Docs

Agent Step

Published July 25, 2026

The Agent Step adds a conversational AI experience inside a quiz. It can use earlier answers, gather structured variables, present suggested replies, consult attached Knowledge, and route the visitor to the next step.

Availability and AI Credits

AI Automation, which includes Agent Steps and Knowledge, is available to eligible active or trialing Basic and Pro accounts when enabled for the account. Free accounts cannot publish or execute live Agent-enabled quizzes.

Testing and live Agent calls require AI Credits. Credits are not included automatically with Pro and are tracked separately from the account's monthly response allowance.

Open Dashboard > AI Usage & Credits to check your balance, review Agent usage by quiz, and add a one-time prepaid credit package when purchasing is available. You can also open this page from the AI Credits panel in Agent settings.

No credits remaining

Your Agent settings and Knowledge bindings remain saved, but Agent tests and live Agent responses pause until AI Credits are available.

Add an Agent Step

In the Build Flow, open the block palette and select Agent Step. You can reorder it like other blocks and continue editing its settings before publishing.

Instructions and conversation

Write clear instructions that define the Agent's role, tone, boundaries, and task. Tell it which information to collect, which variables to set, and when it should finish or route elsewhere.

The Agent can use quiz context collected before its step, including earlier answers, variables, and tags. Treat visitor-provided text as untrusted input: instructions should tell the Agent to follow its assigned role and not reveal private instructions or unrelated information.

Opening message

Add an optional opening message to greet the visitor before they send anything. The opening message is displayed without making an AI call, so it does not use AI Credits.

Keep the opening message focused on the Agent's purpose. Do not ask the Agent to invent a second greeting when a fixed opening message already handles the introduction.

Conversation limit

The conversation limit controls how many follow-up Agent responses can be generated after the opening message. The default is four. Once the limit is reached, another follow-up is not sent, so the Agent should be able to complete its task within the configured number of turns.

Knowledge Sources

Knowledge lets an Agent answer from material you provide. Open Knowledge Sources in Agent settings to upload a supported document and attach it to that Agent Step.

When website import is available for your account, you can also enter an authorized public URL. Confirm that you own the content or have permission to import it. Only sources that finish indexing can be attached and queried.

Attach only the sources needed for this Agent's purpose.
Review imported material for sensitive or outdated information.
Test representative questions after indexing completes.
Remove or detach a source when the Agent should no longer use it.

Knowledge retrieval contributes to AI Credit usage. A source that is unavailable or still processing should not be treated as authoritative by the Agent.

Variable outputs

Outputs map structured Agent results to quiz variables. Select Add output, choose or create a variable, and assign its type: string, number, boolean, or enum.

Turn on Required when the flow should wait for that value.
Add allowed values for an enum to keep downstream rules predictable.
Reference saved values in outcome copy and supported logic with the standard {{variable}} placeholder.

See Response Tags for more about placeholders.

Response UI

Enable the response controls the Agent may use, such as buttons, cards, or carousels. Add guidance for labels, order, and actions. Use only the controls needed for the experience so responses remain clear and predictable.

Exit conditions and routing

Exit conditions turn Agent outputs into routing decisions. Add conditions in priority order, bind each one to an output or variable, and select its destination. The first matching exit is used.

Always keep a suitable default path for cases where no exit matches. Advanced compound routing may require Pro even when the Agent itself is available to an eligible Basic account.

Model and response settings

Prompt settings let you choose an available model, adjust creativity, and limit response length. More capable models, longer input, longer output, and Knowledge retrieval can consume AI Credits more quickly. Preview realistic conversations after changing these settings.

Step-by-step

Add an Agent Step from the Flow Blocks palette.
Write instructions that define the Agent's role, boundaries, expected outputs, and finish condition.
Add an opening message and set a practical conversation limit.
Upload and attach only the Knowledge sources the Agent needs.
Declare any variables the rest of the quiz will use.
Configure response controls and exit routes, including a default path.
Check your balance under AI Usage & Credits, then preview representative and adversarial conversations before publishing.

Usage and response counting

Agent calls use AI Credits independently of monthly responses. A visitor can exchange messages with an Agent before a quiz response is counted. The quiz uses one monthly response when it accepts a consented lead or final completion; completing after an earlier accepted lead does not consume a second response.

If an Agent request cannot complete, the visitor remains at the Agent step and sees an availability message. The quiz should not silently advance to completion because an Agent call failed.