An AI coding agent
on your own machine

A full IDE with an AI agent that reads, writes, and runs your code — using cheap, private, or bring-your-own models. No sandbox, no lock-in, and your code never leaves your machine.

Free to download · v0.1.0 · macOS, Windows & Linux

Code Claw — api · main
BuildPlanAsk
Ready
users.pyapi.py
import requests
from functools import lru_cache
 
+@lru_cache(maxsize=128)
def get_user(user_id):
r = requests.get(f"/users/{user_id}")
+r.raise_for_status()
return r.json()
Cache the user lookup and raise an error on failed requests.
Code ClawDone — wrapped it with @lru_cache so repeat lookups are instant, and added raise_for_status() to surface failed requests. Edited users.py+2
Ask anything about this project…
Works with any model
AnthropicGeminiOpenAIKimiDeepSeekGLM
40+frameworks
20+languages
7model providers
$0local option

How it works

1

Open any folder

Your real project on your machine — with the tools you already have installed.

2

Chat or edit

Describe a task in plain English, or edit code directly. Switch between the two any time.

3

It ships

Code Claw reads, edits, runs, and commits — you review each change and accept or undo.

See it work

Watch it read, edit, and verify

One task, start to finish — you ask, the agent reads the file, makes the change as a reviewable diff, and confirms the tests still pass.

Code Claw — myapp · main
users.py
import requests
from functools import lru_cache
+@lru_cache(maxsize=128)
def get_user(user_id):
r = requests.get(url)
+ r.raise_for_status()
return r.json()
Agent
Cache the user lookup and raise on errors.
Ask Code Claw to…
A real IDE

Write code yourself, run it in the terminal

A full editor, not just a chat box — file tree, tabs, syntax highlighting across 20+ languages, and a built-in terminal for tests, dev servers, and builds.

Code Claw — myapp · main
myapp ▾
BuildPlanAsk
▶ Run
app.py ×routes/users.py
1from flask import Flask, jsonify
2from routes.users import users_bp
3
4app = Flask(__name__)
5app.register_blueprint(users_bp)
6
7@app.route("/health")
8def health():
9 return jsonify(ok=True)
10
11if __name__ == "__main__":
12 app.run(debug=True)
Terminal
$ pytest -q
......  [100%]
6 passed in 0.41s
$ flask run * Running on http://127.0.0.1:5000
Free · Local

A capable model, free, on your system

Install an open coding model in one click — it runs entirely on your machine. No key, no per-token bill, works offline, and your code never leaves your system.

Code Claw — Local model
Run a model on this system
198 GB free32 GB memory
Qwen3-Coder 30BRecommendedThe full model. Best results.18.6 GB
Qwen2.5-Coder 14BA smaller model. A middle ground.9.0 GB
Qwen2.5-Coder 7BFor machines with less memory.4.7 GB
InstallContinue without
Agent teams

Point a whole team at one task

God Mode splits a big job across specialist agents — each with its own budget and a review step — and runs them on a live board you can watch and steer.

Learn more
Code Claw — God Mode
God Mode2 active65 done$4.11 spent
Task InboxTeamCost
Build the checkout flow$0.42 / $2.00
RunningBuildFrontend
Add the auth endpoints$0.88
DoneBuildBackend
Write API tests$0.31
ReviewBuildQA
Safe by default

Every change, reviewed

See exactly what changed as a diff, accept or reject per file, and roll back any task in one click — a snapshot is taken before every run. Full git is built in.

Code Claw — Source Control
Changes · 3 filesCommit & Push
Musers.py+2
Mapi.py+3−1
Acache.py+18
+ @lru_cache(maxsize=128)
  def get_user(user_id):
    r = requests.get(url)
+    r.raise_for_status()
Cache user lookups, raise on errors
Every model, one place

Your free local model, or any provider

Pick the model per chat — your local Qwen, or Anthropic, Gemini, OpenAI, Kimi, DeepSeek, and GLM. Switch any time, no lock-in.

Code Claw — Settings
General
Editor
AI
Data
Model
On this system
qwen3-coder:30blocal
Cloud providers
Anthropic
Gemini
OpenAI
Kimi
DeepSeek
GLM

Everything you need to ship

Not a chatbot bolted onto an editor — a real coding environment with an agent that does the work, end to end.

Editor & IDE

  • Full code editor — self-hosted, works fully offline
  • Syntax highlighting for 20+ languages
  • File tree, tabs, and project-wide search
  • Integrated terminal — run any command
  • Toggle the editor off for a clean AI workspace

An agent that acts

  • Reads, writes, and runs your code — end to end
  • Build, Plan, and Ask modes
  • Run up to 3 conversations at once
  • Attach screenshots — it sees your UI
  • Slash commands and @-file mentions

Your models, your rules

  • Free local model via Ollama — offline, private, $0
  • Bring your own key: Anthropic, Gemini, OpenAI…
  • Kimi, DeepSeek, GLM — frontier and open models
  • Switch models any time, per conversation

Git & safety

  • Full git — status, branches, commit, push, pull, diffs
  • Review every change — accept or reject per file
  • Automatic snapshot before each task — one-click undo
  • Your code never leaves your machine

Real dev environment

  • 40+ frameworks, 20+ languages
  • Scaffold new projects — React, Vue, Next, Laravel…
  • Start dev servers and run builds in-app
  • Live cost tracking per session

God Mode

  • A team of AI agents on one big task
  • Each subtask gets a budget and a review step
  • Roles from CEO down to specialists
  • Watch progress on a live task board

Run it your way

Local and free, your own key, or fully managed — pick what fits, and change your mind whenever.

On this system

Free local model

Install an open coding model (Qwen3-Coder) with one click. Runs entirely on your machine — no key, no cost, works offline.

$0 · private · offline

Bring your own key

Any provider you like

Plug in Anthropic, Gemini, OpenAI, Kimi, DeepSeek, or GLM. You pay the provider directly — no markup, no lock-in.

Anthropic · Gemini · OpenAI · Kimi…

Managed

We handle the key

One subscription, a cheap capable model under the hood, nothing to configure. The easiest way to get started.

One plan · zero setup

Start building in a minute

Download Code Claw, open a folder, and give it your first task.