zo-memory-system

Persistent memory for Ollama personas is established by storing episodic data and facts in a vector store with semantic search. This allows multi-agent swarms to recall context and share memory across sessions without manual re-entry.

Graph-boosted retrieval links related episodic and procedural memories through graph-based connections, enhancing standard vector semantic search. It improves context recall in multi-agent swarms by preserving provenance and linking decisions across tasks.

Cross-persona memory sharing is supported by storing memories in a shared persistent store with embedding management. This enables multiple personas within a swarm to access and build upon facts and procedural steps captured by other agents.

Auto-capture of facts from conversations and swarm outputs is configured by initializing the memory system and adding a persona. It automatically tracks episodic and procedural memory, injecting relevant context into future tasks via memory-gating.

Hybrid search combines full-text search with vector semantic embeddings to retrieve relevant persona memories. A memory-gate mechanism then determines which retrieved context is automatically injected into the active session.

Episodic memory tracking is needed for long-running sessions where preserving context and provenance across tasks is critical. It allows multi-agent swarms to remember past decisions and procedural steps, enabling faster iteration without context loss.