bradsadevnow/claw_harder

This is an open fork of OpenClaw, a runtime experiment project by Brad Bates, released under the MIT license. The repository contains several components: a mirror of the OpenClaw base code, a plugin for OpenClaw II that implements a "manifest-engine" for managing context, a governed runtime substrate prototype, and architecture documentation. The project focuses on building a structured memory system for AI-like runtimes. The memory design distinguishes three layers: STM (short-term memory, described as a rolling epoch window), MTM (session ledger), and LTM (long-term compressed artifacts with a deterministic table of contents). The architecture also includes concepts of "organism-centric state", tracking identity, affective state, and short-term memory as core architectural elements, with a governance layer that controls what gets admitted into memory. Work in progress includes formalizing memory contracts, cleaning up runtime startup and replay behavior to be deterministic, and wiring live modulation through a governed signal flow. The quick-start instructions show how to build and install the OpenClaw II plugin using npm, then configure OpenClaw to use the manifest-engine as its context engine. This is an early-stage research and experimentation repository, likely useful for developers exploring agent runtime architectures, memory management for AI systems, or extending the OpenClaw platform with custom plugins. Written primarily in TypeScript.