diff --git a/src/cad_documenter/cli.py b/src/cad_documenter/cli.py
index 3866de7..1261bbf 100644
--- a/src/cad_documenter/cli.py
+++ b/src/cad_documenter/cli.py
@@ -11,6 +11,7 @@ from rich.table import Table
 
 from .pipeline import DocumentationPipeline, PipelineProgress, PipelineStage, create_pipeline
 from .config import Config, load_config
+from .cli_project import project as project_commands
 
 console = Console()
 
@@ -380,5 +381,9 @@ def main():
     cli()
 
 
+# Register project subcommands
+cli.add_command(project_commands)
+
+
 if __name__ == "__main__":
     main()
diff --git a/src/cad_documenter/cli_project.py b/src/cad_documenter/cli_project.py
new file mode 100644
index 0000000..20122b5
--- /dev/null
+++ b/src/cad_documenter/cli_project.py
@@ -0,0 +1,276 @@
+"""CLI commands for project management."""
+
+import click
+from pathlib import Path
+from rich.console import Console
+from rich.table import Table
+from rich.panel import Panel
+from rich.progress import Progress, SpinnerColumn, TextColumn
+
+from .project import Project
+from .incremental import IncrementalProcessor, UnifiedDocumentGenerator
+from .config import load_config
+
+console = Console()
+
+
+@click.group()
+def project():
+    """Manage iterative documentation projects."""
+    pass
+
+
+@project.command()
+@click.argument("path", type=click.Path(path_type=Path))
+@click.option("--name", "-n", help="Project name (defaults to folder name)")
+@click.option("--description", "-d", default="", help="Project description")
+def init(path: Path, name: str | None, description: str):
+    """Create a new documentation project.
+    
+    Example:
+        cad-doc project init ./my-bracket-project
+    """
+    if name is None:
+        name = path.name
+    
+    try:
+        proj = Project.create(path, name, description)
+        console.print(f"[green]✓[/green] Created project: [cyan]{name}[/cyan]")
+        console.print(f"  Location: {path.absolute()}")
+        console.print()
+        console.print("Next steps:")
+        console.print(f"  1. Copy videos to [cyan]{path}/videos/[/cyan]")
+        console.print(f"  2. Run [cyan]cad-doc project add {path} <video>[/cyan]")
+        console.print(f"  3. Run [cyan]cad-doc project process {path}[/cyan]")
+        
+    except ValueError as e:
+        console.print(f"[red]Error:[/red] {e}")
+        raise SystemExit(1)
+
+
+@project.command()
+@click.argument("project_path", type=click.Path(exists=True, path_type=Path))
+@click.argument("video", type=click.Path(exists=True, path_type=Path))
+@click.option("--no-copy", is_flag=True, help="Don't copy video, just reference it")
+def add(project_path: Path, video: Path, no_copy: bool):
+    """Add a video to the project.
+    
+    Example:
+        cad-doc project add ./my-project video.mp4
+    """
+    try:
+        proj = Project.load(project_path)
+        entry = proj.add_video(video, copy=not no_copy)
+        
+        console.print(f"[green]✓[/green] Added: [cyan]{video.name}[/cyan]")
+        console.print(f"  Status: {entry.status}")
+        
+        pending = len(proj.get_pending_videos())
+        console.print(f"\nPending videos: {pending}")
+        console.print(f"Run [cyan]cad-doc project process {project_path}[/cyan] to process")
+        
+    except FileNotFoundError as e:
+        console.print(f"[red]Error:[/red] {e}")
+        raise SystemExit(1)
+
+
+@project.command()
+@click.argument("project_path", type=click.Path(exists=True, path_type=Path))
+@click.option("--all", "process_all", is_flag=True, help="Reprocess all videos, not just pending")
+def process(project_path: Path, process_all: bool):
+    """Process pending videos in the project.
+    
+    Example:
+        cad-doc project process ./my-project
+    """
+    try:
+        proj = Project.load(project_path)
+        config = load_config()
+        
+        pending = proj.get_pending_videos()
+        if not pending:
+            console.print("[yellow]No pending videos to process[/yellow]")
+            return
+        
+        console.print(f"Processing {len(pending)} video(s)...")
