AI Nervous System - Document Intelligence

Document Intelligence Skills

These skills enable OpenClaw (Moltbot) to interact with the AI Nervous System's document processing pipeline via Telegram, Discord, or other interfaces.

Skills

document_upload

Purpose: Upload a document for AI processing (summarization + vector embedding)

Trigger Phrases:

• "upload document"
• "process file"
• "index document"
• "add to knowledge base"

Parameters:

API Endpoint:

POST http://localhost:8000/upload
Content-Type: multipart/form-data

file: <binary file data>

Response Schema:

{
  "message": "Document uploaded successfully",
  "document": {
    "id": 1,
    "filename": "example.md",
    "status": "processing",
    "summary": null,
    "created_at": "2024-01-01T00:00:00Z"
  }
}

Instructions:

1. Accept file from user (attachment or path)
2. POST to /upload endpoint with multipart form data
3. Return document ID and initial status
4. Optionally poll /status/{doc_id} for completion
5. Return AI summary when processing completes

Example Interaction:

User: Upload this research paper
Bot: Uploading document... Created document #42
Bot: Processing with Ollama (llama3.2:3b)...
Bot: Complete! Summary:
     • Key finding 1
     • Key finding 2
     • Key finding 3

document_search

Purpose: Semantic vector search across indexed documents

Trigger Phrases:

• "search documents for"
• "find files about"
• "what documents mention"
• "search knowledge base"

Parameters:

API Endpoint:

GET http://localhost:8000/search?q={query}&limit={limit}

Response Schema:

{
  "query": "machine learning optimization",
  "results": [
    {
      "id": 1,
      "filename": "ml_paper.md",
      "summary": "This document covers...",
      "similarity": 0.89
    }
  ],
  "count": 1
}

Instructions:

1. Parse user's search intent into query string
2. GET /search with URL-encoded query
3. Format results with similarity scores
4. Highlight high-relevance (>0.7) matches
5. Include document summaries in response

Example Interaction:

User: Search for documents about Python async patterns
Bot: Found 3 relevant documents:

📄 CLAUDE.md (89% match)
   Summary: Project uses async/await patterns with FastAPI...

📄 api_design.txt (72% match)
   Summary: Guidelines for async endpoint design...

📄 notes.md (58% match)
   Summary: Meeting notes discussing concurrency...

document_status

Purpose: Check processing status of a specific document

Trigger Phrases:

• "status of document"
• "is document ready"
• "check processing"

Parameters:

API Endpoint:

GET http://localhost:8000/status/{doc_id}

Response Schema:

{
  "id": 1,
  "filename": "example.md",
  "status": "completed",
  "summary": "AI-generated summary...",
  "created_at": "2024-01-01T00:00:00Z",
  "updated_at": "2024-01-01T00:01:00Z"
}

Status Values:

• pending - Queued for processing
• processing - Ollama is analyzing
• completed - Ready with summary and embedding
• error - Processing failed

Autonomous Behaviors

Proactive File Watcher

The Document Intelligence module includes an autonomous file watcher that monitors:

• ~/Downloads
• ~/Documents/ToProcess

When new .txt, .md, or .pdf files appear, they are automatically:

1. Uploaded to the FastAPI backend
2. Processed with Ollama for summarization
3. Embedded with nomic-embed-text for vector search
4. Added to the knowledge base

Notification Pattern:

Bot: 🔔 New document detected: research_paper.pdf
Bot: Processing...
Bot: ✅ Indexed! Summary: [3 bullet points]

Heartbeat Monitor

Every 30 minutes, the system audits the document pipeline:

• Checks for stuck documents (processing > 10 min)
• Retries failed Ollama tasks
• Prepares daily intelligence briefing

Integration Notes

For Telegram Bot:

@bot.message_handler(content_types=['document'])
async def handle_document(message):
    file_info = await bot.get_file(message.document.file_id)
    # Download and POST to /upload

For Discord Bot:

@bot.event
async def on_message(message):
    if message.attachments:
        for attachment in message.attachments:
            # Download and POST to /upload

Cold War Jazz Aesthetic

All responses should maintain the tactical intelligence aesthetic:

• Use [INTEL], [CLASSIFIED], [BRIEFING] prefixes
• Monospace formatting for data
• Teal (#4a9c94) for success, Amber (#d4a56a) for processing