This guide is intended for administrators. Select the relevant client and prepare the access required for each provider used. Viewing and editing rights remain distinct.
Opening the new pages
Open AI Integrations in the administration menu, then the desired function. The visible entries and panels depend on your rights. Wait for them to load; an access denial requires a permissions check, not the creation of another provider.
AI settings are no longer a section of General Settings. Caller screening and automatic voicemail transcription also have their own AI pages. The former addresses for AI settings, voice agents, voicemail transcriptions, and the chatbot redirect to the corresponding new screens.
| Entry | Usage |
|---|---|
| Call Transcription | Find recordings, listen, and view or request their analyses |
| Voicemail Transcription | Enable automation, read texts, and track jobs |
| Voice Agents | Create, configure, and simulate an agent; see the voice agents guide |
| Conversational Agent | Configure the chatbot per site and open the voice agent simulation |
| Providers | Configure access, assignments, and the client's AI policy |
| Live Transcription | Choose the authorized population, retention, and languages |
| Identify Unknown Callers | Configure screening and open the identification log |
| Text-to-Speech | Configure voice assignments and the brand voice |
| AI Assistant | Ask a question and review proposed changes |
| Installation AI Settings | Manage system values; see the installation guide |
On pages with shared saving, Save saves the modified settings and Cancel restores the previously loaded values. These buttons are unavailable without any changes or while saving is in progress. Check the result: an error does not confirm that all panels were saved. Tests, generations, and forms with their own button are separate operations.
Call transcription and analyses
- Open AI Integrations › Call Transcription.
- Search by number or name.
- Choose All, Less than one minute, or One minute or more.
- Locate the call by its date, caller, recipient, and duration. Pagination lets you browse the list and choose its size.
- Expand the row to display its details, the player, and the analyses. The row icons can also be used to expand or collapse it.
| Command | Result |
|---|---|
| Listen | Loads the audio into the expanded row; then use the player. Download saves the available file. Collapsing the row pauses playback. |
| Transcribe | Requests the transcription if no result is already available |
| Summarize | Requests a summary of the call |
| Analyze sentiment | Shows the tone obtained: positive, neutral, or negative |
| Evaluate quality | Shows the score, its maximum, and the justification when the result is usable; without an entered scale, a default scale is used (greeting, understanding of needs, clarity of responses, courtesy, closing) |
Opening re-reads results already saved without triggering a new generation. An icon with a checkmark indicates an available result. The animation indicates reading or processing in progress; wait for it to finish. In case of failure, Retry re-reads the result or relaunches the failed operation. The icon indicators specify their status.
Viewing requires read rights on recordings; listening requires their download rights, and generation requires their write rights. "Recording unavailable" indicates the absence of the file; a result already saved may remain viewable. An empty list and a read failure are two different states.
The transcription offers Copy and Export as a text file. Translation of the transcription or summary is offered to accounts authorized to modify recordings. Review the texts along with the audio before using them.
Voicemail
Open AI Integrations › Voicemail Transcription. Automatic configuration, viewable texts, and job tracking are presented on this page.
| Field | Purpose | Possible values | Example |
|---|---|---|---|
| Automatic transcription | Enables automatic processing of the client's voicemail messages | Toggle switch; editable according to voicemail rights | Enabled after service configuration |
| Voicemail box | Selects the messages to view | Client-accessible box | Main reception box |
Change Automatic transcription, then click Save in the header. Cancel discards this change. The display of the setting also depends on access to the parameters; the absence of this panel does not necessarily prevent authorized viewing of the texts.
- Choose the Voicemail box.
- Locate a message by its caller and date, then click View.
- Read the saved transcription, copy it, or export it. Use the available translation if needed.
- Use Refresh to re-read the data. Previous and Next browse boxes and messages, in pages of 50.
"No transcription found." does not mean the audio has been deleted. This view reads existing results; it does not offer a button to transcribe the selected message.
Job tracking refreshes every ten seconds. It also offers refresh and pagination. For a successful job, open Transcription to read the text. In case of a read error, check the displayed diagnostic before retrying.
