engram

What This Skill Does

Engram is a persistent, semantic memory layer designed to overcome the limitations of standard AI session windows. By utilizing local-first storage architecture (combining SQLite for relational data and LanceDB for vector search), Engram enables your AI agent to remember facts, decisions, preferences, events, and relationships across different operational sessions. It runs entirely locally on your machine at localhost:3400, ensuring your proprietary data never leaves your environment. Engram utilizes the 'nomic-embed-text' model via Ollama to map your interactions into a searchable, high-dimensional vector space. Key features include automatic memory extraction from raw text, intelligent relationship mapping, semantic search, and context-aware recall that ranks data based on recency, salience, and access frequency.

Installation

To install Engram in your OpenClaw environment, execute the following command: clawhub install openclaw/skills/skills/dannydvm/engram Ensure you have Ollama installed and running with the nomic-embed-text model prior to initialization. The service runs as a local server at http://localhost:3400, with a dedicated web dashboard available at /dashboard for managing and auditing your memory nodes.

Use Cases

Engram is perfect for long-term project management, client relationship maintenance, and complex technical workflows. Use it to store client preferences to avoid repeating onboarding questions, maintain a ledger of project-specific architectural decisions (e.g., database selection logic), track milestones to prevent scope creep, and link related events to provide your agent with a deeper understanding of historical context. By feeding session updates into Engram, you ensure your agent evolves alongside your project.

Example Prompts

1. "Engram, recall any previous decisions we made regarding the authentication flow for the current client project."
2. "Extract all facts and preferences from this meeting transcript and save them to my memory store."
3. "List all events related to the Q1 deployment milestone to help me prepare for our status report."

Tips & Limitations

To maximize performance, always execute engram search at the start of a session to prime the agent's context window. Use engram auto-relate periodically to ensure your memory graph is well-connected, as highly connected nodes receive higher priority in recall queries. Note that Engram relies on a local vector database; performance depends on your available disk I/O and RAM. Ensure your memory is pruned regularly if handling millions of entries, as semantic search overhead can scale with data volume.