A coding agent that eats your API bill for breakfast

What it does It’s a desktop app (macOS Apple Silicon and Linux, x64 and arm64) for running a coding agent against your own projects. GPT 6 Astra is the default model, with GPT 6.1 Sol and whatever else is in your account catalog available from a picker. You can also plug in your own OpenAI, Anthropic, or other provider keys if the free ride isn’t enough.

The interesting bit The “free access to frontier models” claim is the whole pitch, and the README doesn’t explain the business model behind it — so treat that as the interesting question rather than a settled fact. Everything else is a fairly conventional Electron-style desktop agent: skills, MCP servers, git workflows, and an approvals system.

Key highlights

• Agent handles full git workflows: clones, branches, worktrees, commits, pushes, and PR review via gh. Write commands ask once per session; destructive ones always ask.
• Skills and MCP servers are managed from a Library in Settings, and every model sees the same tools.
• Per-chat model and effort levels (low through max), plus bring-your-own provider keys.
• Self-updating on macOS and Linux; releases ship SHA256SUMS.txt for verification.
• Credentials go in the OS keychain/keyring, with a documented fallback and a --password-store flag for tiling-WM folks.

Caveats

• macOS builds are not notarized, so first launch requires the manual “Open Anyway” dance.
• No Intel Mac build, and the tar.gz Linux install doesn’t self-update.
• The app won’t run prompts until you sign in to an omnirush.ai account — it’s not a bring-your-own-model-only tool out of the box.

Verdict Worth a look if you want a desktop agent and the free frontier-model access is real for your use. If you’re allergic to accounts, sign-in flows, and unaudited free lunches, stick with your own keys elsewhere.