AI Builder Assistant
The AI Assistant is a built-in helper within the bot builder that understands your bot's structure and can edit it directly in real-time via an SSE stream.
Opening the Widget
Click the AI Assistant button (bot icon) in the top-left toolbar of the builder. The chat panel will appear on the right.
Capabilities
- Builder Questions — ask about block types, settings, or flow logic
- Flow Analysis — the assistant sees all nodes and blocks in your current session
- Bot Editing — the AI can add, modify, or delete nodes and blocks directly on the canvas
- Auto-layout — arrange nodes on the grid using a command like "organize the flow"
- Node Highlighting — the AI can highlight specific nodes while explaining
Workflow: Plan → Confirmation
Before making any edits, the AI will propose a plan with numbered steps:
- Review the plan
- Click Confirm — the AI will execute the steps one by one
- Click Cancel — no changes will be made
Stopping
The Stop button (red square) cancels the current generation process.
Clearing History
The refresh icon button in the chat header clears the conversation history. The AI will start the conversation over.
Funnel Analytics
The AI can analyze at which steps users drop off and why. This provides specific recommendations for improving the scenario without the need to manually parse statistics.
How to ask
Ask the AI a question in natural language, for example:
- "Why aren't users registering?"
- "At which step do people leave the bot?"
- "Which commands are underperforming? Show me the stats."
- "Analyze the funnel and tell me what needs improvement."
The AI will automatically:
- Retrieve the current scenario flow
- Load statistics for each command
- Check the content of problematic commands
- Provide specific recommendations: which text to change, which button to add, or where to create a connection
Statistics field definitions
| Field | Description |
|---|---|
| Views | Number of times users accessed this command (total over all time) |
| Average Time | Average time (in seconds) from the command display to the user's next action |
| Transitions | Number of times users clicked a button or an automatic transition to the next step occurred |
| Drop-offs | Number of times a user accessed the command but did not proceed further |
| Drop-off Rate | Percentage of visits that did not result in a transition. A drop-off rate exceeding 50% for non-final steps warrants investigation |
| Current users | Number of users currently at this step who have not yet taken an action |
| Final step | Steps with no outgoing connections—the end of the scenario. A high drop-off rate here is normal |
| Utility step | Contains only logic blocks (database, API, variables, conditions). Users pass through automatically; high drop-off is normal |
| User action required | The step contains buttons or an input form—the user must click or type something. If false, the step merely notifies or automatically advances the user |
What the AI flags as a problem
Only steps meeting all the following criteria are flagged as problematic:
- Non-final — has outgoing connections by design
- Non-utility — contains messages for the user, not just logic
- User action required — requires clicking a button or entering text
- Drop-off rate > 50%
Notification steps without buttons have high drop-off by definition—the user reads the notification and moves on, or an automatic transition triggers. This is normal, not a problem.
Typical causes of actual drop-off:
- No navigation button — the user doesn't know what to do next
- Unclear text — the message doesn't explain the next step
- Dead end — the step leads nowhere
- Input form without instructions — the user doesn't know what to type
- Many users "stuck" — likely waiting for input but unsure how to provide it
Applying recommendations
After analysis, the AI will identify specific steps and suggest fixes. You can:
- Ask the AI to implement the changes immediately: "Apply all recommendations"
- Select specific ones: "Fix only the 'Registration' command"
- Reject them and make the changes yourself
Important: Statistics are aggregated from the moment the scenario was first published. A breakdown by period is not supported; the data reflects the bot's entire operational lifespan.
Security
- Canvas editing is available only to authorized users
- The AI does not perform destructive actions without explicit plan confirmation
- History is stored in the database and linked to the session
