1. Sign in and select a workspace
Open the workspace and sign in with an email code. An enrolled pilot identity can create a workspace; existing members can select their authorized workspace. If access has not been granted, ask your pilot administrator. Signing in does not enroll you.
2. Create a namespace
Create a namespace such as Project notes. Give collaborators only the capabilities they need. Ordinary agents should usually receive read and observe; verification, correction, deletion and administration are separate capabilities.
3. Record evidence and a candidate
Add a source locator and the exact supporting excerpt. Save the observation as a candidate and choose when it began to apply. Keep verification separate until an authorized reviewer can attest to the exact text and its evidence.
4. Read it back
Open the saved memory directly to inspect its source and time intervals. Recall may report pending indexing after a write. When recall is pending, choose “Search without waiting for indexing” to request bounded retrieval without the saved-write indexing guarantee. If hybrid search is unavailable, retrieval uses a lexical slice. Keep its degraded or partial status; these results do not prove that no matching memory exists.
5. Try a historical view
Set both time controls in the workspace. After an authorized correction, compare the earlier known-at cutoff with the current one while holding as-of fixed. The temporal guide includes the exact Boston/New York fixture.
6. Connect code or an agent
Use REST for application calls or the MCP resource for a compatible client. Generate requests from the published catalog rather than guessing payloads. Every mutation needs an Idempotency-Key. Tokens are credentials, not permission to choose an arbitrary workspace.