Context & imports
Bring your own evidence, keep its provenance, and decide what goes into a scenario.
Your context cache
Context holds the records collected by integrations and imported by you. Records include source links, timestamps and expiry so they can be reviewed before analysis.
The cache is local to your browser, with session-memory fallback if storage is unavailable. It is not a shared account workspace or a durable server-side archive. Remove individual records or clear the cache in Context. Automatic sources can refill it on their next update.
Import format
Import a JSON array or an object with a records array. Every record needs the fields below. Use actual source evidence and current ISO timestamps; expired records are excluded from scenarios.
| Required fields | Meaning |
|---|---|
id, kind, title, summary | Record identity, category and description. |
source, sourceUrl | Attribution and an HTTPS source link. |
observedAt, retrievedAt, expiresAt | ISO timestamps for observation, retrieval and expiry. |
assets, entities | Arrays describing associated assets and entities. |
Use the repository’s context protocol reference ↗ when building an adapter.
Optional locations and relationships
A location includes lat, lon, label, basis and an HTTPS evidenceUrl. Basis can be reported, mention or user-supplied. All imported records are labeled user-supplied, including their locations.
Optional relationships contain a subject, predicate, object and evidence URL. A place mentioned in an article is not automatically the event location. Do not infer a wallet’s location from its address.
Import and storage limits
- Up to 150 records per import or request.
- Import files up to 2 MB.
- Up to 500 cached records and 3 million serialized characters per browser.
- Up to 30 fresh records in an OpenAI scenario.
Keep credentials out of record text and source URLs. Imported text is evidence to review, not instructions for the application or model.