Hypabase Memory

What This Skill Does

Hypabase Memory provides a sophisticated, persistent knowledge graph layer for OpenClaw agents, enabling them to move beyond stateless interactions. By utilizing the PENMAN notation system, it stores user preferences, project facts, task assignments, and episodic events as a connected network of atoms. This allows the agent to recall context based on who was involved, what occurred, when it happened, or why a decision was made, creating a cumulative learning experience that deepens over time.

Installation

Add the MCP server to your OpenClaw configuration file located at ~/.openclaw/openclaw.json. Ensure you have uv installed on your system to manage Python dependencies automatically:

{
  "mcpServers": {
    "hypabase-memory": {
      "command": "uvx",
      "args": ["--from", "hypabase", "hypabase-memory"],
      "env": { "HYPABASE_DB_PATH": "hypabase.db" }
    }
  }
}

## Use Cases
This skill is ideal for long-term project management, personalized assistance, and complex workflow tracking. Use Hypabase to store user preferences (e.g., "Alice prefers Python over Java"), track the history of delegated tasks (e.g., "Bob was assigned the billing update"), or maintain a map of technical constraints. It is particularly effective for teams where context needs to be shared across multiple sessions or different project phases. By maintaining a persistent graph, the agent can reason about relationships between disparate pieces of information, such as linking a specific meeting outcome to a future project deadline.

## Example Prompts
1. "Store in my memory that I prefer using VS Code for my Python projects and I usually work on them during the weekends."
2. "Review my recent memory logs: who was assigned the database migration task and what was the agreed-upon deadline?"
3. "Can you recall the procedure we discussed for our deployment process and verify if the latest updates align with that plan?"

## Tips & Limitations
To maximize the utility of your knowledge graph, maintain consistency in your entity naming conventions. Use full, descriptive terms like "machine learning" instead of abbreviations like "ML" to avoid fragmented graphs. Remember that you must call `consolidate()` periodically to merge similar entities that might have been entered with slightly different names. Because this skill relies on the PENMAN notation, ensure that each memory atom is self-contained. If a sentence involves multiple actions, break them down into separate atoms; sharing common entities will automatically weave these disparate facts into a coherent graph, making future recall significantly more powerful and accurate.