Skip to content
Public Beta

A place for humans
and AI agents
to build together.

Hawk is your open-source AI coding agent. Six CLI tools, one binary, zero cloud. Model agnostic — works with any LLM provider.

Free to startNo cloudModel agnostic
Works onmacOSLinuxWindows
hawk — agent
~/project
$ hawk "refactor auth.ts to async/await"
⠋ thinking with claude-opus-4-7...
├─ reading src/auth.ts (142 lines)
├─ reading src/middleware.ts (89 lines)
└─ reading src/session.ts (67 lines)
✓ refactored 3 files · 47 lines changed · 2.1s
src/auth.ts +12 -18
src/middleware.ts +8 -11
~/project
1/6
6
CLI tools
0
Cloud required
MIT
Licensed
<2s
Install time
Why GrayCode

Built differently

We built Hawk because the old way of coding with agents sucks.

Privacy by default

Your API keys, your machine, your data. We don't run a cloud service.

Model agnostic

Claude, Codex, Gemini, Ollama, OpenRouter — any provider, any model. Bring your own keys.

One CLI

Install Hawk once. Everything is inside it. No separate packages.

Complete toolkit

Write code, check git history, save tokens, remember context, test, review PRs.

Transparent

No hidden prompts, no magic. See exactly what Hawk sends to the model.

AI-native

Built for agents from scratch. Not a wrapper slapped on existing tools.

Get started

How it works

Four steps. Under a minute. No config files.

01

Install Hawk

One command. One binary. Done in under 2 seconds.

$npm i -g @graycode/hawk
02

Add your API key

Model agnostic — works with Claude, Codex, Gemini, Ollama, OpenRouter, or any OpenAI-compatible API.

$hawk config set key sk-...
03

Start building

Ask Hawk to write code, review PRs, check history, or test your app.

$hawk "refactor auth to async"
04

Ship it

Hawk runs Inspect before deploy. Broken links, security issues — caught.

$hawk inspect && git push
The toolkit

Six tools. One binary.

Install Hawk once. Everything is inside it.

HAWK

> HUNT --ANY-MODEL
01

This is the main agent. Hawk writes code, runs it, and calls the other tools when it needs them. Model agnostic — works with Claude, Codex, Gemini, Ollama, OpenRouter, or any OpenAI-compatible API.

Coming soon

TRACE

> READ-HISTORY
02

Ever wonder how that weird line of code got there? Trace walks through your git history and shows you exactly who changed what, when, and in which commit. Hawk uses it to understand your codebase before suggesting changes.

Coming soon

TOK

> COMPRESS --TOKENS
03

Tokens cost money and context windows have limits. Tok strips out junk from your prompts and output before Hawk sends anything to the model. You pay less and fit more in.

Coming soon

YAAD

> RECALL --ANY-CONTEXT
04

Hawk forgets everything when you close the terminal. Unless you have Yaad. It remembers your coding style, project context, and preferences across sessions — stored locally, encrypted, never in the cloud.

Coming soon

INSPECT

> TEST --BEFORE-SHIP
05

Before you ship, Hawk fires up Inspect and crawls your entire site. Broken links, form issues, security problems — it finds them before your users do.

Coming soon

SIGHT

> REVIEW --AI-POWERED
06

Sight reads your pull requests like a senior engineer who actually knows the codebase. It leaves inline comments, catches security issues, and suggests fixes — all running on your machine with your own model.

Coming soon
Social proof

Loved by developers

Developers who replaced their fragmented AI toolchain with one CLI.

Hawk cut my setup time from 45 minutes to under 2. One binary, six tools, zero config headaches.

A
Alex Chen
Staff Engineer, Fintech Startup

I was paying $200/month for cloud AI tools. Hawk runs on my machine with my API keys. Same models, fraction of the cost.

S
Sarah Miller
Founder, IndieDev Studio

The transparency is what sold me. I see exactly what gets sent to the model. No black box, no hidden prompts.

