Migration notes

🧪 Beta — Available in PR #681

Wave 1 is mostly invisible — it fixes a bug rather than introducing a new surface. But on existing chatbots, a few state mismatches may have accumulated. This page is a checklist to clean them up.

In this guide:

  • Verify each chatbot’s saved model
  • Verify creativity slider matches actual temperature
  • Re-save Playground settings if drift exists
  • Watch for weirdness in the first week

Verify each chatbot’s saved model

For every chatbot you’ve used the Playground with:

  1. Open the chatbot detail page.
  2. Note the model shown on the General / Settings page (the backend model).
  3. Open the Playground.
  4. Note the model in the Playground settings (the localStorage model — pre-Wave-1 may differ).

If they differ, decide which is correct:

  • Playground model is what you want → click Save to agent in the Playground. Now both align via the new save path.
  • Backend model is correct → in the Playground, switch model to match backend, save (or just clear localStorage).

Verify creativity slider matches temperature

Same idea for temperature:

  • Backend stores a numeric temperature (0.0–2.0).
  • Settings page renders the Creativity slider with a 3-step mapping (Precise/Balanced/Creative → 0.1/0.3/0.8).

Before Wave 1, manually-set temperatures (e.g., 0.5) might not have an exact slider position. After Wave 1:

  • Re-saving from the slider snaps to the canonical 0.1/0.3/0.8 values.
  • Re-saving from the Playground (where you can set arbitrary numeric temperature) preserves the exact value.

If you’d intentionally set a non-standard temperature, save from the Playground, not the slider.

Re-save Playground settings if drift exists

Quickest cleanup pass for an existing chatbot:

  1. Open Playground.
  2. Send a test message — verify the response quality is acceptable.
  3. Click Save to agent.
  4. Confirm in DevTools → Network that a PUT /api/chatbots/{id} request fires (the new behavior).

This forces the backend to align with the Playground, so future drift is impossible.

What’s safe to skip

If a chatbot is configured entirely from the Settings UI (no Playground use), it’s already in sync — Wave 1 doesn’t change Settings-page save behavior.

If a chatbot is on a default model and you’ve never customized, no action needed.

Watch for weirdness in the first week

After Wave 1 lands, watch your chatbots’ analytics:

  • Latency: any sudden change to model selection alters latency. Check Performance → latency charts.
  • CSAT: if a wrongly-saved model was actually performing better than the one you intended, you might see a small CSAT shift. Adjust if needed.
  • Improve rate: same — flagged replies may shift up or down.

Most users won’t see anything; the bug was usually hidden by the fact that defaults happened to match across both places.

Rollback

Wave 1 is forward-only — no flag to disable. If a specific chatbot needs to revert to a prior model:

  • Open the Playground.
  • Pick the old model.
  • Click Save to agent.

This uses the same new save path; you’re not rolling back Wave 1 itself, just choosing different settings.

Reporting issues

If you see anything that looks like a Wave 1 regression:

  • The Playground saving a different model than what you picked.
  • Creativity slider snapping to unexpected values.
  • Live conversations using a different model than the chatbot’s saved one.

Contact support with the chatbot ID, the timestamp, and what you observed. Wave 1 is in beta — these reports help us land Wave 2 cleanly.

What’s next

You’ve reached the end of the Guide. Browse top-level for anything you missed.