Memdoor is an open-source coding agent for the terminal. You run it in a project, tell it what to do, and it reads the code, writes the change, runs the build and tests, and fixes what it broke. It uses your own API key from OpenRouter, Anthropic, OpenAI, Gemini, DeepSeek, Groq, xAI or Baseten; nothing passes through Memdoor's servers.
You can also describe a job in steps ("vet and test in parallel, write the release notes, wait for me, then tag and deploy"). Memdoor writes one task file per step and runs them as a graph: independent steps in parallel, a step done when its output exists or its command exits 0, a step where the run stops for your approval, and a failed run that resumes at the failed step.
A decision model in front of the chat model keeps what it reads to what the task needs, which cut input tokens by 26–49% on measured runs.
One static binary for macOS, Linux and Windows. Apache 2.0. Free on your own key; a $10/month plan adds remote control from a phone and workflow state kept on memdoor.ai.
No comments or reviews, maybe you want to be first?