Skip to content
Phone Guides

FreePBX AI integrations: Scribe and administration tools

For a dated comparison of the provider versions, audio and session limits, pricing and data terms cited across the AI guides, see the Asterisk AI provider contract checklist. Sangoma Scribe is treated there as a managed service because its public guide does not expose the underlying Deepgram API settings.

Sangoma Scribe is a commercial FreePBX/PBXact module for transcribing call recordings and voicemail, creating summaries and sentiment results, and presenting those records in FreePBX and Sangoma clients. Its documented workflow starts from an audio file. That is different from a live conversational agent that listens and speaks during a call.

Source-reviewed, 20 October 2026. This guide reports the scope in Sangoma's published Scribe documentation and the current Deepgram data-handling pages. Phone Guides has not installed Scribe, tested a FreePBX system, or sent caller audio to an AI service. The Scribe guide's last-updated date omits a year; confirm module and FreePBX version support with Sangoma before installation.

What Scribe documents

Sangoma's Scribe product page describes transcription, summaries, sentiment analysis, tags, language options and handling for recorded calls and voicemail. Its Scribe user guide gives more operational detail:

Workflow What the guide describes
Voicemail Transcription can be enabled for selected users or groups and shown in the voicemail interface or voicemail-to-email.
Call recordings Transcription can be enabled for selected users or groups; the module can show transcripts beside recordings and in reporting/client views.
Older recordings Administrators can select past voicemail or call-recording files for transcription.
Uploaded audio The module accepts audio uploads for transcription and related results.
Search and follow-up Tags, reports and email sharing help users locate and distribute transcription results.
Generated prompts The guide describes a text-to-speech feature for creating audio through System Recordings. This does not establish dynamic speech generation during a live caller interaction.
External systems The guide says Scribes can be obtained through the FreePBX GraphQL API for possible CRM or LLM integrations. That is transcript-data access; it is not a documented live-call audio interface.

The guide calls one screen a “real-time” playback view. It describes playing an existing recording while its transcript is displayed; it does not describe an agent responding to the caller in real time.

FreePBX does document a process-supervision API for background applications owned by a custom module. Sangoma's Process Management guide describes FreePBX::PM2() lifecycle calls, and the FreePBX 17 pm2 module source contains the versioned implementation. This identifies a place for a custom module to manage a persistent adapter process; it does not supply the AI adapter, ARI call routing, bridge/media handling or provider-session cleanup. See the source-reviewed live voice-agent design, which keeps those call-path claims unverified.

AI tools that administer FreePBX

Frogman is an independently maintained, experimental FreePBX module that exposes PBX administration functions to AI clients through MCP and other interfaces. Its repository describes FreePBX 17, Asterisk 20 or later, and PHP 8.2 or later as requirements. This is a management/control interface: it does not carry caller audio, implement a voice agent, or establish a supported live-call AI path. It is not a Sangoma-supported module or a tested Phone Guides deployment.

Source-reviewed, 20 October 2026. The requirements and features above are from the module maintainer's repository. Security claims below are from the maintainer's GitHub advisory and release record; Phone Guides has not installed, tested or independently audited the module.

Treat AI access to PBX administration as privileged access. The project README describes permission checks and confirmation for changes, but those are maintainer claims rather than an independent security review. Restrict who can reach its interface, grant the least privilege possible, review each enabled tool and keep an independent change/audit trail. Do not grant an AI client root-level fwconsole access or expose an unauthenticated management endpoint. Keep a tested backup and a human approval step for changes to routes, trunks, credentials, firewall rules or system services.

The maintainer's security advisory for CVE-2026-46512 reports dialplan injection through template parameters in versions through 1.6.0, potentially allowing arbitrary dialplan execution, and lists 1.6.2 as patched. The 1.6.2 release notes also state that the four v1.6.1 advisory patches remain in place. If assessing the module, use no release older than 1.6.2, inspect all later security notices and fixes, and confirm the exact installed version before enabling any write permission. A patched version alone is not a security review or a guarantee of safe AI-generated PBX changes.

This guide's dated provider contract checklist covers hosted voice-AI provider API, audio, lifecycle, pricing and data terms. Frogman has no voice-provider contract to compare there; its version, permissions and security advisories belong in the PBX deployment review.