Chatbots and simulation
Open AI Integrations › Conversational Agent. Site chatbot configuration and voice agent simulation are two separate panels, displayed according to your rights.
| Field | Purpose | Possible values | Example |
|---|---|---|---|
| Web livechats | Selects the site to configure | Existing site; the first site is selected if no valid selection is present | Main site |
| Chatbot enabled | Authorizes chatbot responses on this site | Toggle switch | Enabled |
| Automatic response | Authorizes automatic responses until an agent's first reply | Toggle switch | Enabled |
| Personality / instructions | Defines tone, scope, and handoff to an agent | Optional; empty keeps default behavior; 4,000 UTF-8 bytes maximum | Answer about hours then suggest an agent |
Click Save in the site panel. The button is unavailable without a draft, while saving, without write rights to the parameters, or if the text exceeds the limit. Accented characters can occupy multiple bytes. Check for "Saved", then test a conversation on the site. The chatbot uses the Conversational Agent function, the client's policy, and AI quota.
In the Voice Agents panel, Simulate opens the preview. Choose an agent and review its greeting, instructions, responses, fallback, and limits. The edit link opens the voice agents page.
| Field | Purpose | Possible values | Example |
|---|---|---|---|
| Voice Agents | Chooses the saved configuration to simulate | Existing agent | Information reception |
| Message | Prepares a dialogue turn | Non-empty text, 2,000 characters maximum; write rights on agents required to launch | What are your hours? |
Click Simulate, wait, then read the response, the transfer or fallback decision, or the provider diagnostic. A decision other than a response blocks a new turn; Refresh clears the test dialogue to start over. Close closes the preview. The simulation uses text generation that may consume the quota; it does not place a phone call.
Client providers
Open AI Integrations › Providers. Panels may be limited by your rights on AI and secrets. Check authorization to use installation keys, the client's policy, providers, and per-function assignments.
| Field | Purpose | Possible values | Example |
|---|---|---|---|
| Allow artificial intelligence | Authorizes AI usages for the client | Toggle switch | Enabled |
| Data processing location | Limits the usable services | Remote services allowed or On-premises only | On-premises only if a compatible engine exists |
| Provider / Model | Assigns requests for a function | Available choices; a system lock may impose the value | Already configured provider |
- Use Add a provider, then Add on the chosen service, or Edit on an existing provider.
- Follow its guide and fill in all required parameters. Fields vary by service.
- Use Test this configuration to test the draft.
- Click Save, then test the saved key using the provided command.
- Return to per-function assignments, choose the provider and model, then save.
| Provider field | Purpose | Possible values | Example |
|---|---|---|---|
| Provider access key | Authorizes requests | Secret required or optional depending on the service | Server key provided by the provider |
| Service address | Identifies the service | Address required or optional depending on the service | Address from the configuration guide |
| Additional parameters | Specify the account, language, voice, or deployment | Fields displayed for the selected provider | Supported language |
Secrets are not re-read. Saving the client's form replaces its protected configuration: fill in the necessary parameters. Cancel clears the draft. The access test does not guarantee the availability of all models or the credit needed for a generation. Its diagnostic and the latest test results remain visible on the screen.
Remove configuration requires enabling the displayed confirmation. Assignments are not removed along with the key; subsequent requests may fail. A client-specific key takes priority over the installation's key. Local processing mode blocks remote providers; verify that a compatible engine exists before selecting it.
For Summary and Sentiment analysis, you can choose a text generation provider or the audio transcription provider, if it offers these analyses. The latter works on the recording's audio: the summary keeps its Key points, Decisions, and Action items sections, and the tone remains positive, neutral, or negative.
AI notice to caller has a toggle switch and non-empty text, limited to 2,000 characters. Save these changes from the page. The cap and usage reconciliation are described in Quotas and usage; these panels are now located here.
Live transcription
Open AI Integrations › Live Transcription with read rights on AI and parameters. Selections of people, queues, and groups require their respective read rights.
| Field | Purpose | Possible values | Example |
|---|---|---|---|
| Live transcription | Enables the function for the authorized population | Toggle switch | Enabled |
| Retention | Chooses to keep the text with the recording | Toggle switch | According to the client's needs |
| Population | Chooses all users or a selection | All or selection of people, queues, and groups | A service queue |
| Exceptions | Excludes users | Multiple selection | Test account |
| Target language | Limits the offered languages | All or chosen languages | French and English |
A silence in the conversation no longer interrupts the transcription: segments without speech are simply skipped.
