How to chat with conversational agents
Multi-turn assistants embedded in product screens, with platform-managed memory — rolling out with the V1 Beta.
Conversational agents hold a stateful, multi-turn session: you ask in natural language, answers stream in as they are produced, and the platform manages the conversation memory the agent declares (a sliding window or a rolling summary).
Beta
Conversational mode is rolling out with the V1 Beta. The first pilot is the DataHub assistant; the Sessions area of the Console is visible today but disabled until Beta.
Starting a session
- Open the embedded assistant panel in the product screen — for example the chat in the DataHub Business Rule editor.
- A session is created for you. Type your request in plain language: draft, explain, validate, or test.
- Follow the streamed answer. Each turn is recorded as a run with its own trace.
Actions ask before they touch anything
When the assistant needs to act on a live environment — for example evaluating a rule against the DataHub engine — it shows a confirmation card first and waits for your explicit approval. Tool policies (allow / ask / deny) apply in chat exactly as they do in task mode.
Session history
Sessions survive infrastructure restarts: hot state is kept while the conversation is live, and the history is archived afterwards so a session can be reviewed later.
Still stuck?
Tell us what you were trying to do and where it failed — screenshots and run IDs help. A human follows up.