Provider Routing

Eden AI sits in front of several underlying AI engines and forwards a request to whichever one is configured for a given task, such as text extraction or translation. A Beam agent reads a task queued in another system, formats it as a request Eden AI expects, and sends it through following the customer's approved provider preference for that task type, then reads back the result to write into the workflow. Requests that fail against every configured provider, or return a result outside expected bounds, are routed to a person rather than retried indefinitely.

Unified Output Handling

Because Eden AI normalizes responses from different engines into one format, a Beam agent can read the result the same way no matter which provider actually handled the request. The agent takes that normalized output and writes it to the field or record in another system that the customer's rule specifies, such as a transcript field or a translated text column, keeping the workflow the same even if the underlying provider changes later. Outputs missing expected fields, which can happen if a provider partially fails, are held for a person to inspect.

Usage and Cost Monitoring

Eden AI records how many requests were made and against which engine, information a customer needs to keep costs predictable. A Beam agent reads that usage data on a schedule and writes a summary to another system, such as a spend report, following the frequency the customer has agreed to, and applies an approved threshold rule to flag unusually high usage. Usage that crosses the threshold, or a sudden shift toward a costlier provider, is flagged to a person rather than left to keep running unchecked.