A new block kind gathers multiple typed fields through a goal-driven conversation instead of a fixed question sequence. Attach markdown context or Records-backed decision tables as knowledge sources, and let the agent handle the back-and-forth.
Fixed question sequences work fine for simple bots. They start to feel awkward when the path depends on a combination of answers, or when you need to collect something open-ended that takes a few rounds to pin down. The new Collect Mode agent block handles that kind of intake.
Instead of asking questions one by one, a Collect Mode block runs a goal-driven conversation. You define the fields you want, attach knowledge sources (markdown context or Records-backed decision tables), and the agent figures out how to gather them. The conversation adapts to what the respondent has already said rather than marching through a fixed list.
The outputs are ordinary submission fields. Process boards, records, and action blocks pick them up the same way they would any other field.
For cases where you need the respondent to pick from a valid set of combinations, you can back the agent with a Records decision table. The agent enforces valid combinations server-side once the respondent’s answers pin a row. Large option sets use search-as-you-type so respondents never type something invalid.
If AI is unavailable or credits run out, the block falls back to static questions. A credit budget and turn cap prevent the conversation from running away. The authoring UI shows each slot’s resolved behavior — free text or constrained by a table — as you build, so there are no surprises at runtime.
Available on Pro plans and above.
A new intake bot block kind gathers multiple typed output fields through a goal-driven, multi-turn conversation instead of a fixed question sequence. Builders attach markdown context and Records-backed decision tables as knowledge sources; enforced tables validate final slot combinations deterministically server-side. Outputs are ordinary fields written to submission data, so process boards, records, and action blocks consume them unchanged. Credit budget and turn caps mirror action block guardrails; AI outage falls back to static questions. Available on Pro plans and above.
