Getting Started
Architecture at a Glance
MemorySync sits between your application and model as a scoped context service. Your server writes useful facts, retrieves relevant memory, and decides how to use the results.
The application flow
- 1Your server authenticates with an API key and identifies the project and end user.
- 2The server writes a durable fact or event through the memory API.
- 3Before a later model request, the server queries with the user’s current intent.
- 4MemorySync returns relevant records from the same request scope.
- 5Your application selects the useful results and adds them to the model prompt.
Responsibilities by system
| System | Responsibility |
|---|---|
| Your application | Authorization, stable external identifiers, write policy, prompt construction, and final user experience. |
| MemorySync | Scoped memory operations, relevance-based retrieval, metadata filters, and lifecycle APIs. |
| Your model provider | Generate or classify using the context your application sends. |
| Source systems | Remain authoritative for documents, permissions, and transactional records. |
Scope every request
X-API-Keyauthenticates the server-side caller.X-Project-IDidentifies the application or workload.X-End-User-IDidentifies the person represented by a service request.- Use stable opaque identifiers; do not place email addresses or secrets in scope headers.
What crosses the boundary
Send the minimum context needed for the memory use case. A memory can include text, a source label, and application metadata. Query responses return candidate records for your application to inspect and use.
Go deeper
Was this page helpful?