Data: entities, records and record triggers
A typed data model your agents work on, with checked writes, triggers and webhooks.
entities:
Ticket:
fields:
subject: { type: string, required: true }
priority: { type: enum, values: [low, normal, urgent], default: normal, indexed: true }
lease: { type: ref, entity: Lease }
requester: { type: end_user } # set by the runtime: whose record it is
opened_at: { type: datetime, indexed: true }
Field types: string text integer number boolean date datetime enum ref end_user json. Every
write is checked against them, whoever writes it. Unknown fields are refused, types are normalized,
and refs must point at a record the writer can see.
Agents get a data connection with data__list_ticket, get_, create_, update_
and delete_ tools, limited to the access each workspace declares (data: { Ticket: write }).
Creates are idempotent, so a replay after a crash finds the record it made. Filters accept relative
dates like today+30d. All filtering and sorting runs in SQL, and indexed: true fields get a
database index.
Record triggers start a run when a record changes:
triggers:
- { kind: Record, name: Urgent tickets, entity: Ticket, on: [created, updated],
rule: { conditions: [{ field: priority, op: "==", value: urgent }] } }
Each change fires each matching trigger once. A workspace's own runs don't re-trigger it, so it can't feed itself.
Your backend uses /api/v1/solutions/{name}/records/{Entity}: GET with
?where.status=open&filter=opened_at<=now-2h, POST with an Idempotency-Key, PATCH
with If-Match, DELETE, and POST …/bulk to delete or update many records at once (by id,
or every match of a filter, up to 5,000). It gets signed webhooks for the changes it subscribes to.
The solution's Data tab lists records with columns you choose, edits them, deletes or updates many at once (tick them, or select all matching), and opens a ref field's record, which lists the records that point at it.