AIFlow

AIFlow shifts an agent from "answers well" to "leads the conversation". A plain agent waits for the customer. A scripted agent recognises where the conversation is, what the customer is signalling, and nudges it forward — greeting, qualifying, pitching, handling objections, closing.

The editor is a canvas of connected blocks. Three block types, composable like Lego. Customer triggers describe what the customer might do — "mentioned their city", "asked about price", "wants to book". Logic operators — AND, OR, IF, ELSE — let triggers combine into meaningful conditions. Agent actions are the instructions the agent will splice into its reply when a branch fires.

The matching happens live. A modern model (from SuperAdmin) evaluates incoming messages against the triggers on every turn. The owner writes triggers in plain language, not regex — "customer asks about warranty", "customer seems ready to buy" — and the model handles detection.

When a trigger fires, the corresponding actions become the instruction layer on top of the agent's next reply. The agent stays human — it is not reading from a script — but it is guided by the scenario in the background.

The scenarios hub lists every flow in the workspace, with create and edit shortcuts. Flows export to a file and import back, which makes templates, backups, and cross-project reuse easy. Each scenario shows which agents are currently bound to it so the owner can see the blast radius of a change.

Conversation openers have their own dedicated branch — when a customer arrives for the first time, the "start of conversation" actions fire: a specific greeting, an opening offer, a specific tone. That kills the awkward first two messages that come with generic bots.

AIFlow is overkill for a simple FAQ bot, and that is fine — an agent works without a scenario. It is the right tool when a conversation needs to move a customer along a funnel, and it does that job without the owner writing a line of code. The interface is fully multilingual with RTL-aware layout.