In "all" mode, only exceptions remain displayed among the population lists. In "selection" mode, choose the relevant people, queues, or groups. Save changes; Cancel restores the previous setting. The recordings view button opens their section in General Settings. Configuration alone is not a transcription: verify the result during an authorized call.
Screening and identification log
Open AI Integrations › Identify Unknown Callers. The page requires read rights on parameters and users; modifying screening requires write rights on users.
| Field | Purpose | Possible values | Example |
|---|---|---|---|
| Ask unknown callers to identify themselves | Enables the client's screening | Toggle switch | Enabled |
| Greeting sound | Chooses the client's shared announcement | Available sound or Standard announcement | Standard announcement |
| Screening language | Chooses the language of the flow | French or English | French |
| Transcribe the name (AI provider configured) | Enables transcription of the announced name | Toggle switch | Enabled after provider configuration |
Click Save. The unknown caller records their name; the called person presses 1 to accept, 2 to refuse, or 3 for voicemail. Without a response, the flow uses voicemail. If the AI voice is unavailable, the screen may indicate the use of the standard announcement. An unreadable sound library retains the saved value.
The View — identification log button opens the separate log. Select a client to read the events. The log requires read rights on users and refreshes every ten seconds. Each entry shows the caller, the internal number called, the decision, the date, and, if available, the transcribed name. The displayed decisions distinguish Pending, Accept, Reject, and Voicemail. The transcription status indicates processing in progress, a saved result, or an error.
Each entry shows the decision: Accept, Reject, Voicemail (including when the called person did not respond in time), or Caller hung up. To finish recording their name, the caller can also press the # key.
An empty list is not a read error. If an error appears, use Retry. The link to Call Logs is offered to accounts that can view them.
Text-to-speech
Open AI Integrations › Text-to-Speech. Configure the available assignments, then the Brand voice.
| Field | Purpose | Possible values | Example |
|---|---|---|---|
| Voice | Identifies the voice with the service | Compatible identifier; required to save the brand voice | Identifier provided by the provider |
| Language | Specifies the voice's language | Compatible code; required to save | fr |
| Text | Prepares a sound in the preview section | Non-empty text, 3,000 characters maximum | Hello and welcome |
Save saves the voice and its language; Cancel restores previous values. In the preview section, start generation with the provided button, then listen to the resulting sound. Generation is unavailable while loading, processing, or when the text is empty. An error remains displayed for correction. Generating a sound and saving the brand voice are two separate actions.
AI Assistant
Open AI Integrations › AI Assistant with write rights on AI. Wait for the availability check. An unavailable status or error does not constitute an answer to your question.
| Field | Purpose | Possible values | Example |
|---|---|---|---|
| Message | Asks the assistant a question | Non-empty text, 4,000 characters maximum | What settings should I check for this routing? |
Click Send, then wait. If the assistant proposes a change, review the preview, the values, and the expiration. Confirm requests its application; Cancel withdraws the proposal. An expired proposal can no longer be confirmed. While a proposal is pending, sending another question is blocked.
Clear all clears the displayed exchanges and the proposal. The screen indicates that conversations are not saved. In case of a confirmation failure, read the message and check the actual state of the configuration before making a new request.
Verifying the result
Reopen the modified settings and check the client, rights, assignments, and save results. Then verify the relevant usage: a recording's text, a voicemail message, a test conversation, or a screening call. Review AI outputs before using them.
If it doesn't work
An access denial is handled with the rights administrator. For a provider failure, the displayed message indicates the cause:
| Message | What to do |
|---|---|
| The AI provider's key is rejected | Renew the key in Client Providers or in the installation's AI settings, then test it |
| The AI provider's credit is exhausted | Top up the account with the provider or assign another provider |
| The AI provider is receiving too many requests | Try again in a moment |
| Your organization's AI quota has been reached | See Quotas and usage |
| The AI model is temporarily overloaded | Try again in a few moments; when the selected model is saturated or does not respond in time, a backup model automatically takes over |
| The AI provider is unreachable | Try again later; if this persists, check the PBX's network access to the service |
| The AI provider could not process this request | Check the assigned model and retry; if this persists, contact support |
Never transmit the key itself. Also check the function assignment and the processing policy. A missing result, an unavailable file, and a read error require different checks; do not mistake these states for successful processing.
