AI agents (Famulor)

Identifies the agent in Cockpit lists and on Famulor (auto-prefixed with your PBX tag).

Whether this agent answers inbound calls or places outbound calls.

Governs how the assistant interprets time references ("today", "3pm"), scheduling logic, and call log timestamps.

Internal note about this agent - visible and editable from the list view.

The speech-processing engine - the most consequential field on the form.

Primary language for speech recognition, available voices, filler phrases, and model selection.

Additional languages the assistant understands and responds in - useful for multilingual callers who switch language mid-call.

Text-to-speech provider for the assistant's voice. Premium providers offer higher quality, fast providers offer lower latency.

Voice the assistant speaks with. Green-highlighted voices match the primary language for best quality. Use Preview to listen before choosing.

Optional background noise mixed under the voice to make conversations feel more natural. Disable if speech recognition is affected.

Volume of the ambient sound from 0.0 (silent) to 1.0 (loudest). Start low.

Adjusts tone and emphasis based on conversation context to sound more natural and empathetic.

Consistency of voice across the call. Lower for livelier speech, higher for calmer, even speech. Range 0.0-1.0.

How closely output matches the source voice recording. Most relevant for cloned voices. Range 0.0-1.0.

How fast the assistant speaks. Slower (0.7-0.9) for complex content, faster (1.1-1.2) for short operational calls. Range 0.7-1.2, default 1.0.

Speech-to-text provider for transcribing the caller's voice. Choose accuracy vs latency based on your use case. Pipeline mode only.

How the assistant detects when the caller has finished speaking.

Pause after the last word before the assistant responds. Lower = faster, higher = more patience for natural pauses. Range 0-5, default 0.5.

Filters the caller's background noise for clearer transcription. Disable if voices sound cut off or distorted.

AI language model powering the assistant's responses. Fast models suit high-volume calls, reasoning models suit complex decisions.

How the SIP trunk is identified on Famulor: by extension or by a direct E.164 phone number.

The extension callers dial to reach this agent. Only unassigned extensions are listed.

E.164 DID number (e.g. +14155551234). Only used when trunk type is Phone number (DID).

Carrier whose credentials are used to create the SIP trunk on Famulor. Must already be configured on this PBX.

ISO 3166-2 country code for the SIP trunk on Famulor. Determines carrier routing and rate plans.

Number format for outbound calls on the wire. Match the format your SIP carrier requires.

Short sounds ("mhm", "okay") played while the AI processes a response, preventing unnatural silence. Pipeline mode only.

How varied the responses are. Low (0.0-0.3) for stable, predictable business calls. High (0.7-1.0) for more creative conversations. Default 0.1.

When on, the assistant stops speaking the moment the caller starts. Cannot be disabled in Speech-to-speech and Dualplex modes.

How easily the assistant stops when interrupted. Lower = harder to interrupt, higher = more responsive. Range 0-5, default 0.5.

Minimum caller words to register as an interruption, preventing background noise from cutting off the assistant. Range 0-10. Leave empty to disable.

Seconds of caller silence before the assistant prompts again (e.g. "Are you still there?"). Range 7-600, default 8.

Custom re-engagement text. Leave empty to use the language default.

Hard limit on call length. The call ends automatically at this point. Range 20-1200, default 300 (5 minutes).

Call ends if the caller is silent for this long. Range 1-360, default 30.

How long the phone rings before the call is marked unanswered. Keep low to avoid hitting voicemail on outbound calls. Range 1-60, default 30.

Ends the call automatically when an answering machine is detected, avoiding unnecessary AI processing costs. Recommended for outbound agents.

Records calls for quality and analysis. Ensure compliance with local consent regulations before enabling.

Ends the call if nobody speaks at the start - prevents the agent talking to voicemail. Leave empty to disable. Range 1-120.

Message left when the agent reaches voicemail (e.g. company name + callback request). Leave empty to hang up silently. Max 1000 characters.

Define when the AI assistant should transfer a call to a real person. Each rule specifies a trigger condition and a destination.

Warm transfers keep the caller on hold while the assistant briefs a supervisor, then connects the two. Use this when the supervisor needs context before taking the call.

Let the assistant hang up on its own when the conversation is complete. Describe the exact conditions so it neither lingers nor cuts the caller off too early.

Instructions for the assistant describing the conditions under which it should end the call.

Let the assistant book meetings on your calendar during the call. Add one entry per calendar - describe when each should be used.

When on, the assistant waits silently after connect for the customer to speak first. Use for inbound support and reception scenarios.

Opening line the assistant says when answering. Keep brief, introduce purpose, and support {variable_name} tokens. Max 200 characters.

Defines the assistant's role, goals, and rules. Built-in templates are available via the ? icon. Be specific and concise - clear rules outperform long essays.

Pre-call data injected into the prompt and initial message via {variable_name} tokens - enables personalization (customer name, account number, etc.).

When on, the AI analyzes the transcript after each call and extracts structured data for the webhook payload.

Variables the AI extracts from each transcript. Each has a name (webhook key), type (String / True-False / Number), and a description guiding extraction.

Sends extracted call data to an external URL after each call. Use the magic-wand button to auto-fill the Cockpit webhook URL.

Endpoint that receives the post-call payload (extracted variables, transcript, optionally the recording URL).

When on, skips the webhook for missed, abandoned, or timed-out calls - only fully completed calls trigger it.

Adds the call recording URL to the webhook payload. Requires Record calls to be enabled.