Skip to content

Agent Blocks

An Agent Block is a separate block type from Action Blocks. You will find it under the Agent tab in the Intake Bot editor.

Collect and Instruct are no longer two products. They are one Agent Block. You choose a conversation style, define typed outputs, and attach documents or decision tables.

New Agent Blocks require Business or Enterprise. Blocks you already published on Pro keep serving. You cannot author new ones on Pro.

Structured. The block walks the outputs in order. Use this when you know the fields and want a predictable path.

Agent-led. The block can follow the thread, as long as it still lands the typed outputs. Use this when the lead’s wording should shape the next question.

If the block has no outputs, it is an advisor. The style choice stays hidden, because there is nothing to collect. It answers from the sources you attach and declines when those sources do not cover the question.

Outputs use the same field types, validation, and required rules as ordinary questions. Collected values are saved on the submission. Records, routing, exports, PDFs, and action blocks can use them like any other field.

An objective can reference earlier Intake Bot fields that are not private. The editor inserts the variable. Web, test, and channel conversations resolve it from the submission so far.

Attach workspace Knowledge documents, a collection, or inline Markdown. The block cites the passage it used. See Knowledge.

Decision tables ground collection in Record Type rows. Ask columns are questions. Derive columns fill themselves once the answers pin a single row. If the match is still ambiguous, the block asks instead of guessing. Large option sets render as search, so the lead never types an invalid choice.

A credit budget and a turn cap keep the conversation from running away. If AI or credits are unavailable, the block falls back to plain questions so the submission can still finish.

You can place an Agent Block inside a conditional branch at any depth.

Autonomous pursues an objective with a capped loop: understand, a code policy, then one action and one compose, with a hard cap of three model calls. It is in beta. The add-menu card is off on live plans. Do not treat it as generally available.

FreeProBusinessEnterprise
Author new Agent BlocksNoNoYesYes
Already published Pro blocksKeep servingKeep servingKeep servingKeep serving
  • Knowledge: The shared library Agent Blocks read from.
  • Action Blocks: Fire an integration when a lead reaches a step or finishes.
  • Intake Bots: Build the rest of the conversation.