rcarmo/piclaw

pi coding agent in a technicolor web trenchcoat

What it solves

PiClaw provides a comprehensive, self-hosted workspace for AI coding agents. It solves the problem of having to jump between different tools (chat, code editors, terminals, and file viewers) when working with an AI agent by integrating them into a single, stateful web interface.

How it works

It packages the Pi Coding Agent into a containerized environment with a streaming web UI. The system uses SQLite to maintain persistent state for chat history, tasks, and token usage. It supports multiple LLM providers (including OpenAI-compatible endpoints and local llama.cpp presets) and features a modular architecture where capabilities can be extended via optional add-ons for additional tools, skills, and UI panes.

Who it’s for

Developers and power users who want a private, self-hosted AI agent workspace that can interact with their local files, run terminal commands, and manage complex coding projects.

Highlights

  • Integrated Workspace: Combines chat, a CodeMirror 6 editor, an xterm.js terminal, and various file viewers (PDF, CSV, video) in one UI.
  • Persistent State: Maintains session history, encrypted keychains, and SSH profiles via a local database.
  • Built-in Automation: Includes tools for browser automation (CDP), image processing, and Model Context Protocol (MCP) support.
  • Extensible Design: Supports optional add-ons for advanced features like Windows desktop automation, Proxmox integration, and Office document rendering.
  • Multi-Provider Support: Compatible with various LLM backends, including local models via llama.cpp.

Related

  • Project
  • Project
  • Project
  • Project
  • Project