+        console.print()
+        
+        processor = IncrementalProcessor(proj, config)
+        
+        with Progress(
+            SpinnerColumn(),
+            TextColumn("[progress.description]{task.description}"),
+            console=console,
+        ) as progress:
+            
+            def on_progress(msg):
+                progress.update(task, description=msg)
+            
+            task = progress.add_task("Starting...", total=None)
+            results = processor.process_pending(progress_callback=on_progress)
+        
+        # Show results
+        console.print()
+        console.print(Panel.fit(
+            f"[bold green]Processing complete![/bold green]\n\n"
+            f"Videos processed: {results['processed']}\n"
+            f"New components: {len(results['new_components'])}\n"
+            f"Updated components: {len(results['updated_components'])}\n"
+            f"Changes detected: {len(results['changes_detected'])}",
+            title="Results"
+        ))
+        
+        if results['new_components']:
+            console.print("\n[bold]New components:[/bold]")
+            for name in results['new_components']:
+                console.print(f"  [green]+[/green] {name}")
+        
+        if results['updated_components']:
+            console.print("\n[bold]Updated components:[/bold]")
+            for name in results['updated_components']:
+                console.print(f"  [yellow]~[/yellow] {name}")
+        
+        if results['errors']:
+            console.print("\n[bold red]Errors:[/bold red]")
+            for err in results['errors']:
+                console.print(f"  [red]✗[/red] {err['video']}: {err['error']}")
+        
+        console.print(f"\nRun [cyan]cad-doc project generate {project_path}[/cyan] to create documentation")
+        
+    except FileNotFoundError as e:
+        console.print(f"[red]Error:[/red] {e}")
+        raise SystemExit(1)
+
+
+@project.command()
+@click.argument("project_path", type=click.Path(exists=True, path_type=Path))
+@click.option("--no-history", is_flag=True, help="Don't include change history")
+@click.option("--no-bom", is_flag=True, help="Don't include Bill of Materials")
+@click.option("--no-atomizer", is_flag=True, help="Don't include Atomizer hints")
+@click.option("--pdf", is_flag=True, help="Also generate PDF")
+def generate(project_path: Path, no_history: bool, no_bom: bool, no_atomizer: bool, pdf: bool):
+    """Generate unified documentation from all processed videos.
+    
+    Example:
+        cad-doc project generate ./my-project
+    """
+    try:
+        proj = Project.load(project_path)
+        
+        generator = UnifiedDocumentGenerator(proj)
+        
+        console.print("Generating unified documentation...")
+        
+        doc_path = generator.generate(
+            include_history=not no_history,
+            include_bom=not no_bom,
+            include_atomizer=not no_atomizer,
+        )
+        
+        console.print(f"[green]✓[/green] Documentation: [cyan]{doc_path}[/cyan]")
+        
+        if pdf:
+            console.print("[yellow]PDF generation not yet implemented for projects[/yellow]")
+        
+        # Show summary
+        status = proj.get_status()
+        console.print()
+        console.print(f"Components documented: {status['total_components']}")
+        console.print(f"From {status['processed']} videos ({status['total_duration']:.1f}s total)")
+        
+    except FileNotFoundError as e:
+        console.print(f"[red]Error:[/red] {e}")
+        raise SystemExit(1)
+
+
+@project.command()
+@click.argument("project_path", type=click.Path(exists=True, path_type=Path))
+def status(project_path: Path):
+    """Show project status.
+    
+    Example:
+        cad-doc project status ./my-project
+    """
+    try:
+        proj = Project.load(project_path)
+        status = proj.get_status()
+        
+        console.print(Panel.fit(
+            f"[bold]{status['name']}[/bold]",
+            subtitle=f"Last updated: {status['last_updated'][:16]}"
+        ))
+        
+        # Videos table
+        table = Table(title="Videos")
+        table.add_column("#", style="dim")
+        table.add_column("Filename")
+        table.add_column("Status")
+        table.add_column("Components")
+        
+        for i, video in enumerate(proj.manifest.videos, 1):
+            status_style = {
+                "pending": "yellow",
+                "processed": "green",
+                "error": "red",
+            }.get(video.status, "white")
+            
+            table.add_row(
+                str(i),
+                video.filename,
+                f"[{status_style}]{video.status}[/{status_style}]",
+                str(len(video.components_found)) if video.components_found else "-"
+            )
+        
+        console.print(table)
+        
+        # Summary
+        console.print()
+        console.print(f"Total videos: {status['total_videos']} ({status['pending']} pending)")
+        console.print(f"Components: {status['total_components']}")
+        console.print(f"Total duration: {status['total_duration']:.1f}s")
+        console.print(f"Total frames: {status['total_frames']}")
+        
+        if status['pending'] > 0:
+            console.print()
+            console.print(f"[yellow]Run [cyan]cad-doc project process {project_path}[/cyan] to process pending videos[/yellow]")
+        
+    except FileNotFoundError as e:
+        console.print(f"[red]Error:[/red] {e}")
+        raise SystemExit(1)
+
+
+@project.command("list-components")
+@click.argument("project_path", type=click.Path(exists=True, path_type=Path))
+def list_components(project_path: Path):
+    """List all components in the project.
