README.md

---
title: Moltbot Hybrid Engine
emoji: ⚖️
colorFrom: blue
colorTo: red
sdk: docker
pinned: false
app_port: 7860
---

# Moltbot Hybrid Engine
Safe AI agent for legal document processing - Dual LLM backend + file matching + **Clawdbot (OpenClaw)** in the Space.

**Version 7.0.0** - Last updated: 2026-02-14 — Clawdbot installed; gateway runs on port 18789, proxied at `/gateway`

## Required Space Secrets

Set these in Space Settings > Repository secrets:

| Secret | Purpose |
|--------|---------|
| `HF_TOKEN` | HuggingFace API token (get free at hf.co/settings/tokens) - enables HF Inference API for Qwen |
| `MOLTBOT_API_KEY` | API key for authenticating requests (set any strong value) |

## LLM Backends

1. **Ollama** (local in container) - runs qwen2.5:1.5b if binary installs correctly
2. **HF Inference API** (fallback) - uses Qwen/Qwen2.5-7B-Instruct hosted by HuggingFace (requires HF_TOKEN)

## Clawdbot (OpenClaw) in this Space

OpenClaw/Clawdbot is **installed and started** in this Space so you can use the autonomous agent with Qwen and Claude.

- **Control UI / WebChat:** After the Space is running, open: **`https://<your-space>.hf.space/gateway`**
- The gateway runs internally on port 18789; the FastAPI app proxies `/gateway` and `/gateway/*` to it so a single Space port (7860) serves both the API and Clawdbot.
- **To get autonomous behaviour** (Clawdbot working on court bundle changes with Qwen/Claude):
  1. Open the Control UI at `/gateway` and complete pairing/setup (model, API keys).
  2. Add a **skill** that triggers your pipeline (e.g. run the web editor flow, call this Engine, run the bundler). Skills live in `~/.openclaw/workspace/skills/`; you can add a custom skill that calls your court bundle project endpoints or scripts.
  3. Wire that skill to the same Qwen/Claude endpoints the editor uses (this Space’s `/api/generate` and your Claude API).