Documentation
Attelium is a memory layer for AI agents. It exposes an MCP server over a folder of plain-text KGL files, so memory is something you can read, diff and keep in git.
Quickstart
Install, initialise a project, restart your agent, commit your first memories.
How it works
The folder layout, node and link types, and how writes, recall and entity resolution fit together.
Tools reference
All 16 MCP tools, which six are in the core profile, and when to use each.
Search & recall
Hybrid search, the two embedders, and what get_context assembles.
Viewer
The read-only dashboard: graph view, activity feed, sessions, lint and the agent's search log.
Maintenance
Duplicate scans, reflections, reindexing and per-project setup.
Packages and names
| Name | What it is |
|---|---|
| Attelium | The product and this site. |
kgl-memory | The Python package: store, search, tools and MCP server. Console scripts: kgl-memory-init, kgl-memory-mcp, kgl-memory-viewer, kgl-memory-maintain. |
kgl-lang | The KGL parser, linter, OpenCypher query engine and writer. Imported as kgl. |
| KGL | The plain-text knowledge-graph file format. Format docs. |
| Graxeon | A workspace for editing KGL graphs. graxeon.com. |
Status: alpha. Requires Python 3.10+. The tools and layout are tested and in daily use by the author, and may still change between releases. See FAQ & limits.