Release 5.8.0
When your setup team works in Constructor, you can choose a safe change mode, follow an interface tour, run a ready-made skill, and validate a broadcast file against the full client base — without leaving the current section.
Constructor — AI Assistant
New
- Three work modes. Ask — answers and explanations from documentation and instance settings only; Agent — answers plus automatic actions in the system; Plan — a step list with risk level first; actions run only after you click Confirm.
- Skills in the UI. Type
/in the input field to open a catalogue of ready-made scenarios for common tasks: settings, clients, bot scenarios, and more. - Interface tours. For questions like “where is this in the interface?”, the assistant offers step-by-step highlights; in chat — Show or No, thanks. Hints match the section you currently have open.
- Model selection — up to five options in chat; the available set depends on provider keys on the instance and role permissions.
- Chat history — separate topics, new chat, rename and delete (stored in the browser on this computer).
Improved
- AI provider keys on the instance — the root administrator configures them under Menu → Settings → Instance settings → Instance Agent; the model list in chat reflects the configuration.
- Execution steps in chat — while the assistant works, you see stages such as documentation search, web search, and system actions.
Fixed
- Create-user tour — reliable start after the chat offer, wait for the settings page to load, and the password generation step.
Broadcasts
Improved
- Phone file upload now finds all matching clients across the entire base, not only clients of the active bot in the sidebar.
- Channel selection after upload shows how many recipients from the file are reachable for the chosen send bot.
Operator Line
Improved
- Operator statuses in large teams update without delay after workload changes — supervisors see the current state.
Quality Assurance
Improved
- Re-running evaluation on the same dialog with the same checklist gives the same result — easier to compare scores across periods.
- Grouped evaluation — when there are many criteria, the system processes them in batches and produces one summary score instead of fragmented results.
- Checklist criteria limit — the editor shows a counter; add is blocked at the limit; saving shows a clear warning if the limit is exceeded.
Please note
- AI Assistant is available only after enabling Instance Agent Integration in instance settings.
- Web search and URL reading (available since 5.7.0) require separate enablement on the instance and a role permission; disabled by default.
- Agent Read-Only Mode on the instance blocks all AI Assistant automated actions, regardless of role permissions.
- Chat history is stored locally in the browser — topics may not appear on another computer.