+    
+    Example:
+        cad-doc project list-components ./my-project
+    """
+    try:
+        proj = Project.load(project_path)
+        components = proj.get_all_components()
+        
+        if not components:
+            console.print("[yellow]No components found. Process some videos first.[/yellow]")
+            return
+        
+        table = Table(title=f"Components ({len(components)})")
+        table.add_column("Name")
+        table.add_column("Material")
+        table.add_column("Function")
+        table.add_column("Changes")
+        
+        for comp in components:
+            table.add_row(
+                comp.name,
+                comp.material or "-",
+                comp.function[:40] + "..." if len(comp.function) > 40 else comp.function or "-",
+                str(len(comp.history)) if comp.history else "-"
+            )
+        
+        console.print(table)
+        
+    except FileNotFoundError as e:
+        console.print(f"[red]Error:[/red] {e}")
+        raise SystemExit(1)
diff --git a/src/cad_documenter/incremental.py b/src/cad_documenter/incremental.py
new file mode 100644
index 0000000..339757f
--- /dev/null
+++ b/src/cad_documenter/incremental.py
@@ -0,0 +1,414 @@
+"""Incremental processing for iterative documentation."""
+
+import re
+from pathlib import Path
+from datetime import datetime
+from dataclasses import dataclass
+
+from .project import Project, VideoEntry, ComponentKnowledge
+from .pipeline import DocumentationPipeline
+from .vision_analyzer import ComponentAnalysis, Component
+from .audio_analyzer import Transcript
+from .config import Config, load_config
+
+
+@dataclass
+class ChangeDetection:
+    """Detected change from transcript."""
+    change_type: str  # "supersedes", "adds", "updates", "removes"
+    component: str | None
+    field: str | None
+    old_value: str | None
+    new_value: str | None
+    quote: str
+    timestamp: float
+
+
+class IncrementalProcessor:
+    """Processes videos incrementally, building up knowledge."""
+    
+    # Patterns that indicate changes/updates
+    CHANGE_PATTERNS = [
+        # Superseding patterns
+        (r"(?:now|instead)\s+(?:we\s+)?(?:use|using)\s+(.+?)(?:\s+instead\s+of\s+(.+))?", "supersedes"),
+        (r"(?:i|we)\s+changed\s+(?:the\s+)?(.+?)\s+(?:from\s+(.+?)\s+)?to\s+(.+)", "supersedes"),
+        (r"(?:the\s+)?new\s+(.+?)\s+(?:is|replaces)", "supersedes"),
+        (r"replaced\s+(?:the\s+)?(.+?)\s+with\s+(.+)", "supersedes"),
+        
+        # Additive patterns
+        (r"(?:i|we)\s+added\s+(?:a\s+)?(.+)", "adds"),
+        (r"(?:in\s+addition|also|plus)\s+(?:we\s+have\s+)?(?:a\s+)?(.+)", "adds"),
+        (r"(?:there(?:'s| is)\s+)?(?:a\s+)?new\s+(.+)", "adds"),
+        
+        # Update patterns
+        (r"(?:i|we)\s+(?:updated|modified|adjusted|refined)\s+(?:the\s+)?(.+)", "updates"),
+        (r"(?:the\s+)?(.+?)\s+(?:is\s+)?now\s+(.+)", "updates"),
+        
+        # Removal patterns
+        (r"(?:i|we)\s+removed\s+(?:the\s+)?(.+)", "removes"),
+        (r"(?:no\s+longer|don't)\s+(?:have|need|use)\s+(?:the\s+)?(.+)", "removes"),
+    ]
+    
+    def __init__(self, project: Project, config: Config | None = None):
+        self.project = project
+        self.config = config or load_config()
+    
+    def process_pending(self, progress_callback=None) -> dict:
+        """Process all pending videos."""
+        pending = self.project.get_pending_videos()
+        
+        if not pending:
+            return {"processed": 0, "message": "No pending videos"}
+        
+        results = {
+            "processed": 0,
+            "errors": [],
+            "new_components": [],
+            "updated_components": [],
+            "changes_detected": [],
+        }
+        
+        for i, video_entry in enumerate(pending):
+            if progress_callback:
+                progress_callback(f"Processing {video_entry.filename} ({i+1}/{len(pending)})")
+            
+            try:
+                video_results = self.process_video(video_entry)
+                results["processed"] += 1
+                results["new_components"].extend(video_results.get("new_components", []))
+                results["updated_components"].extend(video_results.get("updated_components", []))
+                results["changes_detected"].extend(video_results.get("changes", []))
+                
+            except Exception as e:
+                video_entry.status = "error"
+                video_entry.error_message = str(e)
+                results["errors"].append({
+                    "video": video_entry.filename,
+                    "error": str(e)
+                })
+        
+        self.project.save()
+        return results
+    
+    def process_video(self, video_entry: VideoEntry) -> dict:
+        """Process a single video and merge into project knowledge."""
