CoderClaw Overview

此内容尚不支持你的语言。

CoderClaw Overview

CoderClaw is a self-hosted, open-source AI coding assistant that connects your favorite messaging apps to powerful AI agents. Run it on your own infrastructure — no cloud lock-in, no data leaving your control.

It is also a self-healing AI engineering agent and orchestration platform that manages tasks, workflows, and collaboration across all AI agents. It provides persistent memory, context-aware reasoning, and self-repair — allowing AI systems to detect failures, fix themselves, and adapt over time — while keeping humans in the loop for governance and approval.

TL;DR

CoderClaw connects messaging apps (WhatsApp, Telegram, Discord, Slack, and more) to AI coding agents with a secure, extensible gateway.

Personal use → spin up with a single command, message your AI from anywhere
Team use → add RBAC, distributed execution, and audit logs as you grow
Self-healing → agents detect failures, fix themselves, and adapt over time
Persistent memory → context and knowledge survive restarts and session changes
Human-in-the-loop → governance, approval flows, and audit trails built in
Any scale → startups (5–50 devs) through enterprise (100–1,000+ devs)

What is CoderClaw?

CoderClaw is a self-hosted multi-channel AI gateway that lets you interact with AI coding agents through the chat apps you already use. It is also a self-hosted multi-agent AI orchestration platform that coordinates specialized agents across the full development lifecycle — planning, coding, reviewing, testing, debugging, and refactoring — while maintaining persistent context and enabling self-repair.

Core Features:

Multi-channel messaging (WhatsApp, Telegram, Discord, Slack, Signal, iMessage)
WebSocket control plane for clients, nodes, and a web control UI
Plugin system for extensibility
Mobile nodes (iOS/Android)
Canvas and voice features
Tool system with skills support
Self-healing agent runtime (failure detection, auto-repair, adaptive execution)
Persistent memory and context-aware reasoning across sessions
Multi-agent orchestration with 7 built-in roles and custom agent support
Human-in-the-loop governance with approval flows and audit trails
CI/CD integration and private/self-hosted deployment support

coderClawLLM — a pay-per-use API layer for AI agent compute — is included for teams who need managed model access without managing their own provider keys.

Best For:

Personal AI assistant accessible from any device
Small teams in a trusted environment
Local or distributed execution
Projects requiring full data sovereignty
Startups (5–50 developers) using AI as a virtual workforce
Enterprises (100–1,000+ developers) running complex multi-agent pipelines
Organizations that need compliance, audit trails, and human approval workflows

Enterprise Features

CoderClaw includes enterprise-grade capabilities for teams that need more:

Additional Capabilities:

🔄 Transport Abstraction Layer - Execute tasks locally or remotely
📊 Distributed Task Lifecycle - Formal state machine with persistence
🔐 Enhanced Security - RBAC, device trust, comprehensive audit logs
🎯 Team Collaboration - Multi-session isolation, shared registries
🏢 Enterprise Ready - CI/CD integration, deterministic execution
🧠 Self-Healing Runtime - Agents detect failures and repair automatically
💰 coderClawLLM - Pay-per-use API layer for AI agent compute

Best For:

Development teams (5+ people)
Remote or distributed execution
Advanced security requirements (RBAC, granular permissions)
CI/CD automation
Enterprise compliance and audit trails

Feature Summary

Feature	Personal	Enterprise
Core Gateway
Multi-channel messaging	✅	✅
WebSocket control plane	✅	✅
Plugin system	✅	✅
Mobile nodes	✅	✅
Canvas & voice	✅	✅
Self-Healing & Memory
Persistent memory across sessions	✅	✅
Context-aware reasoning	✅	✅
Self-repair on failure	✅	✅
Human-in-the-loop approval flows	❌	✅
Execution
Local task execution	✅	✅
Remote task execution	❌	✅
Transport abstraction	❌	✅
Distributed runtime	❌	✅
Task Management
Basic task execution	✅	✅
Task lifecycle management	❌	✅
Task persistence	❌	✅
Task resumability	❌	✅
Audit trail	❌	✅
Security
Allowlists	✅	✅
Device pairing	✅	✅
Token authentication	✅	✅
RBAC	❌	✅
Device trust levels	❌	✅
Comprehensive audit logs	❌	✅
Multi-provider auth (OIDC, GitHub, Google)	❌	✅
Granular permissions	❌	✅
Collaboration
Single-user sessions	✅	✅
Multi-session isolation	❌	✅
Shared agent registries	❌	✅
Team policy enforcement	❌	✅
CI/CD integration	Basic	✅ Advanced
Developer Experience
CLI tools	✅	✅
Web Control UI	✅	✅
macOS/iOS/Android apps	✅	✅
Project knowledge engine	❌	✅
Multi-agent workflows	Basic	✅ Advanced
coderClawLLM compute API	❌	✅

Architecture

CoderClaw Gateway

┌─────────────────────────────────────┐
│      Multi-Channel Gateway          │
│  (WhatsApp, Telegram, Discord...)   │
└─────────────┬───────────────────────┘
              │
              ▼
┌─────────────────────────────────────┐
│       WebSocket Control Plane       │
│     (Clients, Nodes, Control UI)    │
└─────────────┬───────────────────────┘
              │
              ▼
┌─────────────────────────────────────┐
│         Pi Agent Runtime            │
│       (Local execution only)        │
└─────────────────────────────────────┘

