", "Status", "Progress dashboard", "status"),
+    ("9", "Verify", "Integrity check", "verify"),
+    ("s", "Sources", "List profiles", "sources"),
     ("0", "Exit", "", None),
 ]