Build something new
Describe an idea. The agent makes a folder, writes the code, and runs it.
“Build something new from scratch… My idea: a habit tracker I can use on my phone”
Harness · Graff built in
Put coding agents to work. See what works.
Harness brings Graff and your other coding agents into one desktop app. Give them a task, follow the changes, and inspect the result. No code yet? Pick a starter.
Apple Silicon · macOS 12+ · Windows x64Other downloadsGitHub Getting started
Use your Codex, Kimi or Grok subscription, or another supported provider. Compare connections · Run your first repository task · Verify the result
Install via terminal
curl -fsSL https://github.com/justrach/codegraff/releases/latest/download/install.sh | shWindows · PowerShell, no WSL needed
irm https://codegraff.com/install.ps1 | iexStart from a template
Start with a sentence.
Open Harness, choose a starter, and add one line about what you want. The agent asks what it needs, then begins the work on your computer.
Research a topic
Get a sourced write-up on anything, with the key takeaways up front.
“Research the topic below and give me a concise, well-sourced write-up… Topic: how small cafés run loyalty programs”
Plan a launch
Positioning, a go-to-market plan, and landing page copy for your product.
“Help me take my product to market… The product: a meal planner for busy parents”
Tour this project
Learn what the code does, how it's organized, and where to start.
“Give me a tour of this project: what it does, how it's organized, how to run it”
Free · macOS and Windows · no account needed to start
Starters for anything
Build, research, or plan a launch from a template. Add one sentence and the agent takes it from there.
Every agent in one place
graff, Claude Code, Codex, Cursor, Devin, Grok, Hermes, Pi, OpenCode and Antigravity. Pick the agent and model for each conversation.
Use the plan you already pay for
Sign in with your ChatGPT (Codex), Kimi or Grok subscription, or use Codegraff credits for many models behind one key.
See every change
Review what the agent changed beside the chat, with files, a browser and a terminal close at hand.
Pick up on any device
Sign in to follow and steer a session from your other computers or your iPhone while it keeps running at home.
Local first
Harness starts without an account. Nothing leaves your machine for sync until you turn it on and choose what to bring.
Harness for iPhone and iPad · Coming soon
Your agents, in your pocket.
Start a task at your desk and keep up with it from your phone. The sessions keep running on your computer; the iOS app follows and steers them.
- Every session from your desktop, live on your phone
- Reply, approve tools and follow the transcript as it streams
- Switch between your machines from one list
Join the iOS waitlistOne email when it is ready to try.
From model to working environment
What is a coding agent harness?
A coding agent harness connects a model to repository context, file edits, terminal commands and tests. It manages the steps between a task request and the resulting code.
Execute
Run repository tasks
Edit files, run terminal commands, call MCP tools, route across models, and coordinate coding-agent workers.
Evaluate
Measure agent outcomes
Score success, cost, speed, tool errors, and verification evidence instead of trusting a confident final message.
Improve
Reuse stronger strategies
Keep higher-fitness prompts and agent strategies so future runs reuse what worked while preserving useful diversity.
Graff records how strategies perform and uses those results to select future configurations. You can inspect the code, evaluation setup and recorded outcomes.
20/21
terminal passes · Graff / Grok 4.6
17/21
terminal passes · Graff / Kimi K3
$0.31
per pass · recorded Grok run
$0.21
per pass · recorded Kimi run
Later FrontierHarness runs on the selected 21-task terminal slice, with extra evaluation instructions. Cost per pass uses historical list-price totals and includes failed attempts. Explore the results and comparison conditions →
Tune it to your team
What would graff save you?
Pick the agent you use today, then set your team's numbers.
Measured on 21 tasks × 3 runs, every run passed, same model on both (one gateway model). Our tasks are not yours: treat this as a ratio, not a promise.
graff against Claude Code
35%less agent time per task
74hagent hours saved / month
$1,408saved each month, out of $2,000 model spend (70%)estimated from prompt tokens per task (only tokens were compared)
$5,559of team time saved each month, at $75 / h
Money saved plus time saved: +$6,967 / month.
Bring the plan you already pay for
On ChatGPT Enterprise or Grok? Run graff on it.
Sign in once and graff runs on your own plan, straight from your machine to the provider. No API key, no Codegraff credits, and your plan's limits and data terms apply.
ChatGPT
Plus, Pro, Business, Edu and Enterprise
graff login chatgptGrok
SuperGrok and X Premium
graff login xaiKimi Code
Kimi Code plans
graff login kimiZ.AI
GLM Coding Plan
graff login zaiBusiness, Edu and Enterprise workspaces may need an admin to allow Codex and device-code sign-in. Connection details →
Graff's core loop
Keep a record of what works. Use it on the next task.
Graff turns completed work into a fitness signal, keeps a diverse archive of strong strategies, and uses that archive to guide future runs.
01Run
Try agent variants.
Graff samples agent strategies and runs them against real repository work.
02Score
Measure the outcome.
Record completion, cost, time and errors for each strategy.
03Select
Keep useful strategies.
Store selected variants in a MAP-Elites archive, grouped by task and provider class.
04Reuse
Reuse selected strategies.
Future runs sample stronger strategies more often while preserving useful diversity.
Graff owns the evolutionary loop. Harness is where you run it, next to the other coding agents you already use.
The short answers.
What do the Graff benchmark results show?+
The later FrontierHarness runs passed 20 of 21 selected terminal tasks with Grok 4.6 and 17 with Kimi K3. Both used additional evaluation instructions. These are results for that task slice and setup, not full Terminal-Bench leaderboard scores. The comparison article includes costs, DeepSWE outcomes, source records and evaluation conditions.
What is Codegraff?+
Codegraff is the project behind Graff, a free, open-source AI coding harness. Graff runs repository tasks with your model and tools, evaluates results, and reuses selected strategies. CodeDB adds local code search and context. CodeDB Pro adds batching and editing tools.
What is a coding agent harness?+
A coding agent harness connects an AI model to the tools it needs to work: repository context, file edits, terminal commands and tests. It also manages execution and records outcomes. Graff combines that execution layer with evaluation and strategy selection.
How does Graff reuse what works?+
Graff evaluates agent variants and stores selected strategies in a MAP-Elites archive. Later runs can draw on those strategies while continuing to explore alternatives. Selection changes the harness configuration; the base model weights stay the same.
What is Harness?+
Harness is the desktop app for macOS and Windows with Graff built in. It also runs Claude Code, Codex, Cursor, Grok, Hermes, Pi and other installed agents in one window. It starts without an account, and nothing leaves your machine for sync until you turn it on.
Your next task starts here
Try Graff on your repository.
Download Harness, connect a model, and give Graff a task. Follow the changes and check the result for yourself.
More about Harness →