robotmem

Overview

robotmem is a specialized MCP server designed for AI robots to learn from experiences. It stores and retrieves episode experiences including parameters, trajectories, and outcomes using a hybrid search approach combining BM25 full-text search with vector embeddings. The system features structured experience retrieval through JSON context filtering and spatial sorting capabilities, making it particularly suitable for robotic applications where spatial relationships matter. It operates entirely on CPU with a local SQLite database, requiring no GPU infrastructure.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• learnlearn(insight, context)

  Record physical experiences (parameters / strategies / lessons)

  Try:Record that grip force of 12.5N yields highest grasp success rateTry:Learn from this experiment that pushing with force 15N worked best
• recallrecall(query, context_filter, spatial_sort)

  Retrieve experiences using BM25 + vector hybrid search

  Try:Find experiences with successful push tasksTry:Retrieve memories about grasping objects at position [1.3, 0.7, 0.42]Try:Show me successful experiments with low final distance
• save_perceptionsave_perception(description, perception_type, data)

  Store perception / trajectory / force data

  Try:Save the grasp trajectory data from this successful attemptTry:Store this tactile perception of a soft object
• forgetforget(memory_id)

  Delete incorrect memories

  Try:Remove this failed experiment from memoryTry:Forget the experiment where the robot dropped the object
• updateupdate(memory_id, new_content)

  Correct memory content

  Try:Update the memory to reflect the correct grip forceTry:Correct this memory with the successful parameters
• start_sessionstart_session(context)

  Begin an episode

  Try:Start a new session for the push taskTry:Begin episode with robot arm-01 and task: grasp
• end_sessionend_session(session_id)

  End an episode with auto-consolidation

  Try:End the current session and consolidate memoriesTry:Finish the experiment and trigger proactive recall

Comparable tools

Installation

pip install robotmem

{
  "mcpServers": {
    "robotmem": {
      "command": "python",
      "args": ["-m", "robotmem", "mcp"]
    }
  }
}

FAQ

What makes robotmem different from other memory systems?

robotmem is specifically designed for physical robots with structured experience storage, spatial retrieval, and context filtering capabilities not found in general-purpose memory systems.

Does robotmem require a GPU to run?

No, robotmem is CPU-only and uses FastEmbed ONNX models for vector search, requiring no GPU infrastructure.

On Hacker News

Recent discussion from the developer community.

• Show HN: I gave my robot physical memory – it stopped repeating mistakes▲ 18💬 4
  Story by robotmem · 2026-03-09

Compare robotmem with