Voice and conversation settings
Choose a voice and an available spoken language, then adjust conversation settings.
The voice engine is configured by BeAI. You choose the voice, language, and behaviour of your assistant, without having to select a provider or technology.
Choose and preview a voice
In Bots → Edit → Language, time zone, and voice, choose a voice and use the audio preview. The sample follows the bot's language, falling back to an alternative language if that sample is unavailable. A preview illustrates the tone; it is not a test of your instructions or of a live call.
Voices are identified by their name, without mentioning any provider or model. Their availability depends on the bot's configuration. Voice identifiers used by the API remain unchanged.
Previews are refreshed when the voice configuration changes. An updated sample automatically replaces the old one upon next playback, without altering your voice choice.
Save and then test your scenario with a real call. The change applies to upcoming calls. Telephony configuration, skills, permissions, recordings, and history logs are retained.
Fine-tune the conversation
Advanced panels allow you to customise settings compatible with the bot: end-of-speech detection, sensitivity, interruptions, and speech rate. Closing a panel preserves the saved values. A custom value always takes precedence over automatic settings.
Under Advanced voice settings, the parameters of the active voice configuration appear directly, with no sub-panel to expand. Speech rate is also available here when supported by the bot. Only compatible settings are displayed, even when the technical choice is reserved for the platform. Hiding other settings does not delete their saved values. If a platform administrator changes the technical configuration, the compatible panel is displayed automatically.
Start with the automatic values. If the bot cuts off hesitations, gradually increase the end-of-speech delay. If short confirmations are ignored, check the sensitivity and audio quality. A short detection delay does not guarantee an immediate response from the AI or an external service.
See Responsiveness and consumption and Identity and instructions.
Limit total call duration
Under Maximum call duration, enable the limit, choose a duration between 15 and 14 400 seconds, the conclusion anticipation, and its instruction. For a three-minute call: duration 180, anticipation 30; the conclusion starts at 150 seconds. The anticipation must remain strictly less than the total duration. Existing bots remain unchanged as long as the option is not enabled.
The timer includes the greeting, silences, AI startup, and tool wait times after answering. The preceding ringing is not included. An interruption, a new turn, or a slow tool does not reset the timer. A call that ends earlier remains finished; the system does not prolong it.
When the closing window begins, the server blocks new topics, new data collection, new business tools and inactivity prompts. Results of actions already in progress are retained without repeating those actions. The bot says goodbye once; the server waits for that response to finish and its audio to play fully before hanging up, without necessarily waiting for the maximum duration. An earlier turn ending cannot trigger this hangup. Speaking after the goodbye does not restart the conversation.
The ceiling remains independent: a monotonic clock, checks every 50 ms and media stopped no later than the limit. Actual line release is confirmed separately by telephone signalling; the network or carrier may add a delay. Speech can be cut off at the hard limit. Prepare a fallback in Calendar and answering service → Announcements → Maximum duration: it is used once if the closing response does not start, stalls or is interrupted. No speech synthesis starts at the deadline. With no active duration-limit fallback announcement, the server hangs up at the deadline if the farewell fails. The service-unavailable message is reserved for other telephony cases.
Technical details retain the start, selected settings, deadline, closing phase, hangup request and confirmed end. The reason duration_conclusion_completed means the goodbye finished and was played; max_duration_reached means the hard limit. New settings apply only to subsequent calls. Standalone closed-hours announcements are not subject to this limit. Credit constraints may end a call earlier.
The same permissions and validations apply to GET/PATCH /api/v1/bots/{id}/conversation:
{"duration_limit":{"enabled":true,"max_duration_seconds":180,"closing_notice_before_seconds":30,"closing_message":"Merci de m’avoir confié ton souhait. Je dois préparer ma visite. À bientôt !"}}
Omitted fields are retained. The post-call SMS, if enabled and eligible, always waits for the confirmed end and not the start of the conclusion.
Actions and business data
Only enable the necessary skills. BeAI controls actions using the bot's permissions and scope; a voice response does not prove that a business action has succeeded. The bot announces a search or action, then confirms its result after execution. Technical parameters and secrets remain server-side.
With the experimental voice configuration enabled by the platform, an action can remain in progress while the bot answers another question. Receipt or an "in progress" status does not constitute success: the bot waits for the corresponding final result before confirming. If the search fails or finds nothing, it informs you; a relevant retry is announced and limited to one additional attempt. It must not chain more than two tool calls without speaking to you, nor re-trigger a write that has already been confirmed or is still in progress. These instructions complement the limits and permissions of each skill.
After hanging up, scheduled wrap-up actions can continue in the background. An action already in progress retains its tracking and must not be run a second time. The summary distinguishes confirmed results from actions whose outcome remains unknown; its preparation remains silent.
For the same request, the assistant reuses searches that have already been run and waits for those that are still in progress. When a delayed result arrives, it answers any questions left open, unless you asked it to keep that result silent. You can explicitly request a new check.
A connection error or a service limit may prevent startup. Check the conversation status and provide its reference to your administrator. Historical logs are not rewritten when a technical change occurs.
Cut-offs and in-progress transfers
The start and end of the caller's speech are taken into account without stopping the audio response on their own. A cancelled action must not cut off another response in progress. Interrupting the bot's speech remains possible when the voice service confirms that the caller has interrupted it.
When a transfer to an authorised destination is already scheduled, a new identical request maintains that transfer without triggering a second one. Another destination does not replace an in-progress transfer. In case of dead air or an abnormal transfer, use Report a bug from the relevant conversation and specify the moment heard; the transcription alone does not allow verifying what was audible.
Permissions and pricing
The organisation's voice rate is uniform in EUR/min excl. VAT, regardless of the technical configuration. Recording, automatic enhancement, telephony, and other services retain their rates. Subscriptions retain their included minutes and their contractual overage rate.
Only a platform administrator can change the engine in the bot's closed technical panel. Clients and resellers do not have this option. PicApp bots retain the configuration mandated by their integration. Technical names do not appear in regular lists or detail views. The platform administrator can view them in the consumption details of a conversation after temporarily enabling Purchase estimates from their avatar.
The standard voice configuration remains the default for new bots. No automatic switch to an experimental configuration is made. A legacy experimental configuration alias is interpreted using the current compatible mode, without variants or new additional choices. This compatibility does not rewrite historical conversations, versions, or prices.
Public API
To create a bot, send your business parameters to POST /api/v1/bots without selecting an engine. To change the voice of an authorised bot:
PATCH /api/v1/bots/{id}/voice
{"voice":"Puck"}
Technical selection fields are reserved for platform administrators. An explicit attempt by another user is rejected with 403, even if the value matches the one already in use. Changing other parameters preserves the existing technical configuration.
Local samples can also be played with GET /api/v1/voices/preview?voice=Puck&locale=fr&bot_id=42 for a bot within your scope. Without bot_id, voices from the standard configuration are offered. A bot outside your scope, an incompatible voice, or a missing sample returns 404. The technical engine catalogue is reserved for the platform; it does not play a part in customer pricing.