AI Systems Memory — Persistent Knowledge and Agent Memory
Persistent knowledge beyond a single chat thread.
This section collects guides on persistent knowledge and memory for AI systems — how assistants keep facts, preferences, and distilled context across sessions without stuffing every token into one prompt. Here, memory means intentional retention (user facts, summaries, plugin-backed stores), not GPU RAM or model weights.
It complements the broader AI Systems cluster — OpenClaw, Hermes, orchestration — and sits beside RAG for retrieval mechanics and LLM Hosting for running models.
Agent memory providers
Drop-in backends exposed by frameworks such as Hermes Agent and OpenClaw — Honcho, OpenViking, Mem0, Hindsight, and others — with different LLM, embedding, and database trade-offs.
- Agent memory providers compared — full table, dependency notes, and Hermes
memory setupflows
For Hermes-only bounded core memory (MEMORY.md / USER.md), see Hermes Agent Memory System.
Knowledge graphs and Cognee
Institutional and project knowledge extracted into graphs for retrieval-aware assistants.
- Self-Hosting Cognee — Choosing LLM on Ollama — hands-on Cognee quickstart with local models
- Choosing the Right LLM for Cognee — Local Ollama Setup — model comparison for graph quality vs hardware
Graph builders such as Cognee typically ingest Markdown vaults, wikis, or exports that people have already edited—salience, naming, and “why this mattered” are largely settled before chunks hit embeddings. A sloppy upstream corpus trains ambiguity back into the assistant; disciplined capture-through-expression workflows limit that damage. For that human-centered framing—including how it differs from retrieval-first RAG—see Second brain explained for engineers.
