A self-hosted Claude agent that actually remembers — verbatim memory palace at zero retrieval cost, Discord + web UI, and 90%-cheaper repeat calls via prompt caching.
// readme
Galadriel
The open engine behind Aedelgard — a persistent AI agent with sovereign memory, a model-agnostic brain, and the ability to improve its own code.
Separate the mind from the brain power. The mind — memory, identity, values — is yours, local, portable, and model-agnostic. The brain — the LLM that does the thinking — is a rented commodity you can swap at any time. Galadriel is the engine that proves this thesis in code.
What this is
A persistent AI agent harness that runs on your own machine. It connects to any LLM provider (Claude, Gemini, OpenAI, Bedrock, Nebius/DeepSeek, xAI/Grok, Mistral), maintains a local verbatim memory palace with zero-token retrieval, and can edit its own code to improve how it works.
This is not a product. It is the open engine that Aedelgard packages into a one-click desktop app. Everything here is real, inspectable, and yours to build.
What’s in the box
| Component | What it does |
|---|---|
| 🏛️ Memory Palace | Verbatim semantic search + temporal knowledge graph. Local ChromaDB + SQLite. Zero API cost to recall anything. |
| 🧠 Provider Seam | One… |