How It Works
How MemorySync Works
MemorySync gives your server a simple loop: identify the workload and end user, save useful context, retrieve relevant memories later, and decide how your application will use them.
The core loop
- 1Identify
Your trusted server selects the project and opaque end-user ID.
- 2Save
Add a durable fact, preference, decision, event, or outcome.
- 3Retrieve
Ask for the context a later task needs.
- 4Inspect
Receive candidates from the same authorized scope.
- 5Use
Select and delimit useful context for your model or product.
Who is responsible for what
Your application
Owns product policy and access.
- Authorize users.
- Choose stable identifiers.
- Decide what to save and how results are used.
MemorySync
Provides the public memory contract.
- Accept scoped operations.
- Return documented customer-visible outcomes.
- Keep private implementation outside the contract.
Model provider
Processes what your application sends.
- Generate or classify from supplied instructions.
- Does not become your source of truth.
Source systems
Remain authoritative.
- Own permissions and records.
- Provide the current document or transaction state.
Three values select the memory space
| Value | Purpose | Safe default |
|---|---|---|
| API key | Authenticates your trusted server. | Load it from a server-side secret store. |
| Project ID | Selects the application or workload. | Keep separate access boundaries in separate projects. |
| End-user ID | Selects the person your server represents. | Use a stable opaque ID, not an email address or token. |
What your application can rely on
- Add returns either a created memory or a successful skipped outcome.
- Query returns a
memoriesarray; an empty array is a valid result. - Bulk add reports
created,skipped, orrejectedfor each submitted item. - Connected-source sync exposes a job resource that your application can inspect.
- Exact storage, ranking, caching, scheduling, and processing internals are not part of the public contract.