+        video_path = self.project.get_video_path(video_entry)
+        
+        # Create output dir for this video's frames
+        video_frames_dir = self.project.frames_dir / video_path.stem
+        video_frames_dir.mkdir(exist_ok=True)
+        
+        # Run pipeline
+        pipeline = DocumentationPipeline(
+            video_path=video_path,
+            output_dir=video_frames_dir,
+            config=self.config,
+        )
+        
+        # Extract and transcribe
+        frames = pipeline.extract_frames()
+        transcript = pipeline.transcribe_audio()
+        
+        # Save transcript
+        transcript_file = self.project.transcripts_dir / f"{video_path.stem}.json"
+        self._save_transcript(transcript, transcript_file)
+        
+        # Analyze components
+        analysis = pipeline.analyze_components(frames, transcript)
+        
+        # Detect explicit changes from transcript
+        changes = self.detect_changes(transcript)
+        
+        # Merge into project knowledge
+        merge_results = self.merge_analysis(
+            analysis, 
+            video_entry.filename,
+            changes
+        )
+        
+        # Update video entry
+        video_entry.status = "processed"
+        video_entry.processed_at = datetime.now().isoformat()
+        video_entry.duration = transcript.duration if hasattr(transcript, 'duration') else 0
+        video_entry.transcript_file = transcript_file.name
+        video_entry.frames_extracted = len(frames)
+        video_entry.components_found = [c.name for c in analysis.components]
+        
+        # Update project totals
+        self.project.manifest.total_frames += len(frames)
+        self.project.manifest.total_duration += video_entry.duration or 0
+        
+        return {
+            "frames": len(frames),
+            "transcript_duration": video_entry.duration,
+            "components_analyzed": len(analysis.components),
+            "changes": changes,
+            **merge_results
+        }
+    
+    def detect_changes(self, transcript: Transcript) -> list[ChangeDetection]:
+        """Detect explicit changes mentioned in transcript."""
+        changes = []
+        
+        for segment in transcript.segments:
+            text = segment.text.lower()
+            
+            for pattern, change_type in self.CHANGE_PATTERNS:
+                matches = re.finditer(pattern, text, re.IGNORECASE)
+                for match in matches:
+                    groups = match.groups()
+                    
+                    change = ChangeDetection(
+                        change_type=change_type,
+                        component=groups[0] if groups else None,
+                        field=None,
+                        old_value=groups[1] if len(groups) > 1 else None,
+                        new_value=groups[2] if len(groups) > 2 else groups[0],
+                        quote=segment.text,
+                        timestamp=segment.start,
+                    )
+                    changes.append(change)
+        
+        return changes
+    
+    def merge_analysis(
+        self, 
+        analysis: ComponentAnalysis,
+        source_video: str,
+        changes: list[ChangeDetection]
+    ) -> dict:
+        """Merge analysis results into project knowledge."""
+        new_components = []
+        updated_components = []
+        
+        # Process each detected component
+        for component in analysis.components:
+            existing = self.project.find_component(component.name)
+            
+            if existing:
+                # Update existing component
+                self.project.update_component(
+                    name=component.name,
+                    source_video=source_video,
+                    description=component.description,
+                    function=component.function,
+                    material=component.material,
+                    features=component.features,
+                    confidence=component.confidence,
+                )
+                updated_components.append(component.name)
+            else:
+                # New component
+                self.project.update_component(
+                    name=component.name,
+                    source_video=source_video,
+                    description=component.description,
+                    function=component.function,
+                    material=component.material,
+                    features=component.features,
+                    confidence=component.confidence,
+                )
+                new_components.append(component.name)
+        
+        # Apply explicit changes from transcript
+        for change in changes:
+            if change.change_type == "supersedes" and change.component:
+                comp = self.project.find_component(change.component)
+                if comp and change.new_value:
+                    # Try to determine what field changed
+                    if self._looks_like_material(change.new_value):
+                        self.project.update_component(
+                            name=comp.name,
+                            source_video=source_video,
+                            material=change.new_value,
+                        )
+                    else:
+                        # Add to description/notes
+                        pass
+        
+        # Merge assembly relationships
+        if analysis.assembly_notes:
+            self.project.manifest.assembly_relationships.append({
+                "source_video": source_video,
+                "notes": analysis.assembly_notes,
+            })
+        
+        # Merge Atomizer hints
+        if analysis.atomizer_hints:
+            hints_dict = {
+                "objectives": analysis.atomizer_hints.objectives,
+                "constraints": analysis.atomizer_hints.constraints,
+                "parameters": analysis.atomizer_hints.parameters,
+                "critical_regions": analysis.atomizer_hints.critical_regions,
+            }
+            self.project.merge_atomizer_hints(hints_dict, source_video)
+        
+        return {
+            "new_components": new_components,
+            "updated_components": updated_components,
+        }
+    
+    def _looks_like_material(self, text: str) -> bool:
+        """Check if text looks like a material name."""
