VoiceThere is a cloud platform for real-time apps built on WebRTC. You ship TypeScript agent logic; we run it on managed agent runners, handle signaling and sessions, and roll out deploys when you promote a build.

Two common shapes of app on the same stack:

Develop locally with open-source WebRTC tooling, then upload → promote → deploy from the dashboard or CLI.

Status: Public launch on voicethere.cloud is in progress. The operator preview at app.voicethere.dev is the same product direction on our staging environment.

Define agents with @voicethere/agent: greet on join, respond to finalized speech, handle barge-in, and stream TTS through the parent runtime. Verify bundles locally before every upload:

npm install @voicethere/agent
npx @voicethere/agent verify

Connect end users with @voicethere/client in the browser or Node for full WebRTC roundtrips against your deployed runners.

Run several clients in one session with shared or per-peer agent logic:

• Shared session agent — one sandboxed process serves many connections; use sessionId to target speak, sendToClient, and broadcastToClients.
• Data-only mode — turn off voice pipelines for DataChannel-first apps (games, live state sync, chat) without STT/TTS.
• Per-peer agents — one isolated agent process per session when you need strong separation.

See the game-sync template in voicethere/agent for a data-only authoritative server example.

flowchart LR
  subgraph you [Your repo]
    A["agent.ts → dist/agent.js"]
    B["Your app backend"]
  end

  subgraph vt [VoiceThere]
    C["VoiceThere API"]
    R["Agent runners"]
    S["Signaling & WebRTC"]
  end

  subgraph clients [End users]
    F["Browser / Node client"]
  end

  A -->|"upload & promote"| C
  C -->|"deploy"| R
  B -->|"start session"| C
  F -->|"join"| S
  S <-->|"media & DataChannel"| R
  R <-->|"your agent logic"| A

1. Build a sandboxed agent bundle with @voicethere/agent.
2. Upload and promote with @voicethere/cli or the dashboard.
3. Deploy so VoiceThere rolls your active build out to agent runners.
4. Start sessions from your backend; clients join with @voicethere/client.

You never manage servers, clusters, or signaling infrastructure — that stays on VoiceThere.

These npm packages are the main developer surface:

CLI quick start:

npm install -g @voicethere/cli

voicethere login --api-key "$VOICETHERE_API_KEY"
voicethere projects create --name my-agent
voicethere build upload
voicethere build promote <buildId>
voicethere deploy --wait

Default API base: https://app.voicethere.dev/api/v1

Other services that power the hosted product are not open source. Use the repos above for SDK issues and contributions.

Available on app.voicethere.dev today (more before public launch):

• Dashboard sign-in, projects, build history, promote and rollback
• Deploy to cloud — roll your active build out to agent runners
• Voice sessions — start a session from your backend; clients get a join token and signaling URL
• Dashboard chat and embed for testing agents
• Multiplayer and data-only sessions — shared agents, session settings, idle timeouts
• Session history, usage metrics, structured session errors
• Early developer docs at /docs

Before GA: expanded documentation, managed TURN/ICE in session mint responses, live session diagnostics, subscription plans, and production regions on voicethere.cloud.

Local development uses the node-webrtc-rust library (separate OSS project):

Build and test on your machine; deploy the same agent bundle to VoiceThere when you are ready.

• Try staging — sign in at app.voicethere.dev, create a project, upload an echo agent, and chat from the dashboard.
• Issues and PRs — open them on agent, cli, or client depending on what you are changing.
• Releases — watch this org and @voicethere/* on npm.