exo
An AI agent harness that can inspect and safely edit its own code, prompts, memory and tooling at runtime, so it can incrementally improve itself and even manage a lineage of its own clones.
exo is a complete AI agent harness, similar in scope to OpenClaw, Pi or Hermes, with tools, tasks and integrations, but with one key difference: it has full visibility into its own code and runtime logs. That visibility lets it incrementally improve any part of itself, from prompts to memory to tooling to its own harness policy, and even clone itself and manage a lineage of clones.
An append-only event log gives it a canonical history of what it has already tried, so self-modification does not get stuck in a loop. A custom agent can be written and run from a Markdown file with exo agent run --agent-file. It requires an OpenAI or OpenRouter API key, plus git and Docker, and installs with a setup script.