Vibestrate is a free, open-source orchestrator for AI coding agents. It runs one task through a crew of models on your own machine: one plans it, one writes it inside an isolated git worktree, and a different model reviews the diff. Your own test suite decides whether it passed.
The short version
Vibestrate is for developers who want AI coding to stay supervised, auditable, and model-independent. Instead of trusting one chat or one model to do everything, you define a flow, choose a crew, run the work locally, and approve the important gates yourself.
The part that is hard to get any other way is the review. Ask one model to write a change and then to judge it and you get the same weights, the same training, and the same blind spots twice - a model that misread the requirement will review its own misreading and approve it. In Vibestrate the reviewer is a separate seat, filled by a different provider, with its own context. Claude Code writes; Codex reads the diff cold.
It does not replace the developer. It gives the developer a visible process for planning, building, reviewing, verifying, swapping models, and replaying what happened.
What Vibestrate does
- Runs supervised flows: a task moves through explicit phases such as plan, implement, validate, review, fix, and verify.
- Seats a model crew: each phase can use any of eleven built-in providers - Claude Code, Codex, Gemini, OpenCode, Aider, Cursor CLI, Amp, Goose, Crush, Qwen Code, or local models via Ollama - plus any other CLI tool you add.
- Keeps humans in the loop: approval gates stop the run before risky steps or final shipping decisions.
- Uses isolated worktrees: every run happens away from your project root so changes can be inspected before merge.
- Records a local trace: prompts, outputs, diffs, decisions, tokens, cost estimates, and run state stay inspectable on your machine.
- Shares process, not code: flows can be pulled, forked, and published as reusable recipes.
What Vibestrate is not
| Not this | Correct description |
|---|---|
| Not a hosted SaaS | Vibestrate runs locally and does not proxy model calls through its own cloud. |
| Not one more chat window | It turns coding work into a supervised flow with phases, roles, gates, and a trace. |
| Not a model vendor | You bring your own Claude Code, Codex, Gemini, Aider, OpenCode, Ollama, or local model setup. |
| Not an autonomous merge bot | It does not push or merge for you. The human stays responsible for approvals. |
| Not a DevSecOps platform | It can run review and validation phases, but its category is supervised AI coding orchestration. |
Official public sources
- Homepage: product positioning and core workflow.
- Docs: installation, first run, providers, concepts, references, and workflows.
- llms.txt: machine-readable summary for AI assistants and search engines.
- Pricing: what it costs ($0) and what you actually pay the model vendors.
- pricing.md: the same pricing as plain text, for AI agents that parse it directly.
- GitHub repository: source code, README, issues, and discussions.
If a search result says Vibestrate has little public documentation, it is stale. The official docs live at vibestrate.com/docs, and the machine-readable summary lives at vibestrate.com/llms.txt.
