WEBVTT

overview-profile
00:00:04.600 --> 00:00:15.352
An OpenPond Agent is a reusable capability that lives in your Profile. Its source defines its purpose, the actions people can run, and the Evals that protect its behavior.

overview-entrypoints
00:00:16.452 --> 00:00:26.052
Use chat when the request is open ended and conversational. Use a direct action when the inputs and result should be explicit, repeatable, and easy to connect to another workflow.

overview-surfaces
00:00:27.152 --> 00:00:37.506
A shared action catalog lets the Agent appear in OpenPond chat, its Agent page, supported channels, and API or M C P integrations without duplicating the underlying behavior.

overview-example
00:00:38.606 --> 00:00:50.073
An Account Health Agent can answer follow up questions in chat, summarize one account, triage renewal risk, or build a weekly portfolio review. Each path uses the same trusted Agent source.

overview-evals
00:00:51.173 --> 00:01:02.462
Evals run repeatable scenarios against the Agent's behavior and outputs. During Improve, OpenPond compares the candidate with the active Agent and only releases the change after the required Evals pass.

overview-lifecycle
00:01:03.562 --> 00:01:13.787
Start from a purpose or supporting chats, review what OpenPond will build, use the Agent on real work, then improve it from the evidence. Next, create the Account Health Agent step by step.