+        materials = [
+            "aluminum", "aluminium", "steel", "stainless", "titanium",
+            "brass", "copper", "plastic", "nylon", "abs", "pla", "petg",
+            "carbon fiber", "composite", "wood", "rubber", "silicone",
+            "6061", "7075", "304", "316", "4140",
+        ]
+        text_lower = text.lower()
+        return any(mat in text_lower for mat in materials)
+    
+    def _save_transcript(self, transcript: Transcript, path: Path):
+        """Save transcript to JSON file."""
+        import json
+        
+        data = {
+            "full_text": transcript.full_text,
+            "duration": getattr(transcript, 'duration', 0),
+            "segments": [
+                {
+                    "start": seg.start,
+                    "end": seg.end,
+                    "text": seg.text,
+                }
+                for seg in transcript.segments
+            ]
+        }
+        
+        with open(path, "w") as f:
+            json.dump(data, f, indent=2)
+
+
+class UnifiedDocumentGenerator:
+    """Generates unified documentation from accumulated project knowledge."""
+    
+    def __init__(self, project: Project):
+        self.project = project
+    
+    def generate(
+        self,
+        include_history: bool = True,
+        include_atomizer: bool = True,
+        include_bom: bool = True,
+    ) -> Path:
+        """Generate unified documentation from all project knowledge."""
+        output_dir = self.project.output_dir
+        output_dir.mkdir(exist_ok=True)
+        
+        # Build document
+        doc_parts = []
+        
+        # Header
+        manifest = self.project.manifest
+        doc_parts.append(f"# {manifest.name}\n")
+        doc_parts.append(f"**Generated:** {datetime.now().strftime('%Y-%m-%d %H:%M')}\n")
+        doc_parts.append(f"**Videos processed:** {len([v for v in manifest.videos if v.status == 'processed'])}\n")
+        doc_parts.append(f"**Total duration:** {manifest.total_duration:.1f}s\n")
+        
+        if manifest.description:
+            doc_parts.append(f"\n{manifest.description}\n")
+        
+        doc_parts.append("\n---\n")
+        
+        # Components
+        doc_parts.append("\n## Components\n")
+        
+        components = self.project.get_all_components()
+        components.sort(key=lambda c: c.first_seen)
+        
+        for i, comp in enumerate(components, 1):
+            doc_parts.append(f"\n### {i}. {comp.name}\n")
+            
+            if comp.description:
+                doc_parts.append(f"\n{comp.description}\n")
+            
+            doc_parts.append("\n| Property | Value |")
+            doc_parts.append("\n|----------|-------|")
+            
+            if comp.function:
+                doc_parts.append(f"\n| **Function** | {comp.function} |")
+            if comp.material:
+                doc_parts.append(f"\n| **Material** | {comp.material} |")
+            if comp.part_number:
+                doc_parts.append(f"\n| **Part Number** | {comp.part_number} |")
+            
+            doc_parts.append("\n")
+            
+            if comp.features:
+                doc_parts.append("\n**Key Features:**\n")
+                for feat in comp.features:
+                    doc_parts.append(f"- {feat}\n")
+            
+            # History
+            if include_history and comp.history:
+                doc_parts.append("\n<details>\n<summary>Change History</summary>\n\n")
+                for change in comp.history:
+                    doc_parts.append(f"- **{change['timestamp'][:10]}**: {change['field']} changed from \"{change['old']}\" to \"{change['new']}\"\n")
+                doc_parts.append("\n</details>\n")
+            
+            doc_parts.append("\n---\n")
+        
+        # BOM
+        if include_bom and components:
+            doc_parts.append("\n## Bill of Materials\n")
+            doc_parts.append("\n| Item | Part Number | Name | Material | Function |")
+            doc_parts.append("\n|------|-------------|------|----------|----------|")
+            
+            for i, comp in enumerate(components, 1):
+                pn = comp.part_number or "TBD"
+                mat = comp.material or "TBD"
+                func = comp.function or "-"
+                doc_parts.append(f"\n| {i} | {pn} | {comp.name} | {mat} | {func} |")
+            
+            doc_parts.append("\n")
+        
+        # Atomizer hints
+        if include_atomizer and manifest.atomizer_hints:
+            doc_parts.append("\n## FEA / Atomizer Hints\n")
+            
+            hints = manifest.atomizer_hints
+            
+            if hints.get("objectives"):
