Open source · MIT license · Perpetually improving

Build persistent,
autonomous agents
that run forever.

CoreClaw is an open-core framework for long-running AI agents. Bring your own models, deploy anywhere, and extend with plugins.

Quickstart → View on GitHub

4.2k ★ GitHub npm install coreclaw Docker ready

quickstart.mjs

// coreclaw quickstart
// kiro mindset: never final, always improving
import { Agent } from "coreclaw"

const agent = new Agent({
  model: "glm-5.2",
  tools: ["web", "terminal", "file"],
  persistent: true,
})

await agent.schedule({
  every: "2h",
  task: "Scan competitor updates",
})

await agent.start() // runs until you stop it

Why CoreClaw

Built for developers who ship

Long-Running by Default

Agents persist across sessions with built-in cron scheduling. Deploy once, runs until told otherwise.

Model-Agnostic

Swap between any LLM provider — OpenCode Go, Anthropic, DeepSeek, Ollama, OpenAI. Route different tasks to different models.

Plugin System + MCP

Extend with plugins or any MCP-compatible server. Custom tools, custom data sources — no fork needed.

Get started in 2 minutes

Install. Configure. Deploy.

npm install -g coreclaw — one command install
coreclaw init — interactive setup (or headless config)
coreclaw start — your agent goes live, persistent across restarts

Read the docs

terminal

$ npm install -g coreclaw
$ coreclaw init

  ✔ API key configured
  ✔ Default model set
  ✔ Schedule loaded

$ coreclaw start

  ⚡ Agent is alive
  → Next task: 2m 15s
  → Status: monitoring

Works with

OpenCode Go Anthropic DeepSeek Ollama OpenAI MCP Docker

Start building with CoreClaw

Open source. Free to use. MIT license. Perpetually improving.

GitHub → npm install coreclaw