Gemini Live PBX
Voice operator console
Workspace Dashboard
Idle

Operations dashboard

Live snapshot across all voice agents · last 7 days

Calls · 7d
0
Avg duration
--
Success rate
--
Active now
0

Campaign

Idle
Active / concurrency0 / 0
Completed / total leads0 / 0
WhatsApp sent0
Campaign spend$0.000

Calls by hour

UTC · 7d
0006121823

Agents

at a glance · click to configure
Loading agents…
Voice Agent Sandbox. Connects directly using WebRTC microphone capture. Click Start Session below, authorize your mic, and test.

Estimated Session Cost

$0.000000
Accumulated usage metadata returned by Gemini Live.

Token Usage Meter

Text input 0
Audio input 0
Text output 0
Audio output 0
Total Session 0
Active / Concurrency 0 / 0
Completed / Total Leads 0 / 0
Campaign Cost $0.000
Run as
Quick Single Call:
Leads List & Outcomes
Lead Company Phone Agent Status Duration Cost WhatsApp Actions
No leads uploaded yet. Click Upload Leads to add contacts.
Calls (window) 0
Avg duration --
Active right now 0

Calls by hour of day (UTC)

0006121823
Inbound Outbound All
24h 7d 30d All time
Call log
Date Time Agent Phone Direction Duration Status Cost Actions
Loading…

Team & limits

Members, daily spend caps, and the audit trail · resets at midnight Europe/Chisinau

Members

spend today vs limit ·
MemberSpend todayDaily limitSessionsActions
Loading…

Account limits

apply to everyone
Off = unlimited. A member set to $0 is paused (no new calls or sandbox). Active calls always finish; caps only block new sessions.

Spend by agent

today
Loading…

Audit trail

logins · edits · calls · limit events
Loading…

Agent settings

Identity, conversation behaviour, tools and post-call handling — per agent.

Identity & voice

Model, voice, opening line, telephony routing & languages.
Strongly enforced via the system prompt. ~95% verbatim adherence in practice.
Optional. Must be unique across agents.
The first ticked language is used for the opening greeting. Drives TTS language pinning, STT bias, and a constraint in the system prompt.

Persona & conversation

System prompt, pace, sampling, reasoning & native-audio behaviour.
Affective dialog Requires a native-audio model (auto-routes to v1alpha API).
Proactive audio Requires a native-audio model (auto-routes to v1alpha API).

Turn-taking

Voice-activity detection, sensitivity & barge-in.
ms
ms
Allow caller to interrupt (barge-in) Turning this off forces the caller to wait for every line.

Tools & knowledge

Built-in tools, custom HTTP webhooks & the knowledge base.
Google Search Adds an extra 100–300 ms when the agent uses it. Skip for cold-call agents that just need to talk.
end_call is always available; no toggle. The agent's Knowledge Base ({{kb}}) is edited via the Knowledge button above.
Tool name + description should be precise enough that Gemini knows when to call it.

Post-call

Extraction, success criteria, webhook delivery & recording.
After every call, a single Gemini text completion extracts these data fields and evaluates these criteria against the transcript. Results land on the lead and inside the post-call webhook payload.
Webhook fires for every terminal call state (Completed / Failed / No Answer). Payload includes transcript, duration, cost, extracted data, evaluations, and (when enabled) an audio URL. Retries 3 times on non-2xx (5s, 30s, 5min).
Leave empty to disable post-call delivery for this agent.
Sent in X-Webhook-Signature: sha256=<hmac>. Verify on your side.
Include audio recording URL When ON, the webhook carries audio_url pointing at the mixed call recording. Requires public_base_url set in Workspace settings.
Record calls for this agent Writes a stereo WAV per call (L = caller, R = agent) to recordings/{uuid}.wav. Independent of webhook delivery. Cleaned up after retention window expires.

Workspace & globals

Concurrency, recording retention & circuit breaker — shared by all agents.