AI & data actions

This is the group that makes a flow smart. These actions bring AI into the conversation — to answer, to classify, to search your knowledge, and to pull structured data out of messy human messages. Most agents lean heavily on this group.

Agentic AI

Agentic AI turns the conversation over to an autonomous agent that can reason, use tools, search your knowledge, remember details, and decide what to do next — all on its own. It's the most capable block in Vivollo, and often the only AI action a flow needs.

Reach for it when you want the agent to handle a request end to end rather than follow a fixed script.

You point it at an assistant (which supplies the persona, model, and knowledge), and it takes over from there. Because it's covered in depth on its own page, head to Agentic AI for the full story and its toolkit.

Ask AI

Ask AI hands a single turn to an assistant for one focused response — without the full autonomous loop of Agentic AI.

Reach for it when you want exactly one AI reply at a specific point in a structured flow.

You choose which assistant answers. If that assistant can call tools, you can route the flow based on which tool it chose — letting a single Ask AI step branch the conversation by what the AI decided to do.

User Intent

User Intent uses AI to sort a customer's message into categories you define, then routes the flow to the matching path.

Reach for it when you want to branch on what the customer means, not on a fact you already hold.

The settings

• Intentions — the categories you're sorting into (up to 25). For each, you give:
  • A short label (e.g. track_order, refund, product_question).
  • An optional description that helps the AI understand what belongs in this category — the more specific, the more accurate.
  • The next step for the flow when this intent is detected.
• A fallback path for when no intent fits.

Good descriptions are the secret to accuracy: "Customer wants the status or location of an existing order" classifies far better than just track_order.

Reply From Documents

Reply From Documents searches your knowledge and uses AI to compose a natural answer from what it finds — citing its sources. This is the classic "answer from our help center" pattern in a single block.

Reach for it when you want grounded, sourced answers to factual questions.

The settings

• Which collection to search.
• Response style — how the answer should read: its tone (you can give custom voice instructions), its length (short, normal, or long), and its format (plain text or markdown).
• Query helpers — optional synonyms that map different phrasings to the same idea (e.g. refund, reimbursement, money back), improving what the search finds.
• Filters — narrow the search by document attributes, the same way the knowledge filters work.
• Routing for each outcome — send the flow different ways depending on what happened: a good answer was found, documents existed but none fit, the collection was empty, or the search failed. This lets you fall back gracefully — for example, handing off to a human when there's no good answer.

Search Documents

Search Documents finds matching items in a collection and returns them — without composing a written answer. Where Reply From Documents answers, this one retrieves.

Reach for it when you want the raw results to work with — to show as a carousel, to feed into a later step, or to check whether anything matches at all.

The settings

• Which collection to search.
• The query — what to look for.
• Filters — optional narrowing by attributes (price, stock, category, and so on).

The results become available to later steps in the flow.

Extract Data

Extract Data uses AI to pull structured fields out of a free-form message, according to a shape you define.

Reach for it when a customer says something messy and you need tidy data from it — turning "my order 10432 came smashed and I need it sorted today" into order_id: 10432, issue: damaged, urgency: high.

The settings

• The fields to extract — you define each field, its type, and whether it's required (this is the shape the AI fills in).
• Routing — one path when extraction succeeds, another when the message doesn't contain valid data for the required fields.

Whatever it extracts becomes variables your later steps can use.

Get Filter Vocabulary

Get Filter Vocabulary retrieves the filterable fields and their possible values from a collection — for example, the list of available brands, sizes, or price ranges.

Reach for it when an advanced flow needs to know what's filterable before building a precise search. In most flows you won't need it directly — the Agentic AI uses this kind of information behind the scenes to build accurate catalog filters on its own.

How they fit together

A few of these actions often appear in sequence. A grounded answer with a graceful exit looks like:

Reply From Documents
   found a good answer        → (continue / resolve)
   nothing relevant           → Send Message → Assign User (human)

And a classic "understand, then act":

User Intent → route → (Reply From Documents / Search Documents → carousel / Extract Data → API Call)

Used well, this group is what makes a Vivollo agent feel like it genuinely understands and helps — not just matches keywords.