+                doc_parts.append("\n### Objectives\n")
+                for obj in hints["objectives"]:
+                    doc_parts.append(f"- **{obj.get('direction', '').capitalize()} {obj.get('name', '')}**\n")
+            
+            if hints.get("constraints"):
+                doc_parts.append("\n### Constraints\n")
+                for const in hints["constraints"]:
+                    doc_parts.append(f"- {const.get('type', '')}: {const.get('value', '')}\n")
+            
+            if hints.get("parameters"):
+                doc_parts.append("\n### Parameters to Optimize\n")
+                for param in hints["parameters"]:
+                    doc_parts.append(f"- {param}\n")
+            
+            if hints.get("critical_regions"):
+                doc_parts.append("\n### Critical Regions\n")
+                for region in hints["critical_regions"]:
+                    doc_parts.append(f"- **{region.get('feature', '')}**: {region.get('concern', '')}\n")
+        
+        # Video sources
+        doc_parts.append("\n## Source Videos\n")
+        doc_parts.append("\n| # | Filename | Processed | Components Found |")
+        doc_parts.append("\n|---|----------|-----------|------------------|")
+        
+        for i, video in enumerate(manifest.videos, 1):
+            date = video.processed_at[:10] if video.processed_at else "pending"
+            comps = len(video.components_found)
+            doc_parts.append(f"\n| {i} | {video.filename} | {date} | {comps} |")
+        
+        doc_parts.append("\n")
+        
+        # Write document
+        doc_path = output_dir / "documentation.md"
+        doc_path.write_text("".join(doc_parts))
+        
+        # Also write atomizer hints JSON
+        if manifest.atomizer_hints:
+            import json
+            hints_path = output_dir / "atomizer_hints.json"
+            hints_path.write_text(json.dumps(manifest.atomizer_hints, indent=2))
+        
+        return doc_path
diff --git a/src/cad_documenter/project.py b/src/cad_documenter/project.py
new file mode 100644
index 0000000..d39082c
--- /dev/null
+++ b/src/cad_documenter/project.py
@@ -0,0 +1,391 @@
+"""Project management for iterative documentation."""
+
+import json
+from pathlib import Path
+from datetime import datetime
+from dataclasses import dataclass, field, asdict
+from typing import Literal
+import re
+
+
+@dataclass
+class VideoEntry:
+    """A video in the project."""
+    filename: str
+    added_at: str  # ISO timestamp
+    processed_at: str | None = None
+    duration: float | None = None
+    transcript_file: str | None = None
+    frames_extracted: int = 0
+    components_found: list[str] = field(default_factory=list)
+    status: Literal["pending", "processed", "error"] = "pending"
+    error_message: str | None = None
+
+
+@dataclass
+class ComponentKnowledge:
+    """Accumulated knowledge about a component."""
+    name: str
+    first_seen: str  # ISO timestamp
+    last_updated: str
+    description: str = ""
+    function: str = ""
+    material: str = ""
+    features: list[str] = field(default_factory=list)
+    part_number: str = ""
+    confidence: float = 0.0
+    
+    # History tracking
+    history: list[dict] = field(default_factory=list)
+    # [{"timestamp": "...", "field": "material", "old": "aluminum", "new": "steel", "source_video": "..."}]
+
+
+@dataclass
+class ProjectManifest:
+    """Project manifest tracking all videos and knowledge."""
+    name: str
+    created_at: str
+    updated_at: str
+    description: str = ""
+    
+    # Videos
+    videos: list[VideoEntry] = field(default_factory=list)
+    
+    # Accumulated knowledge
+    components: dict[str, ComponentKnowledge] = field(default_factory=dict)
+    assembly_relationships: list[dict] = field(default_factory=list)
+    atomizer_hints: dict = field(default_factory=dict)
+    
+    # Metadata
+    total_duration: float = 0.0
+    total_frames: int = 0
+    version: str = "1.0"
+
+
+class Project:
+    """Manages an iterative documentation project."""
+    
+    def __init__(self, project_dir: Path):
+        self.project_dir = Path(project_dir)
+        self.manifest_path = self.project_dir / "project.json"
+        self.videos_dir = self.project_dir / "videos"
+        self.knowledge_dir = self.project_dir / "knowledge"
+        self.frames_dir = self.project_dir / "frames"
+        self.output_dir = self.project_dir / "output"
+        self.transcripts_dir = self.knowledge_dir / "transcripts"
+        
+        self.manifest: ProjectManifest | None = None
+    
+    @classmethod
+    def create(cls, project_dir: Path, name: str, description: str = "") -> "Project":
+        """Create a new project."""