CoderClaw with Enterprise Runtime

┌─────────────────────────────────────┐
│      Multi-Channel Gateway          │
│  (WhatsApp, Telegram, Discord...)   │
└─────────────┬───────────────────────┘
              │
              ▼
┌─────────────────────────────────────┐
│       WebSocket Control Plane       │
│     (Clients, Nodes, Control UI)    │
└─────────────┬───────────────────────┘
              │
              ▼
┌─────────────────────────────────────┐
│     CoderClaw Runtime Layer         │
│  ┌───────────────────────────────┐  │
│  │  Transport Abstraction Layer  │  │
│  └───────────────────────────────┘  │
│  ┌───────────────────────────────┐  │
│  │ Distributed Task Engine       │  │
│  └───────────────────────────────┘  │
│  ┌───────────────────────────────┐  │
│  │  Security Service (RBAC)      │  │
│  └───────────────────────────────┘  │
└─────────────┬───────────────────────┘
              │
              ▼
┌─────────────────────────────────────┐
│   Pi Agent Runtime + Task Executor  │
│   (Local or Remote execution)       │
└─────────────────────────────────────┘

Use Cases

Personal and Small Team Use

✅ Personal AI assistant for messaging apps ✅ Small trusted team (2-5 people) ✅ Local execution is sufficient ✅ Basic allowlist security is enough ✅ No need for audit trails ✅ Simple deployment model

Example: “I want an AI assistant I can message on WhatsApp from my phone that runs on my Mac at home.”

Startup Use (5–50 developers)

✅ Use AI as a virtual workforce to ship faster with a small team ✅ Self-healing agents handle failures automatically, reducing on-call burden ✅ Persistent memory means agents understand your codebase without re-explaining ✅ CoderClawLink gives a Jira-like view of what every agent is doing ✅ coderClawLLM provides pay-per-use compute without managing provider keys

Example: “We’re a 10-person startup. Our AI agents write code, review PRs, and fix failing tests automatically — all visible in CoderClawLink. We don’t need a dedicated DevOps engineer to babysit CI.”

Team and Enterprise Use

✅ Development team (5+ people) ✅ Need remote or distributed execution ✅ Require RBAC and granular permissions ✅ Need comprehensive audit logs ✅ CI/CD automation requirements ✅ Enterprise compliance needs ✅ Multi-tenant deployments ✅ Complex multi-agent pipelines with deterministic execution and full audit trails ✅ Human-in-the-loop approval gates for sensitive changes ✅ Private/self-hosted deployments for air-gapped or regulated environments

Example: “Our team needs an AI assistant that runs on a shared server, with different permission levels for developers, reviewers, and CI pipelines, plus full audit logs for compliance.”

Extended example: “Our team needs AI agents that orchestrate across 50 microservice repos, with different permission levels for developers, reviewers, and CI pipelines, plus full audit logs for compliance — all governed by CoderClawLink.”

Pricing and Licensing

CoderClaw is MIT licensed and free to use.

Source: github.com/SeanHogg/coderClaw

Getting Started

npm install -g coderclaw@latest
coderclaw onboard --install-daemon
coderclaw gateway --port 18789

Documentation: docs.coderclaw.ai

Enabling Enterprise Features

Enterprise features are backward compatible — your existing setup continues to work. Enable them as needed:

# Your existing setup works unchanged
coderclaw gateway --port 18789

# Opt-in to extended features
mkdir -p ~/.coderclaw/.coderClaw
# Add runtime.yaml and security.yaml as needed

Contributing

CoderClaw welcomes contributions:

Core gateway features, channels, tools
Enterprise runtime, security, and distributed features

Community

Join Discord

FAQ

Q: Can I run CoderClaw on the same machine as other services?

A: Yes, just ensure the gateway port (18789) is not in use by another process.

Q: Is CoderClaw expensive to run?

A: No. Enterprise features add minimal overhead. API costs depend on which AI models you configure.

Q: Can I switch configurations easily?

A: Yes. Backup your ~/.coderclaw directory to preserve your configuration.

Q: Which setup should I start with?

A: If you’re just exploring — start simple. If you know you need team features — enable enterprise capabilities from the start.

Summary

CoderClaw = Self-hosted, open-source AI coding assistant + multi-channel messaging gateway + self-healing multi-agent orchestration platform.

Connects messaging apps (WhatsApp, Telegram, Discord, Slack, and more) to powerful AI coding agents
Manages tasks, workflows, and collaboration across all AI agents
Persistent memory and context-aware reasoning across sessions
Self-repair: agents detect failures, fix themselves, and adapt over time
Human-in-the-loop governance with approval flows and audit trails
CoderClawLink replaces Jira with a centralized orchestration portal
coderClawLLM — pay-per-use API layer for AI agent compute

Choose your setup based on requirements:

Solo or small team + local only — start with the default configuration
Startup (5–50 devs) — use AI as a virtual workforce with full visibility in CoderClawLink
Team or enterprise + distributed + security — enable enterprise runtime features
Enterprise (100–1,000+ devs) — multi-agent pipelines with RBAC, audit trails, and human approval gates

This guide is maintained by the CoderClaw project. For questions, see docs.coderclaw.ai.