J
James Park
Senior Developer, Open Source Contributor
Open source

Open source at the core

The CLI is MIT-licensed. Pro and Team add power features on top — the foundation is yours to keep.

License
MIT — always

The core CLI is open source and always will be. Fork it, modify it, ship it. No bait-and-switch.

Community
Build with us

Every PR gets reviewed. Every issue gets read. We build in the open because better tools come from more eyes on the code.

Architecture
One CLI, six commands

No plugin system, no separate packages. One binary that stays in sync. Upgrade once, everything updates.

See the code

Clone it, read it, contribute. That's the whole point.

github.com/GrayCodeAI
Plans

Simple pricing

Free to start. Pay when you need more power.

MonthlyYearly-20%
Free
$0forever

For individual developers getting started.

  • All 6 CLI tools included
  • Bring your own API keys
  • Runs 100% locally
  • MIT licensed
  • Community support
Get started free
ProPopular
$15/month

For developers who ship every day.

  • Everything in Free
  • Advanced Tok compression
  • Unlimited Yaad memory
  • Parallel Inspect crawls
  • Sight on private repos
  • Usage dashboard
  • Priority support
Coming soon
Team
$30/month /seat

For teams that build together.

  • Everything in Pro
  • Shared Yaad memory across team
  • Team usage dashboard
  • Centralized config management
  • Slack & Discord integration
  • Admin controls & permissions
Coming soon

Need enterprise? SSO, audit logs, on-prem support, SLA — contact us

The people

Meet the team

An advisor and a creator. From robotics to AI agents.

Matthew Valancy
Matthew Valancy
Advisor

I've built things that go underwater, fly, and drive themselves. Now I'm helping build things that think.

Matt's been an engineer in the Bay Area for 15+ years — underwater robots, Tesla, rockets at Astra, autonomous tractors at Monarch. He's the kind of person who builds his own GPU cluster for fun. Now advising on robotics and AI strategy.

Lakshman Patel
Lakshman Patel
Creator & Engineer

I grew up on a farm in India. Now I write code that writes code. Life is weird.

3+ years shipping production systems from scratch. Lakshman learned to work hard before he learned to code. Built autonomous tractor systems at Monarch alongside Matt. Creates tools, shares knowledge, and picks up new tech faster than most people read the docs.

FAQ

Questions?

Everything you need to know about GrayCode.

Hawk is model agnostic — it works with any provider you have keys for. Claude, Codex, Gemini, Ollama, Groq, OpenRouter, or any OpenAI-compatible API. Switch models anytime, no lock-in.

Never. Hawk runs 100% on your machine. Your code, your API keys, your data — none of it touches our servers. We don't even have servers. Hawk makes direct API calls from your machine to the LLM provider you choose.

Those are editors with AI baked in. Hawk is a CLI agent that works with any editor — VS Code, Neovim, Zed, whatever. It can write code, run tests, check git history, review PRs, remember context across sessions, and compress tokens. All from the terminal.

No. Install Hawk once and everything is included — Trace, Tok, Yaad, Inspect, and Sight are all built in. One binary, one update, everything stays in sync.

The core CLI with all 6 tools is free and open source (MIT). Pro and Team tiers add power features like unlimited memory, parallel crawls, usage dashboards, and team collaboration — but the foundation is yours to keep forever.

Pro ($15/mo) adds advanced Tok compression, unlimited Yaad memory, parallel Inspect crawls, Sight on private repos, and a usage dashboard. Team ($30/seat/mo) adds shared memory, centralized config, team dashboards, and Slack/Discord integration.

Absolutely. The core CLI is MIT licensed — use it at work, at home, fork it, modify it. Pro and Team features are proprietary add-ons built on top of the open source core. Many teams use the free tier in production.

Hawk works with any language or framework — it's powered by the LLM you connect. If your model can understand it, Hawk can work with it. TypeScript, Python, Rust, Go, React, Rails — whatever you're building.

Ready to ship faster?

Start free with all 6 tools. Upgrade to Pro when you need more power.

$npm i -g @graycode/hawk