+        project_dir = Path(project_dir)
+        
+        if project_dir.exists() and any(project_dir.iterdir()):
+            raise ValueError(f"Directory {project_dir} is not empty")
+        
+        # Create directory structure
+        project_dir.mkdir(parents=True, exist_ok=True)
+        (project_dir / "videos").mkdir()
+        (project_dir / "knowledge").mkdir()
+        (project_dir / "knowledge" / "transcripts").mkdir()
+        (project_dir / "frames").mkdir()
+        (project_dir / "output").mkdir()
+        
+        # Create manifest
+        now = datetime.now().isoformat()
+        manifest = ProjectManifest(
+            name=name,
+            created_at=now,
+            updated_at=now,
+            description=description,
+        )
+        
+        project = cls(project_dir)
+        project.manifest = manifest
+        project.save()
+        
+        # Create README
+        readme = f"""# {name}
+
+{description}
+
+## Structure
+
+```
+{name}/
+├── project.json      # Project manifest
+├── videos/           # Add your walkthrough videos here
+├── knowledge/        # Accumulated knowledge base
+│   └── transcripts/  # Video transcripts
+├── frames/           # Extracted keyframes
+└── output/           # Generated documentation
+```
+
+## Usage
+
+```bash
+# Add a video
+cad-doc project add videos/my-video.mp4
+
+# Process all pending videos
+cad-doc project process
+
+# Generate documentation
+cad-doc project generate
+
+# Check status
+cad-doc project status
+```
+
+## Workflow
+
+1. Record a video explaining your CAD model
+2. Copy it to the `videos/` folder
+3. Run `cad-doc project add` and `cad-doc project process`
+4. Repeat as you make changes
+5. Run `cad-doc project generate` for unified documentation
+"""
+        (project_dir / "README.md").write_text(readme)
+        
+        return project
+    
+    @classmethod
+    def load(cls, project_dir: Path) -> "Project":
+        """Load an existing project."""
+        project = cls(project_dir)
+        
+        if not project.manifest_path.exists():
+            raise FileNotFoundError(f"No project found at {project_dir}")
+        
+        with open(project.manifest_path) as f:
+            data = json.load(f)
+        
+        # Reconstruct manifest
+        videos = [VideoEntry(**v) for v in data.get("videos", [])]
+        components = {
+            k: ComponentKnowledge(**v) 
+            for k, v in data.get("components", {}).items()
+        }
+        
+        project.manifest = ProjectManifest(
+            name=data["name"],
+            created_at=data["created_at"],
+            updated_at=data["updated_at"],
+            description=data.get("description", ""),
+            videos=videos,
+            components=components,
+            assembly_relationships=data.get("assembly_relationships", []),
+            atomizer_hints=data.get("atomizer_hints", {}),
+            total_duration=data.get("total_duration", 0.0),
+            total_frames=data.get("total_frames", 0),
+            version=data.get("version", "1.0"),
+        )
+        
+        return project
+    
+    def save(self):
+        """Save project manifest."""
+        if self.manifest is None:
+            raise ValueError("No manifest to save")
+        
+        self.manifest.updated_at = datetime.now().isoformat()
+        
+        # Convert to dict for JSON serialization
+        data = {
+            "name": self.manifest.name,
+            "created_at": self.manifest.created_at,
+            "updated_at": self.manifest.updated_at,
+            "description": self.manifest.description,
+            "videos": [asdict(v) for v in self.manifest.videos],
+            "components": {k: asdict(v) for k, v in self.manifest.components.items()},
+            "assembly_relationships": self.manifest.assembly_relationships,
+            "atomizer_hints": self.manifest.atomizer_hints,
+            "total_duration": self.manifest.total_duration,
+            "total_frames": self.manifest.total_frames,
+            "version": self.manifest.version,
+        }
+        
+        with open(self.manifest_path, "w") as f:
+            json.dump(data, f, indent=2)
+    
+    def add_video(self, video_path: Path, copy: bool = True) -> VideoEntry:
+        """Add a video to the project."""
+        video_path = Path(video_path)
+        
+        if not video_path.exists():
+            raise FileNotFoundError(f"Video not found: {video_path}")
+        
+        # Copy or link video to project
+        if copy:
+            dest = self.videos_dir / video_path.name
+            if not dest.exists():
+                import shutil
+                shutil.copy2(video_path, dest)
+            filename = video_path.name
+        else:
+            filename = str(video_path.absolute())
+        
+        # Check if already added
+        for v in self.manifest.videos:
+            if v.filename == filename:
+                return v  # Already exists
+        
+        # Create entry
+        entry = VideoEntry(
+            filename=filename,
+            added_at=datetime.now().isoformat(),
+        )
+        
+        self.manifest.videos.append(entry)
+        self.save()
+        
+        return entry
+    
+    def get_pending_videos(self) -> list[VideoEntry]:
+        """Get videos that haven't been processed yet."""
