Build

AI Flows

Connect a team of AI agents, each with its own model and prompt, plus knowledge from your website or files.

What an AI flow is#

An AI flow is a team of steps that answers on its own. Each step does one job, such as finding the right part of your knowledge base, deciding what the customer needs, or writing the reply, and passes its result to the next.

Create one from AI Flows: start with a Blank AI flow or import From an OpenAI assistant (uses your own OpenAI key). Each flow has a Setup tab and a Team builder canvas. Use Run once or Save and preview to test, then Publish.

Knowledge
Search your docs
LLM
Answer with context
Filter
Needs a human?
Text
Reply to customer

A typical support flow: retrieve context, answer, then route on the result.

Step types#

Step	What it does
LLM	Calls an AI model with a system prompt and a prompt. Can answer in a fixed schema (below).
Knowledge	Searches the flow's knowledge (a website you scan, or uploaded PDF, TXT, CSV, Excel or HTML files) and returns the most relevant passages. Tune Top K and Top N to control how many.
Filter	Checks conditions on earlier results. Then go to one step, Otherwise another.
Python	Runs your own Python code for custom logic.
Text	A fixed text step.
Jump	Continues from another step.

Models#

Pick a model per LLM step. Models run inside Officely, so you don't need your own API keys. Available families include OpenAI GPT (GPT-4o is the default), Anthropic Claude, Google Gemini, Meta Llama, Mistral and Cohere.

Use a fast, low-cost model for simple steps such as classification, and a stronger one for the final answer. You can also set the model from a variable to choose it at run time.

Structured answers#

Turn on Answer with a schema to make an LLM step return named fields instead of free text. Field types are Text, Number, True/False and One of… (a fixed list of values). Start from a preset (Lead, Classification, Sentiment, Yes or no) or Paste a JSON example to create the fields.

Later steps read each field as {{AgentName.field}}. Type @ in a prompt to pick one. See Variables.

Example: a Classify step's schema output

{
  "intent": "order_status",
  "urgent": false,
  "order_id": "3234379"
}

Schemas are available on supported models; the option appears on the LLM step when the selected model supports it.

Put an AI flow to work#

After you publish, the Your AI flow is live panel gives you:

a Chat link to share,
Website embed code for your site,
Connect a channel, which takes you to Channels.

To use an AI flow inside a scripted chatbot, add an AI Agent step in Chat Builder and select the flow. Links and embed codes need a paid plan.