fluid-memory

What problem does it solve?

Fluid Memory solves the problem of unbounded, noisy conversational context by converting transient chat history into a semantic, decay-aware memory store that strengthens frequently accessed facts and archives or deletes low-value information automatically.

Core Features & Use Cases

• Semantic vector storage using ChromaDB for embedding-based recall instead of keyword matching.
• Decay + reinforcement model that reduces the weight of stale memories while boosting items accessed repeatedly (Ebbinghaus-inspired).
• Incremental summarization and buffering to aggregate recent dialog and commit compact summaries to the vector store upon OpenClaw memory flush events.
• Maintenance daemon and archival that nightly consolidates and archives low-value memories and hard-deletes long-archived items.
• Use case: Keep a user's preferences and decisions accessible across sessions while automatically forgetting ephemeral or obsolete details.

Quick Start

Ask the agent to remember that the user prefers cola so Fluid Memory will store that preference during the next OpenClaw memory flush.