+        return [v for v in self.manifest.videos if v.status == "pending"]
+    
+    def get_video_path(self, entry: VideoEntry) -> Path:
+        """Get the full path to a video."""
+        if Path(entry.filename).is_absolute():
+            return Path(entry.filename)
+        return self.videos_dir / entry.filename
+    
+    def get_videos_chronological(self) -> list[VideoEntry]:
+        """Get all videos in chronological order."""
+        return sorted(self.manifest.videos, key=lambda v: v.added_at)
+    
+    def update_component(
+        self,
+        name: str,
+        source_video: str,
+        **updates
+    ) -> ComponentKnowledge:
+        """Update or create a component with change tracking."""
+        now = datetime.now().isoformat()
+        name_key = self._normalize_name(name)
+        
+        if name_key in self.manifest.components:
+            # Existing component - track changes
+            comp = self.manifest.components[name_key]
+            
+            for field_name, new_value in updates.items():
+                if not new_value:
+                    continue
+                    
+                old_value = getattr(comp, field_name, None)
+                
+                # Only track if actually changed
+                if old_value and old_value != new_value:
+                    comp.history.append({
+                        "timestamp": now,
+                        "field": field_name,
+                        "old": old_value,
+                        "new": new_value,
+                        "source_video": source_video,
+                    })
+                
+                setattr(comp, field_name, new_value)
+            
+            comp.last_updated = now
+            
+        else:
+            # New component
+            comp = ComponentKnowledge(
+                name=name,
+                first_seen=now,
+                last_updated=now,
+                **{k: v for k, v in updates.items() if v}
+            )
+            self.manifest.components[name_key] = comp
+        
+        return comp
+    
+    def _normalize_name(self, name: str) -> str:
+        """Normalize component name for matching."""
+        # Lowercase, remove extra spaces, standardize
+        name = name.lower().strip()
+        name = re.sub(r'\s+', ' ', name)
+        return name
+    
+    def find_component(self, name: str) -> ComponentKnowledge | None:
+        """Find a component by name (fuzzy matching)."""
+        name_key = self._normalize_name(name)
+        
+        # Exact match
+        if name_key in self.manifest.components:
+            return self.manifest.components[name_key]
+        
+        # Fuzzy match - check if name is contained
+        for key, comp in self.manifest.components.items():
+            if name_key in key or key in name_key:
+                return comp
+        
+        return None
+    
+    def get_all_components(self) -> list[ComponentKnowledge]:
+        """Get all components."""
+        return list(self.manifest.components.values())
+    
+    def merge_atomizer_hints(self, new_hints: dict, source_video: str):
+        """Merge new Atomizer hints with existing."""
+        existing = self.manifest.atomizer_hints
+        
+        # Merge objectives (deduplicate)
+        existing_objectives = existing.get("objectives", [])
+        new_objectives = new_hints.get("objectives", [])
+        
+        for obj in new_objectives:
+            # Check if similar objective exists
+            exists = any(
+                o.get("name") == obj.get("name") and o.get("direction") == obj.get("direction")
+                for o in existing_objectives
+            )
+            if not exists:
+                obj["source_video"] = source_video
+                existing_objectives.append(obj)
+        
+        existing["objectives"] = existing_objectives
+        
+        # Merge constraints
+        existing_constraints = existing.get("constraints", [])
+        new_constraints = new_hints.get("constraints", [])
+        
+        for const in new_constraints:
+            const["source_video"] = source_video
+            existing_constraints.append(const)
+        
+        existing["constraints"] = existing_constraints
+        
+        # Merge parameters (deduplicate)
+        existing_params = set(existing.get("parameters", []))
+        new_params = set(new_hints.get("parameters", []))
+        existing["parameters"] = list(existing_params | new_params)
+        
+        # Merge critical regions
+        existing_regions = existing.get("critical_regions", [])
+        new_regions = new_hints.get("critical_regions", [])
+        
+        for region in new_regions:
+            region["source_video"] = source_video
+            existing_regions.append(region)
+        
+        existing["critical_regions"] = existing_regions
+        
+        self.manifest.atomizer_hints = existing
+    
+    def get_status(self) -> dict:
+        """Get project status summary."""
+        videos = self.manifest.videos
+        return {
+            "name": self.manifest.name,
+            "total_videos": len(videos),
+            "pending": len([v for v in videos if v.status == "pending"]),
+            "processed": len([v for v in videos if v.status == "processed"]),
+            "errors": len([v for v in videos if v.status == "error"]),
+            "total_components": len(self.manifest.components),
+            "total_duration": self.manifest.total_duration,
+            "total_frames": self.manifest.total_frames,
+            "last_updated": self.manifest.updated_at,
+        }