Browser SDK, widgets, Go/C#/Python/Node workers
Own your real-time platform. Voice. Video. AI.
Build once. Deploy anywhere. Own the platform.
Self Hosted Private Cloud Public Cloud Hybrid
Control Plane
Tenant-scoped rooms, JWT/OIDC auth, audit hash chain, admin API, SIEM export.
Media Path
Mesh and SFU topologies, simulcast/SVC, first-party TURN with metering.
Voice & AI Workers
Service participants join rooms like users — STT, moderation, assistants, TTS publish-back.
Observability
Live quality scoring, topology graphs, redacted support bundles, OpenTelemetry.
Provenance
30 years building systems. 10+ years of WebRTC. Production at scale.
Independent new work by Nathaniel Currier — architect of Temasys WebRTC solutions, former CTO of Temasys. Healthcare. Government. Media. Enterprise.
Platform Architecture
Multi-tenant session control, signaling, and adapter boundaries shaped by years of customer integration.
Media & Relay
Mesh/SFU topology policy, TURN metering, and ICE credential design from production relay operations.
SDK & Developer UX
Browser and worker client surfaces informed by SDK adoption, debugging, and onboarding in the field.
Clean Implementation
Ground-up TypeScript monorepo — not derived from Temasys source, private repos, or customer-specific IP.
Note
Note Slide context
Authoritative statement:
PROVENANCE.md at the repository root. TomatoRTC does not contain prior Temasys proprietary implementation material or confidential customer IP.The challenge
Hosted RTC is fast to demo — expensive to own at scale.
$14.9B CPaaS market (2025)
Spend shifting from legacy telco to embeddable APIs.
~39% CAGR · Voice AI agents
Fastest-growing segment inside conversational AI.
Note
Note Slide context
Market figures: CPaaS ~$14.9B (2025) — Metrigy; WebRTC ~$12.6B (2025) — Mordor Intelligence; conversational AI ~$14.5B (2025) — The Insight Partners; voice AI agents ~$2.5B (2025) — Grand View Research. Figures are directional for sizing conversations, not TomatoRTC revenue claims.
The solution
One protocol from prototype to enterprise rollout.
Prototype 1
Browser SDK, Node workers, chat, data channels, and TURN relay out of the box.
Pilot 2
Mesh and SFU topologies, regional redirects, cascaded SFU links, bandwidth-aware TURN.
Production 3
Tenant isolation, RS256/JWKS, short-lived credentials, audit trails, E2EE transform foundations.
Enterprise 4
OIDC, storage, telemetry, effects, and AI workers through swappable adapters — your servers, your rules.
AI voice in the room
Real-time AI. Not batch processing.
TomatoRTC workers join with role: "service" — subscribe to human audio tracks inside the SFU/mesh path, run your STT/LLM/TTS adapters, and publish structured WorkerOutput events back into the room.
Speech → Understanding
Response → Audio Generation
Structured room output
- Human participants publish live audio into the room over mesh or SFU paths.
- A service worker joins the same room via WorkerRoomClient with role service.
- The worker subscribes to remote audio, runs VAD turn detection, then STT, LLM, and TTS adapters.
- Structured WorkerOutput events and optional synthesized audio publish back into the room.
Note
Note Slide context
Voice AI agents market ~$2.5B (2025) → ~$35B (2033), ~39% CAGR — Grand View Research. Browser STT/VAD capabilities vary by platform; validate via kitchen-sink AI perception demo before customer commitments. Worker STT/TTS adapters are BYO provider — TomatoRTC ships the room subscription and output wiring.
AI voice · integration surfaces
Four hooks from browser perception to room output.
Browser VAD + Perception
@rtc-sdk/client-ai — client-side voice activity and perception events on supported browsers.
Per-Room Supervisor
AiWorkerSupervisor spawns workers via POST /api/rooms/:roomId/ai-workers.
Transcription Adapter
Streaming STT from remote audio tracks — wire Whisper, Deepgram, or your stack.
Voice-Out Publish
Workers can publish synthesized audio tracks back — agent speaks in the same room mix.
Platform
Modular stack. Production SFU today. Deeper media ownership tomorrow.
