Your vault for model context
Ghost Vault is a user-held local memory replica for LLMs: hybrid retrieval (pgvector, full-text, weighted fusion), optional encryption at rest, a small REST and OpenAPI surface for tools like ChatGPT and Gemini, and an MCP server (gvmcp) for editors and agents. It pairs with the GitS ecosystem without replacing your host model.
Claude custom connector (Anthropic)
Use Ghost Vault from claude.ai, Claude Desktop, and the mobile app with one hosted remote MCP URL. Register https://…/mcp-<token>/ under Customize → Connectors; Anthropic’s cloud calls your public HTTPS edge. The walkthrough below matches the UI paths in the tutorial.
What you get
A focused datastore and API boundary between your data and the models that use it. For deep docs and the CLI, see the Ghost Vault repository on GitHub. News and long-form write-ups live on the shared GitS blog.
Hybrid retrieval
Vector search, full-text, and a weighted gated fusion path so answers stay grounded in what you actually stored.
Encryption
Optional encryption at rest—enable once; the mode is fixed per database so expectations stay clear.
OpenAPI & MCP
Wire ChatGPT Actions, Gemini tools, or gvmcp in Cursor and other MCP clients with the same backend.
gvctl & ops
CLI for unlock, init, and common REST flows; Docker and local go run paths for your environment.
Blog & contact
Shared with GitS: one blog and one contact path for the product family.
Open source
Issues and contributions welcome; follow the repo for releases and integration notes.