Grinta
A local-first autonomous coding agent that plans, edits files, runs commands, debugs failures and validates results end-to-end, keeping control, execution and session state on the user's own machine.
Grinta is a local-first coding agent built to complete long, failure-prone software tasks rather than stopping after the first plan. It can inspect a repository, plan a change, edit files, run shell commands and git, use language servers and debug adapters, debug failures, validate the result, and continue until the task is complete, offering a terminal UI as well as a non-interactive, pipe-friendly runner with chat, plan and agent modes.
Its control plane, command execution, session history and checkpoints stay on the local machine, while inference can use a hosted provider, an OpenAI-compatible endpoint, Ollama, LM Studio or vLLM. It is built around retries, circuit breakers, stuck detection and explicit lifecycle states so it can recover from provider timeouts, malformed tool output or context-window limits instead of silently stopping, with task tracking and completion gates meant to reduce false-positive done results, and a durable event ledger recording actions and outcomes for recovery, debugging and audit. Confirmation levels, command risk analysis and workspace boundaries bound what it can do.
It runs on Linux, Windows, macOS and WSL2 under Python 3.12 or 3.13, installed with pipx, and is MIT licensed.