Session coordination stays in the orchestration server; media engines, auth, storage, and workers plug in through adapters.
client-ai VAD, WorkerAdapter STT/LLM/TTS, supervisor API
Control plane Orchestration Server Rooms · Auth · Audit · Admin · AI workers
Production SFU · First-Party SFU (Hardening) · TURN
OIDC/JWKS · Postgres/Mongo · OTel · SIEM
Room state, auth, audit, admin
Media routing, relay, storage, telemetry
Note
Note Slide context
† First-party integrated SFU, media-plane RTP termination, and related cross-browser production hardening are active engineering work — available for evaluation and pilot deployments; see technical deck and production-readiness docs for current status.
Positioning
Enterprise capabilities. Self-hosted economics.
Closest to LiveKit self-hosted — stronger tenant isolation, audit trail, and voice-AI room model. Best for B2B products that already have a backend team.
Capability
TomatoRTC
Typical Hosted
Self-host
✓
✕
AI Runtime in room
✓
Partial
Data ownership
✓
Limited
Cost predictability
✓
✕
Tenant isolation
✓
Shared
BYO AI providers
✓
Locked
Positioning · ICP
Who TomatoRTC is built for — and who should look elsewhere.
Great fit
- ✓ B2B SaaS adding realtime + voice AI to an existing product
- ✓ Teams with TypeScript/Node strength and DevOps capacity
- ✓ Regulated or cost-at-scale buyers (healthcare, fintech, edtech)
- ✓ Teams outgrowing Twilio Video's narrowed vertical focus or Vonage's Ericsson-owned managed model
- ✓ Platforms needing custom tenancy, audit, or BYO AI providers
Look elsewhere
- ✗ Mobile-only MVP with no backend team
- ✗ Need vendor HIPAA BAA on week one
- ✗ Want global edge with zero infrastructure work
- ✗ Need composited recording + RTMP streaming out of the box
- ✗ Want a turnkey voice-agent SaaS with no room integration
Commercial model
Predictable licensing. Predictable infrastructure. Predictable scaling.
Self-Managed
Your infra + platform license
- ✓ Annual or term license for the platform source and release train
- ✓ You operate signaling, SFU, TURN, and AI workers on your cloud
- ✓ Pay VMs, bandwidth, and ops — no per-minute platform tax
Managed Solution
Fixed cost usage + management fee
- ✓ TomatoRTC operates regional clusters on your behalf
- ✓ Monthly fee = actual infrastructure usage + predictable management fee
- ✓ Includes upgrades, monitoring hooks, and runbook-aligned operations
Commercial model · TCO
Illustrative crossover vs hosted per-minute RTC.
Illustrative 36-month TCO — hosted per-minute vs self-managed license + infra. Actual crossover depends on usage profile; not a guarantee.
Note
Note Slide context
Pricing is quoted per deployment profile (participant scale, regions, TURN volume, AI worker concurrency, support tier). Optional Compliance package (+$35k/yr SOC 2 evidence kit) — see certification readiness deck. Contact sales for a scoped proposal.
Evaluation
Production tested. Enterprise proven. Prove it in an afternoon.
Technical buyers can validate signaling, media, TURN relay, voice workers, and diagnostics before procurement. The kitchen-sink demo is the fastest path to conviction.
Kitchen Sink Demo
Two-tab calls, SFU probe, AI perception, topology, chaos lab — full-stack on localhost.
Worker Examples Code
Node bot, Python listeners, transcription/moderation/assistant service participants.
AI Worker API REST
Per-room supervisor — spawn STT/assistant workers from admin console or REST.
Docs & Gap Analysis Docs
Production-readiness labels, platform gap analysis, and the full technical deck.
Start with a technical eval. Put voice AI in the room on your terms.
Run the kitchen-sink demo, review the technical deck, then scope self-managed licensing or a managed deployment.
- 1 Technical Evaluation
Run the kitchen-sink demo locally — full-stack signaling, SFU, TURN, and AI perception in one session.
- 2 Architecture Review
Review the technical deck — understand the room model, AI worker integration, and SDK surfaces.
- 3 Proof of Concept
Deploy a room with a live voice AI worker. Validate latency, quality, and your BYO AI providers.
- 4 Commercial Discussion
Scope licensing or managed solution pricing for your regions, participant scale, and support tier.
Own the UX Own the protocol Own the AI boundary