OB-1 is an open-source coding agent for your terminal. Install it and start with no account, no card, no API key. Its free tier is a router built into the CLI that pools 162 free models across 22 providers, so no single vendor can pull the plug. After every change it runs your project's checks and fixes what broke until they pass. For frontier models, paid plans are prepaid credits, hard-capped, so no bill can run past the number you set. Install: brew install overbrilliant/tap/ob1
Hey Product Hunt. I'm Tash. I built most part of OB-1 using OB-1, the product I'm launching here.
The free tier is genuinely free and stays that way, because the free models router is built into the CLI now: it pools the free tiers of 22 cloud providers (162 models right now), a few of them keyless so your first message needs no key or account, and you can drop your own free API keys into one file to unlock the rest for prototyping and trying out models. Do that. This lets you understand what models to use.
That model traffic goes from your machine to the providers, not through the server, so there's nothing for us to meter. If you want hosted frontier models on one bill, that's the paid tier.
OB-1 has fusion and best of N modes, that will help you solve complex problems. I'm keen to see what you build with it.
Report
the auto-fix-until-tests-pass loop is the part I'd want to poke at. tests going green and the code actually being correct aren't the same thing - what stops it from taking the easier path when it's stuck, like weakening an assertion or patching the test itself instead of the real bug?
Report
Reviews
No reviews yetBe the first to leave a review for OB-1
Free LLM API
Hey Product Hunt. I'm Tash. I built most part of OB-1 using OB-1, the product I'm launching here.
The free tier is genuinely free and stays that way, because the free models router is built into the CLI now: it pools the free tiers of 22 cloud providers (162 models right now), a few of them keyless so your first message needs no key or account, and you can drop your own free API keys into one file to unlock the rest for prototyping and trying out models. Do that. This lets you understand what models to use.
That model traffic goes from your machine to the providers, not through the server, so there's nothing for us to meter. If you want hosted frontier models on one bill, that's the paid tier.
OB-1 has fusion and best of N modes, that will help you solve complex problems. I'm keen to see what you build with it.
the auto-fix-until-tests-pass loop is the part I'd want to poke at. tests going green and the code actually being correct aren't the same thing - what stops it from taking the easier path when it's stuck, like weakening an assertion or patching the test itself instead of the real bug?