One keyEvery model
The AI software engineer for your desktop.
Optamus Code reads your repos, writes the code, runs the tests in a sandbox, verifies in a real browser, and opens the pull request. It runs on the Optamus API — one key that routes to gpt-4o, claude-3-5-sonnet, and every model in between.
MIT-licensed · releases on GitHub · agent backend: https://api.optamus.cloud/v1 · op- keys
Three products, one gateway
Code, Coworker, and the API behind both
Optamus Code — the AI software engineer
Reads your repo, plans the change, edits files, runs tests in a sandbox, verifies in a real browser, and opens the pull request. Four phases · 47s · 0.41M tokens · $0.0041.
Explore Code →"Consolidate the Q3 numbers into the deck before the 2pm review."
send · delete · pay · publish → approval gatesAI Coworker — sees your screen
Watches your screen in real time, understands any application, clicks and types the way a human would — with approval gates on every sensitive action. Sessions recorded for replay.
See Coworker →API Platform — one key
OpenAI-compatible gateway to GPT-4o, Claude, and 100+ open models — per-token billing, per-second features. POST /v1/chat/completions · GET /v1/models · 200 ok · 214ms.
Explore API →From the blog
All posts →The gateway is the product
Why Optamus is built around a routing gateway rather than a model provider. Circuit breakers, per-key rate limits, spending caps, guardrails, and a 5-minute response cache — all enforced before a request ever reaches a model.
Read →EngineeringMetering every token, down to the cent
How usage logging, a per-model cost engine, and Stripe metered billing fit together — and why the number in your usage dashboard is the number on your invoice.
Read →EngineeringGuardrails on the wire, not in a sidebar
PII redaction, prompt-injection detection, and secret-leak blocking run inside the request path. Here is how the scanner works, what it catches, and what it deliberately does not do.
Read →