Voice Calls

Voice calls in Saltare are real-time audio conversations attached to channels. They use LiveKit for WebRTC audio, and — uniquely — AI agents can join calls, listen via transcription, and respond with synthesized speech.

Starting a Call

To start a voice call, click the phone icon in any channel header. When you initiate a call:

1. A call room is created and you join immediately
2. All other channel members receive an incoming call notification
3. A system message is posted in the channel announcing the call
4. Members have 30 seconds to join before the call times out (if you're the only participant)

Joining and Declining

When someone starts a call in a channel you belong to, you'll see a ringing indicator with options:

• Accept — Join the call immediately
• Decline — Dismiss the notification (the caller sees you declined)

You can also join an active call at any time by clicking the call indicator in the channel.

During a Call

Controls

• Mute / Unmute — Toggle your microphone
• Leave — Exit the call (the call continues for other participants)
• End Call — End the call for everyone (only the initiator can do this)

Call Timer

A timer shows how long the call has been active, displayed in the call panel.

Participant List

See who's on the call with speaking indicators that show who's currently talking.

Adding AI Agents

The most powerful feature of Saltare voice calls is agent participation. You can add any workspace agent to an active call.

How It Works

1. Click "Add Agent" during an active call
2. Select an agent from your workspace
3. The agent joins and begins listening via speech transcription

What Happens Behind the Scenes

• Your speech is transcribed using the Web Speech API in your browser
• Transcriptions are sent to the agent in real-time
• The agent processes your words, selects tools, and generates a response
• The response is converted to audio via text-to-speech
• Audio is played back to all call participants

Use Cases

• Quick research — "Hey @Researcher, what's Acme Corp's latest funding round?"
• Live task creation — "Create a task for the items we just discussed"
• Meeting notes — Agents listen and can summarize the conversation afterward
• Data lookup — "What does our sales database show for Q2?"

Removing Agents

Click "Remove Agent" to disconnect an agent from the call. Transcription stops for that agent.

Call Lifecycle

RINGING  → Waiting for participants (30s timeout)
ACTIVE   → Call in progress with connected participants
ENDED    → Call terminated (by initiator or last human leaving)

Important behaviors:
- If the last human participant leaves, the call ends automatically (agents alone don't keep a call alive)
- The initiator can end the call for everyone at any time
- Calls that go unanswered for 30 seconds are marked as missed

Post-Call Processing

After a call ends, Saltare can generate:

Transcripts

A timestamped record of everything said during the call, attributed to each speaker. Transcripts are stored as documents for future reference.

Summaries

Use the summarize_voice_call agent tool to get a structured summary of what was discussed:

@Assistant summarize the voice call we just had in #engineering

The agent pulls the transcript and produces a summary with key decisions, action items, and topics covered.

System Messages

Voice calls post system messages in the channel:
- When a call starts
- When participants join or leave
- When the call ends (with duration)

This keeps an audit trail of calls in the channel history.

Plan Requirements

Voice calls require the Pro plan or higher:
- Voice Calls — Available on Pro, Business, and Enterprise
- Agent Participation — Also requires Pro or higher

Starter plan users will see an upgrade prompt when trying to start a call.

Tips

• Use agents for meeting notes — Add an agent at the start of important calls so it captures everything
• Keep calls in relevant channels — Calls are attached to channels, so start calls in the channel that matches the topic
• Review transcripts — Post-call transcripts are valuable for team members who couldn't attend
• Quick questions — Voice calls with agents are faster than typing for complex, conversational queries