Anvil
A self-hosted, provider-agnostic AI coding pipeline that runs clarify, plan, build, review and PR stages, mixing whichever LLM provider is cheapest per stage.
Anvil is an open-source, self-hosted coding-agent pipeline that carries a feature end to end through clarify, plan, build, review and PR stages, across every repository in a project. Each stage can use a different provider, for example planning on Claude, building on Ollama, reviewing on GPT and shipping on a local model, using the user's own API keys with no markup and no hosted plan.
Every HTTP adapter is hand-written with a plain fetch call rather than a vendor SDK, so the project avoids depending on the Anthropic SDK, the OpenAI package, LangChain or the Vercel AI SDK. It speaks the Model Context Protocol and can run entirely offline through Ollama, or through OpenCode's flat-rate subscription in place of per-token cloud billing.
It installs as an npm CLI; anvil init scaffolds a home configuration directory, a models file and a project configuration file, and anvil dashboard opens a web UI showing pipeline orchestration, live agent activity, a knowledge graph and a running cost ledger.