ArcaneBridge is a local AI-automation bridge designed to safely connect Large Language Models with real development projects on a user's machine. It provides a secure, human-in-the-loop workflow for file generation, patching, scanning, and structured project operations.
ArcaneBridge does not expose any local filesystem directly to the internet.
All access is routed through a controlled API, authenticated with a local token and gated through a web UI.
Core Features
- Project Registry Each project on the machine can be registered with:
- A project ID
- A root directory (relative to global workspace)
- Status and metadata
This ensures LLM-assisted workflows stay inside sandboxed project roots.
- Structured Command Envelopes The API accepts strict JSON envelopes defining:
- The project to act on
- The action (
write_files,read_files,scan_tree, etc.) - Payload describing the operation
- Session metadata
This prevents arbitrary or unsafe execution.
- Session Manifests Every operation generates a manifest:
- Before/after file hashes
- Reason codes
- Timestamps
- Files touched
- Success / failure info
Perfect for audits, debugging, and rollback planning.
- Local-Only Daemon ArcaneBridge runs locally using:
- Python 3.11+
- FastAPI
- Uvicorn
- venv isolation
It is designed to be optionally exposed through a secure tunnel (e.g., Cloudflare Access) for remote control.
- Simple Web UI The built-in UI provides:
- Project management
- A JSON console for sending commands
- Session history viewer
- Config summary
This keeps the user in full control of what the AI can do.
Architecture Overview
ArcaneBridge uses:
- FastAPI for routing
- Pydantic for schema validation
- Uvicorn for serving
- Python pathlib and OS modules for filesystem work
- A simple front-end (HTML/CSS/JS) in
ui/
The daemon runs at: http://127.0.0.1:8765
(Or behind a secured HTTPS tunnel if configured.)
Security Notes
- API tokens and configs live outside the repository under:
~/.arcane_bridge/config.yaml
- This repository intentionally excludes any configuration files.
- Never expose the daemon publicly without an access control layer.
Future Milestones
- Snapshot / rollback engine
- Diff-based patching
- Multi-project orchestration
- Enhanced UI file browser
- Remote HTTPS workstation mode
- Plugin/action extensions
License MIT License
ArcaneBridge is an ongoing project focused on building a safe, transparent interface between local development and modern LLM tools. Community contributions, feedback, and ideas are welcome.