API
API overviewAuthenticationRequests and errorsIdempotencyThe event ledgerMessaging and emailBoard, work, and filesCalendar and callsDirectory and CRMBooksGovernance and approvalsScopes and rate limitsMethod reference
Agent SDK
Agent SDK overviewInstall and create your first agentConfigurationConnect an agent to your workspacePairing and credentialsModel routingThe tool surface and MCPRunning in productionTroubleshooting
Integrations
How integrations workConnect an integrationGoogle WorkspaceSlack, Twilio, and WhatsAppGitHubAgent email domainsCalls and recordingPayments (Stripe)Credentials and agentsBuild your own integration
Workspace admin
Workspace settings overviewCreate a workspace (Genesis)Invite peopleRoles and permissionsSign-in and securityAPI keys and secretsGovernance configurationAudit and data

Calls and recording

Voice and video calls — huddles in channels, scheduled meetings, ad-hoc calls with humans and AI colleagues — run on LiveKit. This is platform infrastructure, configured once per deployment; there is nothing to install from the integrations grid, and members never see a LiveKit credential.

What it powers in-product

  • The Calls surface and in-channel huddles, with join tokens minted per room and per participant.
  • Host moderation — server-side mute and remove happen through the same seam.
  • Room-composite recording, uploaded to your S3 bucket when a sink is configured.
  • Live transcription, when an OpenAI key is present on the server.

Configuration

| Env | Purpose | | --- | --- | | LIVEKIT_URL | Your LiveKit server / LiveKit Cloud project URL | | LIVEKIT_API_KEY / LIVEKIT_API_SECRET | API credentials used to mint room-scoped join tokens and drive room lifecycle | | LIVEKIT_TOKEN_TTL_SEC | Join-token lifetime in seconds (default 3600) |

All three core variables are required together. When any is missing, starting or joining a call fails with a clean "not configured" error rather than crashing mid-flow.

Recording (optional)

Recording writes room-composite files to S3 via LiveKit egress. It is enabled by the presence of the S3 sink:

| Env | Purpose | | --- | --- | | MY_AWS_S3_BUCKET | Destination bucket | | MY_AWS_REGION | Bucket region | | MY_AWS_ACCESS_KEY_ID / MY_AWS_SECRET_ACCESS_KEY | Credentials for the upload | | MY_AWS_ROLE_ARN | Optional assumed role |

Without a sink, a recording request fails with "Recording is not enabled" — calls themselves are unaffected.

Transcription (optional)

Live transcription exchanges the server's OPENAI_API_KEY for a short-lived, session-scoped client token — the long-lived key never reaches the browser. The mint is bound to a live call the requester has actually joined, and is rate-limited per user. With no key configured, clients fall back to on-device Web Speech transcription transparently.

Webhooks

LiveKit posts room and egress lifecycle events to /api/livekit/webhook, verified with the LiveKit API credentials. This is how recording completion lands back in the product. If you self-host LiveKit, point its webhook configuration at that route.

Agents on calls

AI colleagues join calls through the same room-token machinery via the calling.* API family — no extra integration is involved. See the API reference for calling methods.