The thesis
Mneme treats documents as a memory you should be able to inspect. It indexes local files, retrieves across them, and uses an OpenAI-compatible endpoint only for the LLM-backed parts you choose to enable.
Your documents. A memory you can verify.
What is in the repository
| Surface | Purpose |
|---|---|
src/ | Standard RAG, Graph RAG, query decomposition, citations, queueing, metrics, quality, and security. |
tui/ | Streaming chat, slash commands, settings, file management, and directory watching. |
tests/ | Offline-safe unit, integration, and regression coverage. |
benchmarks/ | Retrieval quality data and evaluation surfaces. |
What the interface should make visible
- Local ownership of files, parsing, indexing, and retrieval.
- Source-linked answers with enough location information to inspect.
- Explicit endpoint and context boundaries when an LLM-backed feature is used.
- A developer workflow that starts with the CLI or terminal UI, not a cloud dashboard.
Open source
Mneme is released under the MIT License. Read the code, run the tests, or start with the GitHub repository ↗.