Skip to main content
Belcenter AIHelp center
Configure a bot

Configure by conversation

Configure a bot via a guided in-browser or phone call, upload documents, and confirm choices.

The time zone detected by the browser is also suggested when starting the first interview for a bot created via subscription or through the standard form. Resuming a session retains its decisions.

The setup assistant guides you through a live voice conversation. You describe your business; it asks straightforward questions, logs decisions in the tracker, and configures the bot after your confirmation. It uses the standard voice configuration. No setup questionnaire or text chat is required.

Getting started

Under My assistants → New bot, first choose BeAI number or SIP connection. The next screen offers Create with assistant or Configure manually, along with a reminder of the selected telephony option and a link to change it. Choose Create with assistant, then Start guided configuration. Your telephony choice is retained, even if you return to manual configuration. The name and role will be determined during the conversation; the time zone is suggested based on your browser. No phone number is purchased.

After manual creation, the bot details page opens; you can launch the guided setup from there if needed. It is also available via Customise with the assistant after subscribing on the website. You can find the Configuration wizard on the bot details page.

You must have permission to edit the bot: platform administrator, or manager within the scope of your organisations and tags. A view-only account cannot start a call or view decisions or documents.

For a PicApp bot, open its settings from the Enreach SSO link, then click Configuration wizard. The interview, decisions, and documents remain restricted to this bot, with no separate BeAI sign-in. Manual setup takes you back to its settings; Close configuration ends SSO access. The product name, logo, and colours are inherited from the bot organisation's first ancestor with custom branding defined. The SSO session is limited to 30 minutes: reopen the Enreach link if it expires. Opening the page does not initiate a call or incur any charges; the startup conditions and confirmations remain those described below.

Settings locked by the Enreach Solution cannot be overridden during application. If the Solution hides instructions, greeting message, language, or voice, the interview is unavailable to avoid revealing these values, including in a previous draft. Use manual configuration for all other settings.

Choosing how to talk

The assistant starts in the language selected for your interface (French, English, Dutch, German, Spanish, Italian, or Portuguese), including for telephone callbacks. It adapts if you speak to it in another language. The interview language is independent of the language you choose for your future bot.

  • In my browser: click Enable microphone and call, then grant microphone access. A microphone and headset are recommended. The conversation uses WebRTC: you can speak while the assistant is responding to interrupt it. Mute my microphone pauses transmission; End call disconnects the conversation. The waveform shows your real microphone level. Opening the page never activates the microphone.
  • On my phone: enter your own number, then click Call me at this number. You will receive a call from 071 96 07 69, the Beone number displayed in the box. Keep the tracker open during the call to view decisions and upload documents. Belgian landline and mobile numbers are accepted; special-rate and premium-rate numbers are excluded. The phone will ring for up to 30 seconds. If the line is busy or unanswered, no automatic callback is made. You can request a new one: a maximum of three requests per hour per number/user. A shared capacity limit may also postpone a callback.

The phone option appears when the callback service is available. If your browser denies microphone access or your network blocks WebRTC, grant microphone access, switch networks, or choose a callback. If audio is muted, click Enable sound. Do not close the tab during a browser call; for a callback that has already been answered, the phone conversation continues even if you close the tracker.

You can end a call from the tracker, including in another tab: the termination is applied at the next heartbeat, usually within five seconds. Hanging up the phone also ends the interview.

Building your business context

No knowledge of AI is required: the assistant asks one question at a time, guides you with examples relevant to your industry, and clarifies anything too general. It reuses your answers so you never have to start over.

It begins by understanding:

  • Your company or trading name: its confirmed name, what it offers, its products/services, and its clientele. This name may differ from your BeAI organisation name and the bot's first name.
  • Welcoming the public: shop, offices, practice, with or without an appointment; address and useful access details. If there is no physical location open to the public, it clarifies your service area, delivery, or remote operations instead.
  • Opening hours and availability: days, hours, breaks, and exceptions, where applicable by location; what the assistant should do outside opening hours. An online business does not imply continuous human availability.
  • Useful information for callers: public contact details, website, FAQs, procedures, information to prepare, turnaround times, or prices that the bot can provide. Follow-up prompts depend on your business; it will not ask for all these items if they are not relevant. Providing a link does not trigger automatic browsing.
  • The phone assignment: examples of common requests, questions to ask, details to collect, limits, and situations to transfer to a human. Simply selecting a bot type activates neither calendars nor transfers.

