memory-system

To persist AI agent context across multiple sessions, this system uses a three-layer memory stack storing short-term work-in-progress, mid-term session summaries, and long-term knowledge. It enables persistent personalization for multi-turn interactions and long-running tasks.

The best way to manage cross-session memory for personalized assistants is using a three-layer architecture: a short-term buffer, date-slotted JSONL session summaries, and SQLite with optional vector indexing for semantic search and personalization.

You store and retrieve memory entries via CLI using memory_cli.py. The CLI-driven operations allow you to maintain cross-session continuity by saving current session entries and retrieving related memories through minimal YAML configuration.

Yes, cross-session memory with vector indexing works with SQLite. The long-term memory layer uses SQLite for storage and applies optional vector indexing to enable semantic search and personalization across sessions.

Yes, you can use Ollama-backed embeddings for semantic memory search. The memory-system supports Ollama as an optional embedding provider alongside a minimal YAML and CLI configuration covering memory, llm, and embedding providers.

Yes, chromadb is required as a dependency for cross-session memory operations. The system also requires requests and pyyaml, and includes initialization checks and error handling to manage memory across multi-session workflows.