The usual way to give an agent access to an application adds a layer: custom endpoints, logic rewritten so a model can follow it. Two issues he opened on the RESTHeart repo, #615 and #616, skip that layer entirely. The agent discovers what it can do by reading the schema the API already enforces. A write gets validated against that schema before it reaches the database. A business rule, a warehouse shortage calculation, stays a MongoDB aggregation written by whoever understands the domain; the agent calls it directly.
The permissions never change: the ACLs already governing every other client apply to the agent too.
Top comments (0)