import { CodeAgent } from '@devclaw/core'; const system = new CodeAgent({ mode: "autonomous" }); // analyzing workspace architecture...
import { CodeAgent } from '@devclaw/core'; const system = new CodeAgent({ mode: "autonomous" }); // analyzing workspace architecture...
## DevClaw Terminal Instance ## DevClaw v2026.4.1 initialized. git add . && git commit -m "AI applied architecture refactor"
## DevClaw Terminal Instance ## DevClaw v2026.4.1 initialized. git add . && git commit -m "AI applied architecture refactor"
function blockCopilot() { return true; } // Loading local Generative Model... try { await executeWorkflow("T3 Stack Migration"); } catch(e) { autoFix(e); }
function blockCopilot() { return true; } // Loading local Generative Model... try { await executeWorkflow("T3 Stack Migration"); } catch(e) { autoFix(e); }
{ "quality": "stable", "nameShort": "DevClaw", "dataFolder": ".devclaw" } // Your data, your control.
{ "quality": "stable", "nameShort": "DevClaw", "dataFolder": ".devclaw" } // Your data, your control.
DevClawDevClaw
Back to Docs

Getting Started

DevClaw is the AI that actually does things — a VSCode-based IDE with autonomous agents that code, test, and deploy independently.

What you get

  • A native VSCode fork with first-class agent support
  • Bring-your-own-model (OpenAI, Anthropic, local LLMs, custom endpoints)
  • Open source — read it, fork it, ship it
  • macOS download today, Windows + Linux on the roadmap

Quick start

  1. Download DevClaw for macOS from the download page.
  2. Install by dragging the app to /Applications.
  3. Launch DevClaw — it'll create a .devclaw/ config directory in your home folder on first run.
  4. Configure your AI provider in Settings (bring your own API key for OpenAI, Anthropic, or any OpenAI-compatible endpoint).

What's next

  • Installation details — system requirements, manual install, troubleshooting.
  • More guides coming soon (agent setup, custom model endpoints, contribution guide).

Philosophy

DevClaw is built on three ideas:

  1. AI should ship code, not just suggest it. Agents write, test, and commit. You review.
  2. Your data is yours. Run local models. Use your own keys. No telemetry you didn't opt into.
  3. Open beats closed. Source on GitHub, issues public, roadmap visible.