Docs
Run OB-1 free, local, or hosted.
This is the launch documentation map for developers evaluating OB-1. The CLI is Apache-2.0, provider-neutral, and keeps working without the hosted backend.
brew install overbrilliant/tap/ob1ob1Start freeQuickstart
Install OB-1 and run ob1. The first message is answered by a free cloud model with no key and no account; add keys to ~/.ob1/keys.env to activate more providers.
Install
Homebrew, curl installer, npm, and source checkout paths for macOS and Linux, plus post-install checks.
OpenFree models
OB-1 embeds a free-models router in-process, with no second server to run. Keyless providers answer with no setup; add keys to ~/.ob1/keys.env for higher limits. Free users get newly released free models after 30 days, while hosted plans unlock them immediately. Automatic failover, rate-limit cooldowns, and provider health checks are built in.
OpenFree-tier capacity
The launch ledger behind the free-tier headline: what counts, what needs provider verification, and how to avoid overclaiming free-tier capacity.
OpenProviders and models
Use the free default, a custom endpoint, OpenRouter, Ollama, LM Studio, llama.cpp, vLLM, Groq, Gemini, or the hosted frontier catalogue. OB-1 keeps the wire OpenAI-compatible.
OpenCore concepts
Sessions, model routes, permissions, sandbox modes, checkpoints, rewind, and the local memory model.
OpenMemory and checkpoints
OB-1 keeps durable project memory in readable files, records relationships and reflections, and creates checkpoints so work can be inspected and rewound.
OpenMulti-agent modes
Two modes — Solo (default) and Fusion best-of-N — plus read-only subagents for parallel investigation. Fusion runs several candidates and selects the one that passes your project's real tests and checks, not a model's self-vote. On a verified failure, Solo escalates to Fusion once, so you only pay for extra compute when the checks prove you need it. We measured our own heavier modes against a single agent at equal compute and deleted the ones that lost.
OpenCommands
Slash commands for everyday work: /mode solo|fusion, /review (an independent reviewer that tries to refute its own findings first), /deep (an adaptive generate-vs-refine search), /escalation, and /eval — the compute-matched test that decides whether a mode earns its tokens.
Skills and MCP
Markdown skills capture reusable procedures. MCP connects external tools over stdio, HTTP, or SSE while staying inside OB-1's tool controls.
OpenHosted API
Subscribers get /v1/models, /v1/chat/completions, /v1/search, /v1/usage, and usage CSV export with bearer-token auth.
Configuration
OB1_BASE_URL plus OB1_API_KEY override everything. Provider env keys such as OPENROUTER_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, and GROQ_API_KEY can be used without writing secrets into settings.
Compare OB-1
See how OB-1 differs from Claude Code, Gemini CLI, opencode, and aider across free usage, local models, memory, routing, and hosted options.
OpenChangelog
Readable release notes for the free-by-default launch path, provider routing, docs, demos, and contributor work.
OpenTroubleshooting
If free models are slow or rate-limited, open /free in OB-1 to check provider health, switch routing strategy, or add provider keys. Keys live in ~/.ob1/keys.env and take effect on your next message. If the browser tool needs Playwright and the browser binary is missing, install it then; the OB-1 install should not fail without it.