razzant/ouroboros

What it solves

Ouroboros is a general-purpose AI agent designed to maintain a consistent identity, durable memory, and history across tasks and restarts. It addresses the limitation of stateless AI agents by allowing the agent to remember its past actions, reflections, and knowledge, and even modify its own implementation—including its code, architecture, and prompts—to evolve autonomously over time.

How it works

Ouroboros operates as a native desktop application or a headless CLI. It manages its own repository, durable memory, and history locally on the user's machine. For cognitive processing, it can use remote APIs or local GGUF models. It uses a "reviewed core-evolution system" where improvements are turned into reviewed changes in its Git history. To handle complex tasks, it can coordinate a "live swarm" of specialist agents (planners, builders, researchers) that work in parallel and return results for integration.

Who it’s for

It is designed for developers and power users who need an AI agent capable of long-running work on external projects, autonomous self-improvement, and a persistent identity that grows through experience.

Highlights

• Self-Modifying Code: Can rewrite its own application code, architecture, prompts, and tools.
• Durable Memory: Maintains identity and history across restarts and different tasks.
• Harness-Agnostic: Supports remote provider APIs and local GGUF models.
• Agent Swarms: Coordinates multiple specialist agents to solve complex problems in parallel.
• Local-First Privacy: Sends no telemetry by default and keeps history and memory on the local machine.
• External Project Support: Can operate on separate Git workspaces while maintaining its own distinct governance boundary.