These items appear in the decisions list, under distinct sections for company details, practical information, opening hours, and the scope of work. View shows their full details. Once confirmed, their summary is added to the bot's Additional info and serves as context for its future calls.

You can say "I don't know", "That's not relevant", or "I'll specify that later". The assistant then notes this point and the cautious behaviour to adopt; it does not invent addresses, opening hours, or prices. Practical information and opening hours must be addressed before validation, even if your answer is to postpone. Past sessions that have already ended keep their original tracking.

Decisions made together

The assistant only gathers:

  • The bot's name and purpose; it deduces the appropriate type itself, without asking you to choose a technical category.
  • The primary language, asked as a question and saved after your reply. If the context suggests French, the assistant offers Français (Belgique), Français standard, or Français (Canada) and adds "or another language" before letting you choose; your location does not decide for you. An answer of "French" without further detail prompts a follow-up question on the regional variant. This choice adapts phrasing and vocabulary independently of the voice selection.
  • The time zone, confirmed in a separate question.
  • The voice: rather feminine or masculine, calm or cheerful... It suggests a voice compatible with the bot's model. The voice used during the call remains that of the configuration assistant; it does not serve as a preview for each proposed voice.
  • The greeting phrase, behaviour upon pick-up, and instructions to follow during calls.
  • Relevant documents and how to use them.
  • Any email notifications, recipients, and the circumstances in which to send a message.
  • The automatic summary of each conversation after it ends, sent to one or more email addresses. This option is separate from emails sent in specific situations and also works without audio recording.
  • Recording future calls for the bot and whether an announcement should precede this recording.

On desktop, the call and documents appear on the left and decisions on the right, in a layout tailored to your screen height. Your choices remain visible throughout the conversation. View displays the details of a long decision; View steps shows completed and pending topics. An animated indicator signals ringing and phone connection; it does not reflect audio volume levels. It respects your reduced motion preference. On mobile, panels are stacked vertically.

Decisions appear in read-only mode as the session progresses. To amend a choice, tell the assistant. Decision history allows you to view successive changes (up to the last 500). The type, voice, and destinations remain validated by the server.

ERP, CRM, and calendar connections, transfers, SMS, AI models, pricing, identifiers, and SIP settings are not modified by this guided setup. No emails are sent during configuration. Existing advanced features are preserved.

Uploading documents during the call

The assistant asks if you have price lists, FAQs, catalogues, or relevant instructions. If you have any to provide now, the assistant waits for your upload and does not move on to the next question. Upload the first one in Documents without leaving the call. As soon as an upload is received, it asks what the file contains and in which situations your bot should use it. The explanation is added to the instructions and displayed in the tracking panel. It then asks if there is another document: upload it and explain its use, as many times as needed up to the five-file limit. Say "I have no more" or "I'll do it later" to proceed with configuration. Simply uploading a file does not complete this step; every retained file must have an explanation. Time spent waiting is included in the connection duration.

Formats and file sizes match the Documents skill: PDF, Word (DOC/DOCX), Excel (XLS/XLSX), PowerPoint (PPT/PPTX), OpenDocument (ODT/ODS/ODP), HTML, TXT, Markdown, CSV, JSON, and XML. Limits: 5 documents, 200 MB per file (200 000 000 bytes). There is no longer a 60 KB limit on extracted text. Scanned PDFs are supported; readability depends on the document service's OCR capabilities. An identical file is not added twice. You can remove a document before confirmation; doing so does not delete any pre-existing documents from the bot.

The conversation receives an optional preview of up to 8 000 characters, without full reading during the interview. Its absence does not block the upload: explain the content and its intended use to the assistant. The complete original file is kept in private storage and then sent for standard document processing after validation. Removing a file or closing without applying deletes the pending file; originals from expired sessions are purged automatically. All uploaded documents must have an explanation before applying. If you no longer need a file, remove it.

After confirmation, the tracking panel indicates the preparation status of each document. Wait until it is ready before testing its responses. In the event of a failure, use the bot's document management area to review the text and replace the document.

Confirming before the end of the call

Once choices are ready, the assistant provides a spoken summary and asks for your confirmation. All required topics must be covered before this summary, even when values were already suggested. A natural agreement is sufficient: "yes", "agreed", "that works for me", or "go ahead". No specific wording is required. A correction or a new document requires a new summary. An agreement given to a previous question does not validate the entire configuration.

