Download
Features Pricing Docs Enterprise GitHub [150K] Download
Introducing Hackimi 2.0

Code, with an agent that
never loses the thread.

Hackimi is the open-source AI coding agent that lives in your terminal and your IDE. It reads, greps, edits, and tests across your whole repo — local-first, and completely yours.

Read docs → Download

Free for personal use · macOS, Windows, and Linux · MIT licensed

curl -fsSL https://hackimi.sh/install | bash

Requires macOS 12+, Windows 10+, or any modern Linux. No GPU required to start.

3.4×
faster PR merge time on teams that run Hackimi
9.2M
lines indexed per workspace, per day
27min
average agent session, end to end
[x]

What is Hackimi?

Hackimi is an open-source coding agent you can run anywhere — your laptop, your own GPU box, or a cloud cluster when you need more. It reads your entire repository into context, plans edits across files, runs tests, and shows you every change as a reviewable diff.

One agent, two frontends: a TUI inside your terminal, and a panel inside your editor. Same session, same plan, same privacy. Read the docs or start with a one-line install above.

Features

Everything between your first tab press and your merge button.

Same agent, both worlds — terminal native, IDE friendly, CI ready.

Infinite context

Indexes your whole repo — the code, the tests, and the 9-month-old decision in the comments.

make clean && hackimi index

Agent from anywhere

Queue work from your editor, CLI, or CI. Local or cloud GPUs — same session, same plan.

hackimi deploy app --gpu 4

Rules that stick

One rules file for the team. The agent inherits your conventions, your CI gates, and your API style.

rules: "no fetch without abort signals"

Time-travel diffs

Every agent edit is a reviewable diff. Approve, reject, or fork the session and keep both branches.

hackimi session ls --recent 30

Private by default

Your code never trains anything, and never leaves your machine unless you deploy it yourself.

--privacy "yours, absolute"

Measured, not marketed

Real benchmarks from real repos, published monthly. Try it on your own codebase first.

hackimi benchmark run .
[+]

Built for privacy first

[x] Agentic editing Edits across files with a plan you can review — diff by diff.
[x] Local-first Runs on your machine; cloud only when you ask for it.
[+] Zen access One bill for Claude, GPT, Gemini — switch mid-task. Now in open beta.
[x] Terminal native The same agent in your IDE, your shell, and your CI pipeline.
[-] Open source MIT licensed — fork it, run it, own it, audit it.

Doesn't just autocomplete your code. Understands your codebase.

Other tools see the file you're in. Hackimi sees the convention, the failing test, and the commit you made at 1am — and it knows what to do next.

See what's included →
HackimiOther AI tools
Understands your whole repo✓✗
Inherits your team rules✓✗
Runs reviewable agent diffs✓✗
Local & cloud runtimes✓✗
Your data stays yours✓✗
Testimonials

What engineers say after their third sprint with it.

“It reads an 18-month-old pull request thread and gets it. I stopped explaining our codebase to my tools.”
MCMaya ChenStaff Engineer, Loop
“The agent found a race condition my team had chased for two quarters. In 40 minutes. We run it on every PR now.”
DKDario KovačVP Engineering, Kestrel
“Onboarding shipped faster than the task ticket. I wrote one rules file, and every junior's agent passed review.”
ASAna SilvaTech Lead, Kite
Pricing

Pay for compute. Never for your own code.

Hobby
$0 — forever

For personal projects and learning. Runs entirely on your machine.

Download
  • ✓ Unlimited tab completions
  • ✓ Local agent sessions
  • ✓ Community support
Most popular Pro
$20 /mo

For engineers who ship weekly. Agents on laptop or cloud GPU.

Get Pro
  • ✓ Everything in Hobby, plus:
  • ✓ Cloud agent, 30k context
  • ✓ Team rules & shared sessions
  • ✓ Priority support
Teams
$40 /seat · mo

For teams that want one reviewable plan across every engineer.

Start 14-day trial
  • ✓ Everything in Pro, plus:
  • ✓ SSO and seat management
  • ✓ Admin dashboard & audit log
  • ✓ Dedicated support engineer
[?]

FAQ

Hackimi plans edits across your whole repository before touching a file, and every change lands as a reviewable diff. Other agents autocomplete; Hackimi audits itself.
Yes. Point Hackimi at any open-weights model — hackimi-v2, Llama, Qwen, DeepSeek — or at the Zen marketplace for hosted frontier models. No lock-in.
Everything runs locally by default. Your code, prompts, and outputs never leave your machine unless you deploy them yourself, and no model is ever trained on your data.
The agent is MIT licensed and free forever, for personal use and commercial use alike. You only pay for compute when you need more than your own hardware provides.
Newsletter

Get release notes in your inbox. No spam, ever.

On subscribe you agree to processing per our privacy policy.

Try Hackimi now.

Download for free

Free 2-week Pro trial — no credit card, no fine print.