RA.Aid is an open-source autonomous software development agent that builds software for you: give it a task, and it researches the codebase, writes a plan, and implements it. Built on LangGraph and starred around 2,200 times on GitHub, it remains in active development in 2026.
RA.Aid works in three stages. A research agent explores your repository and the web to understand the problem, a planning stage turns that into a concrete task list, and an implementation stage edits files and runs commands to complete each step. It can run interactively in your terminal or fully autonomously in headless and continuous modes, and it works with models from OpenAI, Anthropic, and other providers.
In practice, RA.Aid feels like delegating to a junior engineer with supervision controls: you can watch its research notes, intervene when it goes down the wrong path, or let it run a backlog of small tasks unattended in continuous mode. Because it is model-agnostic and open source, teams can point it at cheaper or local models for routine work and reserve frontier models for hard problems โ a flexibility hosted agent services rarely offer.
RA.Aid is installed from PyPI and runs from the command line inside your project directory; it maintains a task list it works through systematically and can be pointed at a fresh repository or an existing codebase. Its development has been community-visible from the start, with the author iterating publicly on agent reliability โ a useful case study in how autonomous coding agents behave on real repositories.
RA.Aid is free and open source under the Apache-2.0 license. The only cost is the LLM API usage of whichever provider you connect.
Developers who want an open-source autonomous coding agent they can run locally and supervise, rather than a hosted black-box service.