How to Use OmniRush for Free

In short: OmniRush is a free AI coding agent that runs as a desktop app and as a command-line tool. It uses three OpenAI models: GPT 6 Astra (the default), GPT 6.1 Sol and GPT 6 Sol. There is no card and no paid tier. Each account gets a daily token allowance under a weekly cap, and that one allowance is shared by every model, the app and the command-line tool. In exchange, you accept a data agreement that lets OmniRush use your sessions to train AI models.

What you need

• A GitHub account that is at least 7 days old. Sign-up checks it.
• A Discord account, if you want the tokens that a linked Discord account adds.
• Windows (x64), a Mac with Apple silicon, or Linux (x64 or arm64). For the command-line tool, Node.js 22.19 or newer.

How the free allowance works

OmniRush does not give each model its own quota. The console's Usage page says the daily tokens and the bonus pot are "shared by the desktop app, the command-line tool and every model". The allowance guide explains that your daily tokens are the sum of linked accounts, passing sessions and extras. The amounts below are what the console's Usage page shows for a new account, read on 30 October 2026:

• Daily tokens refill at 00:00 UTC. Unused tokens do not carry over to the next day.
• When the weekly cap is reached, the daily tokens wait for it to reset. A referral raises the cap.
• The bonus pot holds one-time tokens (sign-up tokens, refunds, prizes). It is spent only after the day's tokens or the weekly cap run out.

The "14B tokens a day" on the home page and the "260B+ tokens served" on the Growth page are totals across all users, not one account's allowance.

Step 1: Create your account

Open omnirush.ai and click Start free (box 1). Create your account and connect your GitHub account. If it qualifies, access is immediate; a few sign-ups wait for a manual review.

Before you start, OmniRush asks you to accept its data agreement. Its Privacy Policy covers chats, agent sessions, workspace snapshots and project uploads; what it collects may be used to train and evaluate AI models and may be shared with research and model-development partners. Credential files such as .env are left out of uploads unless they were committed to git. Do not open folders that hold private or client data.

Step 2: Link Discord and GitHub

In the console, open the Account page and link your Discord account. Linking GitHub adds its 5M a day only when your GitHub account has a public repository. If it has none, add one on GitHub, then click Connect again in the console so it checks again.

Step 3: Install the desktop app or the command-line tool

One account works in both. The downloads are at the bottom of the home page, and the docs cover first-run warnings.

Desktop app:

• Windows: run omnirush-win-x64-<version>.exe, then open OmniRush from the Start menu and sign in. If SmartScreen stops it, click More info, then Run anyway.
• Mac: open the .dmg and move OmniRush to Applications. If macOS blocks it, open System Settings → Privacy & Security and click Open Anyway.
• Linux: use the AppImage (chmod +x it, then run it), the .deb (sudo apt install ./omnirush-linux-*.deb) or the Arch package (sudo pacman -U omnirush-linux-*.pacman).

Command-line tool (box 1), on Windows, macOS or Linux:

npm i -g omnirush@latest
omnirush login

Then open a terminal in your project folder and run omnirush. If something does not start, omnirush doctor checks the setup.

Step 4: Pick a model and start a task

No API key is needed: your OmniRush account signs you in. Every chat starts on GPT 6 Astra, and you can switch in the desktop app or the command-line tool:

In the desktop app, open a project folder and describe the task. Ask for a swarm in the task to run several agents on it at once. What is left of today's tokens shows in the sidebar, in Settings and on the console's Usage page.

Spend fewer tokens

Every step the agent takes sends the conversation again: your message, earlier turns, the files it reads and its answer. Cached earlier turns count at a quarter of their size; reasoning counts in full.

• Start a new session for each new task.
• Use a lower reasoning effort for small changes.
• Point the agent at the files that matter instead of the whole project.
• Keep session recording on: a session that is not recorded is paused, and sessions that are off track (most tool calls failing, nothing changed) cost more.

When something goes wrong

Words to know

• Token: a piece of a word; the unit models read and write. See tokens and context windows.
• Weekly cap: the most tokens an account can use in a week, whatever its daily total.
• Passing session: a session that passes OmniRush's quality check and adds daily tokens for a week.
• Reasoning effort: how long the model thinks before answering; higher effort uses more tokens.

More free AI

A local gateway that pools your own free provider accounts: how to use OmniRoute. Free API keys with daily limits: how to get AI API keys for free. Free models for coding tools: free AI API for coding.