LIWM: Auditable, provenance‑based memory for AI agents
LIWM gives LLM agents a structured memory that records where each fact came from, how confident the agent is, and when it should expire. It works as a lightweight CLI and plugin system that plugs into popular agents like Claude Code, Gemini CLI, and Cursor, adding zero‑dependency provenance tracking. Developers can audit and debug agent behavior, ensuring that malicious or uncertain information is flagged and managed. Compared to plain markdown notes, LIWM surfaces contradictions, caps confidence, and supports scoped, decaying memory, making agent reasoning more reliable.
View on GitHub →vyas-devgna/liwm-agent-framework