Check product and module support first

Scribe is a commercial module and requires a valid licence, according to Sangoma's guide. That guide says the module was released in 2024 for FreePBX/PBXact versions 15, 16 and 17. Since the page's update stamp only shows “Aug 12” without a year, treat that list as the guide's stated range, not confirmation of the currently supported releases. Sangoma's licence guide lists a 1,000-minute monthly shared allowance and a 1,000-minute top-up valid for three months. The FreePBX product page lists $24.99 per month for 1,000 minutes but does not state the currency or top-up price. Check the module entry, release notes and current portal price for the exact FreePBX/PBXact release before planning a change; the dated provider checklist records these plan terms and the remaining unknowns.

The guide describes installation through Module Admin or the PBX command line, then enabling voicemail and call-recording workflows at user or group level. The same manual also documents uploading audio for transcription, reprocessing older recordings and voicemail, and generating system-recording audio from text for IVRs and announcements. It describes these as licensed Scribe features, not as a live conversational agent or streaming call interface. See the dated AI provider contract checklist for the public Scribe facts and the unresolved API, audio-limit, price and data terms. Follow the matching current module instructions and take a configuration backup first. Do not replace generated FreePBX files with standalone Asterisk examples; see Asterisk and FreePBX configuration boundaries.

Pay attention to the voicemail email setting. Sangoma says enabling Scribe's voicemail-to-email option changes the PBX mail command and can overwrite an existing custom mail command. Record the current value and its owner before changing that option, and verify the resulting email on a non-sensitive test mailbox.

Know where call data goes

The Scribe guide says FreePBX sends recording files to its partner over HTTPS for transcription or analysis and names Deepgram as its transcription and sentiment partner. HTTPS protects the connection in transit; it does not answer where the module sends data, how long the partner retains it, who can read transcripts, or which model-improvement settings apply to this managed integration.

Deepgram's current data and privacy guide says the endpoint determines processing geography. Its global endpoint is the default; its Australian endpoint is api.au.deepgram.com. The guide also says requests are included in model improvement by default and describes mip_opt_out=true for zero retention of audio, text and transcripts after processing. These are Deepgram API controls. Sangoma's public Scribe guide does not establish that a FreePBX administrator can choose the endpoint or pass the opt-out setting through Scribe. Do not assume Scribe uses Australian processing or zero retention because those options exist in Deepgram's direct API.

Before enabling Scribe on caller recordings, ask Sangoma to confirm the module's current processing region, audio and transcript retention, deletion process, model-improvement setting, subprocessors, and whether those settings can be selected by the customer. Review current terms for your account and intended use. Limit transcript and recording access to the users who need it; transcripts may contain the same sensitive details as the original audio. Check the separate handling of transcripts emailed or exported to another system.

Sangoma's published Data Processing Agreement, version 3.1 dated 11 October 2024, lists sub-processors for named services, but its Annex F does not name Scribe or FreePBX in the applicable-services column. That document does not identify Scribe's provider or region; use the terms that apply to the specific Scribe account and deployment. See the dated AI provider checklist for the current source review and open questions.

Treat summaries and sentiment as aids

Transcription can mishear names, numbers, negation, accents, cross-talk and specialist terms. Summaries can omit qualifications, and sentiment labels are model-generated classifications rather than proof of a caller's state or intent. Review the recording before using an AI result as evidence for a complaint, staff assessment, financial decision or service action. Keep a path for correcting or disputing a transcript.

For a language other than English, the Scribe guide says transcription supports multiple languages but summaries and sentiment are not supported for languages other than English. Check the current module guide and test the exact language and recording conditions before relying on those outputs.

Plan a safe first check

  1. Confirm the PBX release, Scribe module release, licence and current support statement with Sangoma.
  2. Start with a permitted, non-sensitive recording and a small test user group. Confirm the source recording, transcript, summary, sentiment result and access controls separately.
  3. Inspect voicemail email settings before enabling Scribe's mail integration. Send a test message and verify recipients and contents.
  4. Record which provider region and retention settings Sangoma confirms for this module. If the settings cannot be selected or documented, treat the data path as unresolved and decide whether the module fits the use case.
  5. Confirm where exported or emailed transcripts go, who retains them and how access is removed when a user leaves.

For a live AI caller interaction, use the separate Asterisk external media and speech guide as a starting point. It covers Asterisk media paths and an AEAP speech example; it does not establish a FreePBX-to-live-agent recipe. Do not use Scribe's transcript export, recording playback, upload, rewind, or system-recording text-to-speech feature as evidence that it supplies a live agent.

Sources and limits

These sources document a managed recorded-audio product and a separate experimental PBX administration module. They do not establish Phone Guides execution, exact current module support beyond the listed vendor guide, the Scribe account's region or retention configuration, or a supported live AI call path.