Midas — Privacy Policy

Last updated: 2026-06-08

Midas is a local-first memory connector. It is designed so that your data never leaves your own computer. This policy describes exactly what Midas does and does not do with your data.

What data Midas handles

Midas stores the text you (or Claude on your behalf) choose to remember — facts, decisions, preferences, constraints, and conversation turns — together with derived metadata (a vector embedding, an importance score, timestamps, a kind label, and a session id).

Where it is stored

All memories are stored locally on your machine in a single SQLite file that you control (default: ~/.midas/memory.db, configurable). Midas does not operate any server, account, or cloud storage, and there is no telemetry or analytics.

Data collection

The Midas developer collects no data. We do not receive, see, or have access to your memories, prompts, embeddings, or usage. There is no sign-up and no API key.

Network usage

• No memory data is ever transmitted over the network. Capture, recall, forgetting, and context assembly all run entirely offline using local computation.
• The only outbound network activity is infrastructure, not your data:
  • One-time download of the open-source embedding model (BGE ONNX, from Hugging Face) the first time the local embedding backend is used. Select the hashing backend to avoid this entirely.
  • Package installation via uvx/pip from PyPI when the connector is first run. Neither of these sends any of your memories or prompts.

Third-party sharing

None. Midas does not share data with any third party because it never collects or transmits your data in the first place.

Data retention and deletion

• You control retention. Memories persist in your local SQLite file until you remove them.
• Midas can bound storage automatically by forgetting the lowest-value memories (no LLM involved).
• You can delete a single memory (forget), erase everything about a topic (forget_matching — dry-run preview first, and the deleted records are returned as an audit trail), run a retention/audit pass (maintain), or erase everything (forget_all). Deleting the SQLite file removes all stored memories permanently.

Contact

Questions or security reports: open an issue at https://github.com/vornicx/Midas/issues or contact the maintainer via the GitHub profile at https://github.com/vornicx.