Skip to content

Agent Blocks

An Agent Block talks with the lead between your questions. It answers from sources you attach, or it works toward an objective you set. Either way it ends in an outcome, and later steps route on that outcome.

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

New Agent Blocks require Business or Enterprise. Blocks you already published on Pro keep serving, except Collector blocks (see Collector blocks were removed). You cannot author new ones on Pro.

Advisor answers the lead’s questions from the sources you attach. When the sources don’t cover a question, it says so instead of guessing.

Autonomous pursues an objective you write. It holds the conversation until it reaches an outcome, and it can ask short Discovery questions along the way. It is in beta. See Autonomous (beta).

An objective can reference earlier Intake Bot fields that are not private. The editor inserts the variable, and the conversation fills it from the submission so far.

An Agent Block does not collect typed fields. When you need a value in a fixed shape (a date, a phone number, one option from a list), add an ordinary question after the block. You get the usual field types, validation, and required rules, at lower cost and with predictable answers. Route on the block’s outcome to pick which questions come next.

Advisors can keep relevant details the lead volunteers, without asking extra questions. Autonomous blocks can also ask a few short probes. These Discovery answers show under the Agent Block in submission details. They are self-reported context: Submission Intelligence, exports, webhooks, and Action Block mappings can use them, but routing does not read them directly.

Attach workspace Knowledge documents, a collection, or inline Markdown. See Knowledge.

A reference table attaches Record Type rows. Each question queries the table by its category values, so the bot gets every matching row and how many there are.

Before an Agent Block sends a reply, each sentence is checked against the Knowledge it read and what the lead said. A store, product, location, time, price, or contact detail that isn’t there, or that belongs to another place, is removed. The bot says it can’t find that information instead.

If the first search misses the question, the bot searches once more by the kind of place that would hold the answer, then replies.

In submission details and in the Agent Block test, a check mark under each reply shows whether it was verified. Evidence lists what the reply states, each with the line from your Knowledge or the table row behind it, and which document and section it came from.

You can place an Agent Block inside a conditional branch at any depth. Its outcome works like any other answer: later conditions, routing, Records, and Action Blocks can use it. Submission details show the outcome each Agent Block ended with.

Autonomous treats the objective as the success condition. Each turn runs in a fixed order: understand the message, apply a code policy, take one action, compose one reply. A turn makes at most four model calls. Custom outcome values are checked against the list you defined.

You can start an Autonomous block from a catalog profile. The profile’s objective, skills, outcomes, and guidance are copied onto the block. Later catalog edits don’t change blocks that are already live.

Autonomous is available on Business and Enterprise, in beta.

The Collector type is gone, along with typed outputs, decision tables, Ask and Derive columns, and the guided search picker. Existing Collector blocks no longer appear in the editor and no longer run in live flows. Answers they already collected stay in the submission data.

To rebuild one, add the same fields as ordinary questions, after an Advisor or Autonomous block if you still want the conversation in front of them.

FreeProBusinessEnterprise
Author new Agent BlocksNoNoYesYes
AutonomousNoNoBetaBeta
Already published Pro blocks (not Collector)Keep 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.