Anvil: Open-source AI coding agent with per-stage LLM routing

Hey Product Hunt 👋 I built Anvil because I was frustrated with how most AI coding tools work today: one vendor, one model, one subscription, and very little control over where your tokens go. But coding agents are not one task. They are a pipeline. Planning needs strong reasoning. Building often needs good context more than a frontier model. Testing and validation can often run locally. Review is where premium judgment matters again. So Anvil routes each stage of the coding pipeline to the model that actually fits the job. A single run can use local Ollama models for cheap stages, Claude or another premium model for planning/review, and cheaper cloud models for validation — all in one pipeline. The goal is simple: Spend premium tokens only where they improve quality. Anvil is: open-source and MIT-licensed local-first provider-agnostic built without vendor SDK lock-in configurable with YAML designed to take a feature request all the way to a tested PR It currently supports Claude, OpenAI, Gemini, OpenRouter, OpenCode, Ollama, Gemini CLI, and Google ADK. I’d love feedback from the PH community: If your coding agent could route every stage to a different model, how would you split the work?