Skip to main content

​
What is the Encrypted Memory Vault?

The Encrypted Memory Vault is Rainy MaTE’s local, privacy-first storage system. It ensures that the knowledge your agents gather and index from your workspaces remains entirely secure and entirely under your control. When your agent learns facts about your projects, reads your files, or stores snippets of code for later retrieval, this information is never stored in plaintext.

​
Key Features

  • AES-256-GCM Encryption: All memory content and metadata are encrypted at rest using military-grade AES-256-GCM encryption.
  • Keychain Backed: The master encryption key is securely generated and stored in your operating system’s native keychain (e.g., macOS Keychain). Rainy MaTE does not hardcode or transmit this key.
  • Vector Embeddings (3072 Dimensions): Memories are vectorized using Gemini (gemini-embedding-001) or OpenAI embeddings. This allows the agent to perform semantic searches (finding memories based on meaning rather than exact keyword matches) instantly.
  • Local-First Database: The encrypted vault relies on a local Turso/libSQL database (rainy_cowork_v2.db), meaning your memories never leave your machine unless you explicitly configure a remote sync.

​
How it works

When the agent decides to “remember” a piece of information:
  1. The text is sent to your selected Embedding Provider (e.g., Gemini) to convert the content into a 3072-dimensional vector.
  2. The original text content is encrypted locally using the master key retrieved from your system keychain.
  3. The encrypted text, along with the semantic vector and access metadata, is stored in the local SQLite database.
When the agent needs to “recall” information:
  1. The agent’s query is vectorized.
  2. The vault performs a lightning-fast cosine similarity search across the vectors to find the most relevant memories.
  3. The matching memories are decrypted locally and provided to the agent as context.

​
Managing your Embeddings

By default, Rainy MaTE uses Gemini for generating embeddings since it provides a generous free tier with high-quality 3072-dimension vectors.
  • You can configure your Embedding Provider (Gemini or OpenAI) in the Settings menu.
  • Note: Changing the provider will affect how new memories are vectorized. Older memories will retain their original vectors.