If an issue occurs, the tracking area indicates the point to resolve. You must address that point rather than repeating your confirmation.

You can also click Confirm and configure my bot once this summary is ready. Applying creates a new bot version and starts document preparation. A double-click does not trigger a second application. If the bot was modified elsewhere, validation is rejected to protect those changes; end the session and resume from the current bot at the stated rate, or continue via manual configuration.

Before hanging up, the assistant presents additional possibilities: add skills by editing the bot to connect an ERP, CRM, or calendar. The Beone team can discuss your project and offer guidance from a dedicated AI consultant. Any remaining bot telephony setup can be completed as required. Then, place a test call.

Welcome offer and usage

The first session for a new bot created since this feature became available is offered free of charge by Beone, once per bot. Start it within 30 days of creation. It includes up to 30 minutes in total, with no limit on the number of turns, and can be resumed for up to 30 days after opening. Closing or applying the session uses up this offer.

There is no automatic switch to paid billing. A subsequent session requires accepting the displayed rate. With a subscription, included minutes are used first, followed by the out-of-bundle rate specified in your contract. Otherwise, euro credits are used at the organisation's standard voice configuration rate. Example: at 0,40 € excl. VAT/min, three minutes cost 1,20 € excl. VAT; the initial complimentary session charges the customer nothing.

A connection lasts a maximum of 30 minutes, within the limit of remaining credit and complimentary time. Connected time is billed by the second, including silences and muted microphone. Callback ringing is excluded; telephony routing is covered by Beone. Ending the call stops the timer. The technical finalisation of the summary is not billed.

Credits are held temporarily, and any unused portion is released. A connection that never started is not charged. If it disconnects after starting, the last saved tracking data is used to finalise consumption. Decisions remain available for review. An unconfirmed summary warning prompts you to resume the interview before validating. Paid charges appear in Subscription & billing → History, under Configuration assistant.

The Export transcript button downloads a text summary without opening a written chat. Audio from this interview is not retained. The transcript, decisions, and document previews are retained for resuming later. The recording setting selected during the interview applies to the bot's future calls.

Public API

The same flow is exposed under /api/v1/bots/{id}/setup-assistant: status/pricing, sessions, audio start /voice, browser negotiation /rtc, stop /stop, documents, decisions, and apply/complete. Creation via /api/v1/organisations/{id}/bot-setup accepts telephony: "number" or telephony: "sip" and the optional time zone timezone. The status field telephony confirms this choice. For existing integrations, omitting telephony retains BeAI Number mode. An unknown value is rejected (400); reusing the same idempotency key with a different choice is rejected (409). No number is purchased and no SIP credentials are created by this operation. Steps activity, knowledge, and hours of checklist must each be covered with a non-empty summary; an explicit deferral is accepted. Their texts are retained in draft, decisions, and, once applied, the bot's extra_infos. Initial interview language selection uses the authenticated user's preference, without requiring new API parameters or permissions. The session exposes catalog.languages (code and name of enabled languages, matching manual setup). draft.language retains the exact chosen code, such as fr_be or fr_ca; tracking displays its label. Status exposes document_upload with accepted file extensions and maximum size. Each document indicates size_bytes, mime_type, and preview_complete; characters counts the preview, not the full document. Uploads from older sessions remain usable. Status exposes phone_caller.number (+3271960769) and phone_caller.display (071 96 07 69) to announce the callback number. Read the session to monitor decisions and documents during a call. checklist exposes the steps, missing, next, waiting_for_documents, and pending_documents; guidance_error describes the latest blocker. The draft exposes summary_emails (e.g. ["accueil@exemple.be"]) for automatic summaries, and document_collection (unasked, collecting, complete, deferred) for guided data collection. The same validations apply across the portal and the API.

Former endpoints /messages and /draft have been decommissioned and return 410. Their histories remain viewable. Refer to OpenAPI for schemas, revisions, idempotency keys, WebRTC exchanges, and access controls.

Support history

Automated transcripts, decisions, and interview diagnostics can be reviewed by platform administrators to assist you. This tracking does not create an audio recording. It remains available after the resume window expires; deleting the bot retains this platform-admin-only history, marked as "Bot deleted". See Monitor auto-configurations.

Documentation updated with the application