profClaw

Self-hosted AI agent engine for developers.
35 providers · 72 tools · 22 channels · runs on anything.

Docs · Report Bug · Discord

Table of Contents

• Why profClaw
• Features
• Quick Start
• Interactive TUI
• Slash Commands
• Headless Mode
• Deployment Modes
• Where It Runs
• Configuration
• AI Providers
• Chat Channels
• Integrations
• Architecture
• Development
• License

Why profClaw

Most AI coding tools are either cloud-only (Cursor, Devin) or single-purpose CLIs (Aider, SWE-Agent). profClaw is different:

• Local-first — runs on your machine, your data stays local
• Multi-provider — 35 AI providers including Ollama for fully offline usage
• Interactive TUI — streaming markdown, syntax highlighting, slash commands, and model selector in the terminal
• Multi-agent orchestration — routes tasks to the right agent (Claude, GPT, Ollama) based on capability scoring
• Real task queue — BullMQ with dead letter queue, retry with backoff, priority scheduling
• 22 chat channels — talk to your AI through Slack, Discord, Telegram, WhatsApp, Teams, or 17 others
• Cost tracking — per-token budget management with alerts at 50/80/100%
• 72 built-in tools — file ops, git, browser automation, cron, web search, canvas, voice
• Scales to edge — pico mode runs on a Raspberry Pi Zero 2W in ~140MB RAM

No other open-source tool combines task orchestration, project management, cost tracking, and a first-class TUI in one self-hosted package.

Features

Quick Start

npm install -g profclaw
profclaw init
profclaw chat --tui

profclaw init scans your project, detects your stack, and writes a PROFCLAW.md context file. It also auto-detects any AI provider keys in your environment.

profclaw chat --tui opens the interactive TUI with streaming responses, syntax-highlighted code, slash commands, and a live model selector.

Full setup wizard

npx profclaw onboard

Picks your AI provider, sets up config, and starts the server — zero to running in under 5 minutes.

Docker

docker run -d \
  -p 3000:3000 \
  -e ANTHROPIC_API_KEY=sk-ant-xxx \
  -e PROFCLAW_MODE=mini \
  ghcr.io/profclaw/profclaw:latest

Docker Compose

git clone https://github.com/profclaw/profclaw.git
cd profclaw
cp .env.example .env
docker compose up -d

Want free local AI? Add Ollama:

docker compose --profile ai up -d

One-liner (macOS/Linux)

curl -fsSL https://raw.githubusercontent.com/profclaw/profclaw/main/install.sh | bash

Interactive TUI

profClaw ships with a first-class terminal UI — no browser required.

profclaw chat --tui
# or
profclaw tui

The TUI provides:

• Streaming markdown rendered live as the model responds
• Syntax-highlighted code blocks for 100+ languages
• Slash command picker — type / to browse all 50 skills with descriptions
• Live model selector — switch providers mid-conversation with Ctrl+M
• Tool execution panel — see every tool call, argument, and result in real time
• Session history — scroll back through previous conversations
• Keyboard-driven — full navigation without a mouse

┌─ profClaw v2.x.x ─────────────────────────────────────────────┐
│  [Chat]                               [Tools]                  │
│                                                                 │
│  You: Review src/auth.ts for security issues                    │
│                                                                 │
│  Agent: I'll analyze auth.ts for security issues.     read_file │
│                                                       ───────── │
│  ## Security Analysis                                 auth.ts   │
│                                                                 │
│  **Critical**: JWT secret read from env without      analyze   │
│  fallback guard on line 42. If `JWT_SECRET` is       ──────── │
│  undefined the app will sign tokens with `undefined`. auth.ts  │
│  ...                                                            │
└─────────────────────────────── [claude-sonnet-4-6] [pro] ──────┘

Slash Commands

Skills are pre-built expertise modules invoked with / in any chat interface or the TUI:

See Skills docs for all 50 built-in skills and how to create your own with a plain Markdown file.

Headless Mode

Run agents from scripts, CI pipelines, or other services:

# One-shot message
profclaw chat "Summarize the last 10 git commits"

# JSON output for scripting
profclaw chat "List open TODOs in src/" --output json

# Run a skill headlessly
profclaw skill run commit --message "feat: add auth"

# Agent task with file context
profclaw agent run "Review src/ for security issues" --files src/

# Verify your setup
profclaw doctor

REST API:

curl -X POST http://localhost:3000/api/chat/message \
  -H "Content-Type: application/json" \
  -d '{"message": "List open GitHub issues", "provider": "anthropic"}'

Deployment Modes

profClaw scales from a Raspberry Pi to a full production server:

Set via PROFCLAW_MODE=pico|mini|pro environment variable.

Where It Runs

profClaw requires Node.js 22+. For bare-metal embedded devices (ESP32, Arduino), see MimiClaw (C) or PicoClaw (Go).

Configuration

The setup wizard (profclaw onboard) handles everything interactively. Or set environment variables:

# AI Provider (pick one)
ANTHROPIC_API_KEY=sk-ant-xxx
OPENAI_API_KEY=sk-xxx
OLLAMA_BASE_URL=http://localhost:11434

# Deployment
PROFCLAW_MODE=mini
PORT=3000

# Optional
REDIS_URL=redis://localhost:6379   # Required for pro mode

Run profclaw doctor to verify your configuration at any time.

See .env.example for all 130+ options.

AI Providers

35 providers with 90+ model aliases:

Chat Channels

Integrations

Architecture

src/
  adapters/       AI agent adapters (tool calling, streaming)
  chat/           Chat engine + execution pipeline
    providers/    Slack, Discord, Telegram, WhatsApp, WebChat, ...
    execution/    Tool executor, sandbox, agentic loop
  core/           Deployment modes, feature flags
  integrations/   GitHub, Jira, Linear webhooks
  queue/          BullMQ (pro) + in-memory (pico/mini) task queue
  providers/      35 AI SDK providers
  skills/         Skill loader and registry
  mcp/            MCP server (stdio + SSE)
  tui/            Interactive terminal UI (ink + blessed)
  types/          Shared TypeScript types

ui/src/           React 19 + Vite dashboard (mini/pro only)
skills/           Built-in skill definitions (Markdown)

Key design decisions:

• Mode-aware feature flags — the same binary scales from pico to pro; features are gated at runtime
• Adapter pattern for providers — adding a new AI provider is a single file implementing the ModelAdapter interface
• Skill-as-Markdown — skills are plain .md files with a structured header; no code required to create new ones
• BullMQ for production queues — optional Redis dependency; falls back to in-memory queue in pico/mini mode

See Architecture docs for a deep dive.

Development

git clone https://github.com/profclaw/profclaw.git
cd profclaw
pnpm install
cp .env.example .env
pnpm dev

See CONTRIBUTING.md for full guidelines.

Security

See SECURITY.md for our security policy and how to report vulnerabilities.

License

AGPL-3.0