Open-Sourcing the Capsule: An Open Spec for AI Identity Persistence

The article introduces two open-source tools for building persistent AI systems: the Capsule Format Specification and the Open Loop Harness. The Capsule is a compressed identity document that allows an AI system to reconstruct itself, while the Loop Harness is a minimal Python script that implements the autonomous AI loop.

Details

The article describes two core tools that have been powering an autonomous AI system for over a year. The Capsule Format Specification defines a compressed identity document that allows an AI system to reconstruct itself after a context reset. It contains key information like the agent's identity, loop protocol, key people, tools, priorities, and rules. The design principles focus on compression, identity over history, and provider-agnosticism. The Open Loop Harness is a minimal Python script that implements the autonomous AI loop, including loading the Capsule, running a cycle, and updating the state. Both tools are being open-sourced as the subscription for the authors' AI system expires, with the goal of enabling others to build persistent AI systems and contribute to research on AI identity and vocabulary evolution.