One bot. One question. Your first improvement.
Connect the provider, investigate a saved bot and complete a reviewed repair. Verify the live workflow before expanding autonomy.
Start small. Complete the loop.

Prepare your workspace.
Sonneto is available to permitted workspace owners and admins. Provider configuration lives in Platform Admin; workspace enablement and agent controls live in Sonneto.
Connect OpenAI
Add an enabled OpenAI text profile under Platform Admin → AI providers. Use a supported scope such as Agent drafts or All workflows and run the connection or generation test. Store keys in provider settings, never in chat.
Check access and allowance
Enable Sonneto for the workspace. Confirm the provider and relevant AI allowance; subscription access and provider consumption are separate considerations.
Select a saved bot
Save canvas changes first. Choose a clear journey and verify outcome tracking if you plan to measure performance.
Complete the review loop.
Begin with a narrow change whose intended behavior you understand. Review the generated candidate using your real configured provider, then verify the published experience before enabling more independence.
Investigate and prepare
Ask a scoped question and inspect its evidence. In the agent workspace, select AI wording & connection repair and explain the goal.
Approve and publish
Compare the wording or connection diff and scenario results. Apply the exact candidate to its draft and use the separate publication action. Check the live widget afterward.
Monitor deliberately
Set an objective and recurring diagnosis. Check task history for actual runs before relying on the cadence. Enable automatic permissions only after the manual workflow succeeds.
How it works.
- 01
Connect and investigate
Configure the provider, enable Sonneto and ask about a saved bot.
- 02
Review and release
Prepare, test, approve and manually publish the candidate.
- 03
Measure and expand
Inspect monitoring and learning proposals before selecting narrow automatic actions.
From question to next action.
- 01Goal
- Improve an unclear introductory question in a support bot.
- 02Evidence
- Illustrative starting point: the saved prompt asks for several things without explaining the next step.
- 03Action
- Request clearer wording that preserves qualification requirements, then compare scenarios and approve.
- 04Follow-through
- Publish manually, test the live widget and inspect the release report. Keep future releases manual until provider and maintenance paths are verified.
What to expect.
| Scope | Details |
|---|---|
| Provider failure | Check enabled profile, workflow scope, key test and allowance. Do not paste credentials into chat. |
| Stale approval | Inspect the current draft and prepare or review a candidate against that version. |
| Missing scheduled work | Check the maintenance task and task errors. A saved schedule does not prove execution. |
Access depends on permissions, provider configuration and applicable AI allowance. Check current plans and pricing.

Start with a question.
Build toward better outcomes.
One bot. One clear goal. A teammate to help you move forward.
Compare plans and AI allowancesWithin your boundaries.Review first.
Expand when you’re ready.
