What's changing in Wave 1

Until now, Playground and production lived in slightly different worlds. You could try gpt-4o in the Playground, hit Save to agent, and find that production conversations were still using the old model. Wave 1 closes that gap.

In this guide:

• The bug Wave 1 fixes
• New: gpt-4o in the model list
• Creativity → temperature mapping
• “Save to agent” round-trip
• What changes for end-users

The bug Wave 1 fixes

Playground “Save to agent” used to write to your browser’s localStorage only. The backend record for the agent kept the old model and temperature, so live conversations on every channel used the previous settings.

Symptom: you’d test gpt-4o happily in the Playground, save, and customers continued chatting with gpt-4o-mini.

After Wave 1, Save to agent:

1. PUTs the new model + temperature to the backend (PUT /chatbots/{id}).
2. Then writes to localStorage as a Playground convenience.

Live conversations pick up the new settings immediately on the next message.

New: gpt-4o in the model list

gpt-4o is the backend’s default for new chatbots. Until Wave 1 it wasn’t visible in the frontend’s AVAILABLE_MODELS list, which caused wizard validation to reject newly created chatbots.

gpt-4o is now a first-class option in the Playground model selector and the chatbot wizard.

Creativity → temperature mapping

The chatbot settings page exposes a 3-step Creativity slider:

Behind the scenes, the slider value (0/1/2) maps to a temperature (0.1/0.3/0.8) sent to the backend. Direct numeric control is still available in the Playground if you need a precise value.

“Save to agent” round-trip — what’s new

Before Wave 1:

Playground → localStorage only
            (backend never updated)

After Wave 1:

Playground → PUT /chatbots/{id} → backend
            ↓ on success
            localStorage cache

Same for the Compare → Save to main flow, which now also persists to the backend before mirroring to localStorage.

Initialize Playground from backend

When you open the Playground for the first time on a new device (no localStorage cache), the Playground now hydrates from the backend — same model and temperature as production. No more drift between devices.

What end-users see

• No visible change in chat quality from the bug fix itself, except that whatever you saved actually applies now.
• gpt-4o-trained bots will use gpt-4o consistently across channels.
• Creativity-slider changes propagate to live chats on the next message.

What’s coming in later waves

Wave 1 is foundational. Future waves are scoped but not yet shipped:

• Multi-step reasoning improvements.
• Better tool-call planning when multiple AI Actions are available.
• Configurable per-action retry policies.
• Inline reasoning traces in the agent dashboard for debugging hard cases.

Watch the What’s new page for landings.

Beta caveats

• Available to all users; no opt-in.
• Behavior may evolve — read this section after each release for changes.
• Open issues / surprises: check the in-app changelog or contact support.

What’s next