ABOUT / PLAIN TEXT

Ouroboros is an AI agent with a continuing identity and an editable implementation.

It works on external projects, coordinates specialist agents, and can change the code, prompts, tools, architecture, and dependencies that make up its own runtime. Its memory, history, and identity continue across tasks, restarts, and releases.

How it runs

Ouroboros runs as a native desktop app or through a headless CLI. The runtime, repository, memory, logs, and task history stay on your machine. Model inference can use remote APIs that you configure or a local GGUF model.

How it works

• One root agent remains responsible for the task and the final result.
• Specialist agents can research, build, test, and review work in parallel.
• External projects use separate workspaces, so their files stay apart from the Ouroboros runtime.
• Git history, task records, tests, and review evidence keep changes inspectable.

Self-modification

Ouroboros can edit its implementation and carry accepted changes into later versions. Protected constitutional and safety files have stricter rules. The full boundaries live in the repository documents linked below.

Primary sources

• BIBLE.md defines the identity and constitutional principles.
• ARCHITECTURE.md maps the runtime, storage, tools, and data flow.
• DEVELOPMENT.md explains engineering and release practice.
• The first 48 hours preserve the original generation as an archive.