What is aesop?

Aesop is a system that runs autonomous AI coding agents on a repository, coordinating multiple smaller AI workers to build software largely on their own. It picks tasks from a backlog, splits the work so each agent owns a separate set of files, runs the agents in parallel, checks the results before merging them, and remembers what it learned for the next round. Its state is saved to disk, using SQLite plus files rendered into git, so progress survives a machine restart and can be shared across a team. By default it relies mostly on Anthropic's cheaper Haiku model to keep costs down, but it can also connect to other AI backends such as Ollama or OpenRouter, and it adjusts how strict its safety checks are depending on how capable the connected model is. The project claims it was used to build a large part of its own codebase, over 220 merged changes across many commits, through its own automated build cycle, supervised by a human who sets goals and approves things like publishing new versions. It includes a command line dashboard for watching agent activity, cost, and a task board in real time, a health check tool that reports whether the environment is set up correctly, and a background process that regularly backs up work, scans for accidentally leaked secrets, and restarts stalled agents. A pre push check can block a commit if it detects a leaked secret. Installation is done through npx, and after that the user copies a skills folder into their Claude configuration to enable two custom commands used to run the agent cycles. The project is source available under the PolyForm Strict License 1.0.0, meaning anyone can read, run, and use it, but modifying or redistributing the code is not allowed without permission from the author.