Discover the open-source persistent memory layer for AI agents. Let your AI remember facts, track milestones, and persist context across sessions locally.
Most local AI sessions are ephemeral. The moment you close Claude Desktop or Cursor, your agent forgets the structural decisions, API signatures, and project context it just spent an hour learning. You start the next session by repeating yourself.
Today we are introducing the open-source AgenticMemory layer to AgenticStore MCP Tools. This memory layer allows your AI to persist facts, save session checkpoints, and track project milestones indefinitely.
The memory store acts as a local key-value and hierarchical document store situated in your ~/.config/agentic-store/memory directory. Instead of using an external database, simple JSON and Markdown files store everything so you can review and edit them in plain text.
Facts Store
Save key-value pairs (API styles, language preferences) using memory_write and memory_read.
Session Checkpoints
Snapshot active branches, plans, and open files using memory_checkpoint.
Changelogs
Ask the agent to summarize and append its own work to an immutable JSONL log.
Milestones
Maintain real progress tracking in plan.md across multiple chat threads.
To prevent context bloat, the memory system is split into two layers: Storage Primitives (6 tools) and Productivity Layer (6 tools). You don't have to explain to the LLM how to manage files; the MCP tools handle the heavy lifting.
Integrate it via PyPI
Install the memory module globally. No signups, API keys, or subscriptions required.
pip install agentic-store-mcp --upgrade