View history
Search, filter, and view the presented outbound CLI, actions, errors, tokens, and costs for a conversation.
The Conversations list brings together inbound calls, outbound calls and written exchanges accessible to your organisation.
Search and filter
Use search and filters to target:
- a date range;
- a bot or a runtime node if you are a platform administrator;
- the channel and call direction;
- an outbound campaign;
- language, status or voicemail detection.
Filters apply automatically. The date picker offers ranges such as this week, this month, this quarter or this year, with navigation to the previous or next period.
Read details
A conversation's details may display:
- the recording and its timeline;
- caller and bot messages;
- tool results;
- the summary, language and sentiment;
- transfers, emails, SMS or other actions;
- a user-readable initialisation error.
The list and details group Actions by type: a distinct icon for each type and the number of identical uses. For example, two lookups share an icon, whereas the transfer and note each keep their own. Hover over the icon or give it keyboard focus to open a styled tooltip showing each action, its result and its timestamp in your time zone. The tooltip scrolls if necessary and closes with Escape. Older actions without a date display "Timestamp unavailable". Exports, the API and statistics retain all individual uses.
Platform administrators can view system messages intended for diagnostics.
These messages are excluded from the transcript detail and export for other
users, both in the interface and in GET /api/v1/conversations/{id}. Caller and bot messages,
the summary, and actions remain accessible according to conversation permissions.
The purchase estimation option does not grant access to diagnostics.
Live monitoring retains operator instructions and business states;
technical report generation errors remain restricted to the platform.
After a voice call, the final transcription may replace the user side in the background. Bot messages remain unchanged. The processing status appears in the details. If bot recording was not enabled, the capture required for processing remains temporary and is not available for playback.
Find the outbound CLI
In the list, open Columns and enable Presented outbound number. This column is hidden by default; your choice is retained. The same number appears in the details of an outbound call and in the CSV export. It is distinct from the recipient’s number and the inbound number reserved for the bot.
This is the identity sent by BeAI when placing the call: BeAI number, verified custom number, or number presented by standard SIP configuration. Changing the bot configuration afterwards does not alter this historical value. However, an external PBX may rewrite the number presented to the recipient.
For earlier BeAI calls, the number is included when it was retained in the billing details. Without this historical data, the list displays a dash and the details show ‘N/A’. No current number is inferred to complete past calls. Transfer legs using a BeAI number each display their own number in the telephony billing details.
The same conversation access permissions apply. In the API, responses
from GET /api/v1/conversations and GET /api/v1/conversations/{id} contain the
read-only field outbound_presented_number, for example
"outbound_presented_number": "+32475123456". It is null for an inbound
conversation, another channel, or an unavailable historical number.
Understand tokens and costs
Purchase estimates are hidden by default. A platform administrator can enable them temporarily from their avatar menu via Purchase estimates, at the bottom of the menu after My orders and before Log out. This entry uses the same layout as the other options, with a $ symbol on the left and a small tickbox on the right. Separators isolate it from orders and logout. The conversation detail breakdown then displays the operations, quantities, unit prices, and estimated costs. Technical names remain absent from lists, even for administrators. The detail view of
GET /api/v1/conversations/{id} exposes these same details to platform
administrators in token_details and token_cost_details, with the explicit option include_purchase_estimates=true. Without this option, estimations remain hidden. The interface preference is not saved in cookies or in the database; it is disabled once the browser is closed. Depending on browser capabilities, closing all BeAI tabs or navigating to another page may also reset it: re-enable it if necessary.
Inbound text tokens include the instructions, tools, and history processed by the model. They do not correspond to the number of words spoken. The voice engine re-processes the context on every turn: these usages accumulate, even if two turns have identical counts or if the context shrinks. Each report is counted only once.
Final transcription remains a distinct line item and is added to the call cost. The total also includes self-improvement whenever it was run.
The metering fix applies to calls processed by the updated runtime. Earlier calls, whose per-turn usage reports were not retained, cannot be recalculated exactly. Their estimations must not be used to compare models with newer calls.
Export or act on multiple rows
Select rows and open Actions. Available operations depend on your permissions and the table. The export includes the current selection or the scope provided by the view.
Tool details
Platform administrators see a compact row for each tool event, regardless of conversation origin: calls, WhatsApp or Messenger. The name and status remain visible; the i button opens the available arguments, result and duration. Secrets are masked. Details missing from older logs are indicated as not retained, without fictitious reconstruction. Diagnostics remain restricted to platform administrators in the console, API and exports.
For Picapp tools, prepared indicates that the transfer is awaiting the voice announcement and has not yet been sent. Request accepted indicates that Enreach accepted the request, without confirming that the recipient answered. Awaiting response and uncertain result remain distinct from a rejection or failure. These statuses also apply to historical events when viewed, preserving raw responses. In the API, messages[].tool_trace.status uses prepared, accepted, pending, and uncertain respectively; a Picapp outcome ok: false may therefore correspond to a normal, not yet confirmed step.
Receipt of the final report uses the same layout: final report received — completed, with an i button to view retained details (summary, language, sentiment and voicemail detection). This also applies to historical conversations upon viewing them, without replaying processing or altering their summary. In the API, this event appears in messages[].tool_trace with name: "final_report_received", status: "completed" and the report in result. The same access restrictions apply.
Included minutes and profitability
For a subscription call, the list and detail distinguish retained media duration, consumed units, included seconds and the extra actually charged excluding VAT. A 0 EUR extra means these units are covered by the subscription; it is not a loss per call. The existing 60-second minimum is retained: 49 seconds of media can consume 60 seconds. Elapsed time between start and end includes preparation that differs from media time. Missing historical data remains unavailable, without reconstruction using current rates.
The list, detail and CSV export share these values with usage in GET /api/v1/conversations, GET /api/v1/conversations/{id} and GET /api/v1/conversations/export.csv. The API CSV uses the same filters and permissions as the API list; purchase estimates require a platform administrator and include_purchase_estimates=true.
A platform administrator can enable Purchase estimates, then open Period profitability under Subscription and billing. Periods of ended contracts remain accessible. The report counts each initial, monthly and prorated charge confirmed paid and applied once, excluding VAT, adds extras actually charged and subtracts available estimated costs. Unpaid charges, wallet top-ups and late prorated charges credited to the wallet do not become period revenue. No theoretical subscription revenue is attributed to calls.
The report identifies missing or partial costs and differences against the consumption counter. The known-cost balance remains a gross subtotal; an estimated margin with incomplete data remains undetermined. Net margin is never claimed: unrecorded telephony, storage, indexing and other costs are excluded. USD costs use the current report rate shown at read time; EUR costs are not converted and other currencies are marked unavailable. This rate does not rewrite historical settlements. For calls spanning periods, media and costs are shared by consumed units, overage extras by overage seconds and other extras by consumed units.
The same report is available at GET /api/v1/organisations/{id}/subscription/reporting?include_purchase_estimates=true&page=1, with 20 periods per page. Permissions and explicit activation are identical: access is denied without either, and an invalid page is rejected. The daily alert assesses subscription calls by period, without a loss alert based on a zero extra. Available estimated costs exceeding paid revenue and extras are still flagged when period estimates are complete. This view changes no rates, minimums, contracts, history, balances or payments.
A reservation for a future period, settled at 0 without a consumed segment in that period, is not a missing call. A call entirely outside the allowance is still counted; a missing historical settlement remains unknown.