Monitor autoconfigurations
Review guided conversations, decisions, confirmations, errors, and consumption from platform administration.
In Administration → Platform → Autoconfigurations, a platform administrator can view configuration sessions across all organisations. Client and reseller administrators do not have access to this screen. Their standard permissions for monitoring their own bot remain unchanged.
Find a session
Search by bot name, organisation name, or the full session ID. You can filter by bot or organisation ID, status, creation dates, and the presence of an uncompleted or failed attempt. The filter dates cover UTC days; displayed times are in Brussels time. The list is paginated, from 10 to 100 sessions per page.
Resumed attempts for the same interview remain grouped together. Each row displays the channel used (browser or telephone), the number of attempts, rejected tools, duration, and usage. Refresh reloads the data; Auto-tracking updates it every ten seconds while the tab is visible. In the event of an error, the latest data remains displayed and auto-tracking stops; the data shown may therefore be outdated.
View details
Open View to find:
- Available transcripts, grouped by connection and in the order received, with statements from both the user and the assistant.
- The draft, topics already covered, and decision history by revision.
- Uploaded documents and their explained purpose to the assistant, without their full contents.
- Configuration tools, whether they were accepted or refused, and explanations of any blocks, notably confirmation steps.
- The actual application of the configuration and its version number, distinct from a simple agreement noted in the transcript.
- Finalised duration, provisional duration of an ongoing connection, and finalised client debits in euros or subscription seconds. A complimentary session is flagged as Provided free of charge by Beone; this amount does not represent the supplier cost borne by Beone.
Viewing these details does not modify the bot, close any connection, or trigger any calls or billing.
Estimate the cost incurred by Beone
After temporarily enabling Purchase estimates from the avatar menu, the Estimated AI cost column displays voice engine token costs, including for a complimentary session. The breakdown shows input and output tokens, their text/audio distribution, the purchase price per million, and the cost of each category, per attempt and for the entire interview. Rates correspond to conversations and are locked at the first report of each attempt. A resumed session may therefore carry a different rate if the pricing tier has changed in the meantime.
Counters are accumulated from provider reports, which may include context re-used across exchanges. This is an estimate, not a provider invoice. The currency displayed is that of the purchase rate card (for example, USD), without automatic conversion to euros. Telephony, storage, and document indexing are not included.
Partial estimate indicates an ongoing or interrupted connection, incomplete counters, an attempt without counters, or a missing rate. The amount displayed is then the known subtotal. Tokens unavailable specifically indicates an older interview whose counters were not retained: the dash does not mean its cost is zero. If attempts use different currencies, view their amounts separately; no total mixing currencies is displayed.
This data is restricted to platform administrators and affects neither the complimentary initial interview nor customer charges.
Understanding statuses and limitations
To resume indicates an open interview without a known active connection; Connecting means an attempt is ongoing. Tracking stopped appears after more than three minutes without an update: this is not proof that a call is still active. Configuration applied confirms a successful application; a finished conversation may remain To resume if it did not lead to this application. Closed without applying and Recovery expired remain viewable.
Previously recorded sessions are included. This is an automatic transcription, without audio recording: speech recognition errors or the final words of an interrupted connection may be missing. Older interviews may lack a checklist or timestamps for tools; this information is not reconstructed. The screen displays the last 500 decisions and the last 100 tool results per connection.
The expiry of the resumption window after thirty days does not delete the history. Deleting a bot now retains its sessions, transcriptions, decisions, described documents, and costs in this history. The notice Bot deleted replaces the link to its profile; its previous ID, last recorded name, and organisation remain viewable and usable in filters. An archived session no longer allows restarting configuration or document indexing. Sessions already deleted prior to this update cannot be recovered without a backup.
Equivalent API
With a personal API key from an unlocked platform administrator:
GET /api/v1/bot-setup-sessions?bot_id=192&status=open&page=1&limit=25
GET /api/v1/bot-setup-sessions/{sessionId}
The collection returns data, meta (page, limit, total, pages), and filters. The detail is wrapped in data. The bot.deleted field distinguishes deleted bots; bot.id preserves their previous ID. With include_purchase_estimates=true, each session and attempt can expose provider_cost: status (estimated, partial, unavailable), model, currency, total (number or null), tokens (object or null), reports, items (category, tokens, unit price, and cost), and calculated_at. For instance, 1 000 audio input tokens at 3 USD/million and 500 output tokens at 12 USD/million yield total: 0.009 and currency: "USD". amount_eur remains the customer charge. Filters are search, bot_id, organisation_id, status (open, applied, closed, expired), activity (pending, failed), from, to, page, limit. IDs must be positive; dates follow AAAA-MM-JJ and the end date cannot precede the start date. An invalid filter returns 400, an account without platform permissions 403, an unknown session 404. An API key does not retain permissions of a user whose role has been revoked.
The full schema is available in the API documentation. Observe a ten-second polling interval for tracking. Responses are not cached and do not expose connection tokens or technical callback parameters.