Aeion Realtime vs Twilio Video, Daily.co, Vonage Video API
Twilio charges $0.00167/participant/minute for video. At 100 active users × 50 hours/month, that's $10,005/month — just in WebRTC infrastructure. Aeion Realtime is self-hosted on your own Hetzner VPS — zero per-minute fees, plus AI participants that take real actions via 89 MCP tools.
Head-to-Head Comparison
| Capability | Aeion Realtime | Twilio Video | Daily.co | Vonage Video API |
|---|---|---|---|---|
| **Pricing model** | Self-hosted (fixed VPS cost) | Per participant-minute | Per participant-minute | Per participant-minute |
| **100 users × 50 hrs/mo cost** | **$0 infrastructure** | ~$10,000/mo | ~$5,000/mo | ~$8,000/mo |
| **AI participants** | 89 MCP tools available to every agent | No | No | No |
| **TURN relay** | Self-hosted coturn (included) | $0.0004/min TURN | $0.0004/min TURN | $0.0004/min TURN |
| **Transcription** | Whisper.cpp free (or $0.006/min cloud) | $0.02/min (Voice Insights) | $0.02/min | $0.025/min |
| **Recording** | Included (VPS storage) | $0.003/min | $0.004/min | $0.005/min |
| **HIPAA compliance** | BAA-ready (self-hosted) | BAA available ($5K/mo extra) | BAA available | BAA available |
| **Self-hosted LiveKit** | ✅ Embedded (free) | ❌ No | ❌ No | ❌ No |
| **Compliance profiles** | HIPAA / PCI / GDPR-strict | HIPAA only (BAA extra) | HIPAA only | HIPAA only |
| **Multi-tenant isolation** | Per-tenant coturn allocation | Shared infrastructure | Shared infrastructure | Shared infrastructure |
| **Audit signing** | HMAC-SHA256 + 7-year retention | Basic logs | Basic logs | Basic logs |
| **Pre-built agents** | Review Producer, AI Tutor, Helpdesk Triage, Spatial NPC, Telehealth Scribe | No | No | No |
| **Session types** | meeting, review, match, scene, stream, support, telehealth | Video only | Video only | Video only |
| **BYO LiveKit cluster** | ✅ External LiveKit option | ❌ No | ❌ No | ❌ No |
The Per-Minute Billing Trap
The math that kills SaaS margins:
``` Twilio Video pricing: Video: $0.00167/participant/minute Audio: $0.0008/participant/minute TURN: $0.0004/participant/minute Recording: $0.003/minute Transcription: $0.02/minute
Scenario: 100 active users, 50 hours/month each
Total participant-minutes = 100 × 50 × 60 = 300,000/minutes/month Video: 300,000 × $0.00167 = $501/month TURN (30% of users): 90,000 × $0.0004 = $36/month Recording: 300,000 × $0.003 = $900/month Transcription: 300,000 × $0.02 = $6,000/month Voice Insights: $3,000/month (minimum)
Total: $10,437/month just in Twilio ```
That's $125,244/year. For the same infrastructure you could run a dedicated Hetzner AX82-NVMe (AMD Ryzen 9 7950X, 128GB RAM, 2× 2TB NVMe) at $89/month. Even adding 3× LiveKit nodes for horizontal scaling at $89 each = $356/month total. You save $124,888/year by self-hosting.
Aeion Realtime:
Hetzner AX82-NVMe: $89/month
LiveKit embedded: free (no per-minute fees)
Whisper.cpp transcription: free (local)
Recording: VPS storage (unlimited within disk)
Total: $89/month for unlimited participants
The VPS that runs Aeion OS already handles your web server, database, Redis, and background workers. Adding LiveKit to the same VPS adds zero hardware cost — only LiveKit configuration time.
Daily.co — Better Pricing, Same Fundamental Problem
Daily.co's strengths: Cleaner API than Twilio, better developer experience, reasonable pricing.
Daily.co pricing:
Video: $0.00125/participant/minute
Recording: $0.003/minute
Transcription: $0.02/minute
HIPAA: $2,500/month add-on
100 users × 50 hours/month with Daily.co:
Video: $3,750/month
Recording: $900/month
Transcription: $6,000/month
HIPAA: $2,500/month
Total: $13,150/month
Why "better pricing" is still bad: Whether it's $10K/month (Twilio) or $13K/month (Daily.co), you're still paying variable per-minute fees that scale with usage. Growth = higher bills. Success = higher bills. The financial model is fundamentally at odds with the product's value proposition.
Aeion Realtime: Fixed cost. Growth = same bill. Success = same bill.
Daily.co's other limitations:
- No AI participants
- No pre-built agents (Review Producer, AI Tutor, etc.)
- No multi-tenant isolation (shared infrastructure)
- No self-hosted LiveKit option
- No BYO TURN server
- Basic audit logging (no HMAC signing, no 7-year retention)
Daily.co is the best of the per-minute vendors. But "best of a bad model" is still a bad model for serious SaaS businesses.
The AI Participants Advantage — Beyond WebRTC
Twilio, Daily.co, and Vonage are video infrastructure. They provide the pipe — nothing more. You still need to build everything that happens inside the session.
Aeion Realtime is the full stack:
Universal room / peer / JWT session primitive
↓
Every session gets: cross-module attach, compliance profiles, participant tracking
↓
AI participants with summon / dismiss lifecycle
↓
LLM turn orchestration with 30s checkpointing
↓
Tool dispatcher routing 89 MCP tools
↓
Agent takes real actions inside the session
The difference in capability:
- Twilio: User joins video call. Someone takes notes manually. Notes are written after the call. Action items get lost.
- Aeion Realtime: User joins video call. AI agent joins simultaneously. Agent watches the conversation, takes notes in real-time, creates follow-up tasks in CRM, flags action items, sends summary after the call.
89 MCP tools means the agent is useful in every context:
- Collab meeting: agent creates notes, updates deal stage, sends follow-up email
- Helpdesk session: agent classifies issue, suggests KB articles, escalates if needed
- LMS class: agent answers questions, tracks comprehension, generates class report
- Telehealth: agent writes clinical notes, surfaces patient history, creates follow-up
This is not a chatbot in a sidebar. This is a full participant in the session that takes actions via the same tools your team uses.
Self-Hosted TURN — The 20-30% Connectivity Problem
20-30% of users are behind restrictive NATs. STUN alone doesn't work for them. They need TURN (relay). Twilio and Daily.co charge per-minute for TURN.
Coturn self-hosted (included in Aeion Realtime):
``` Listening: port 3478 (UDP + TCP) TLS: port 5349 (DTLS) Realm: aeion External IP: auto-detected
Throughput: 1Gbps on a modern VPS Cost: $0/month (coturn is open-source, runs on your VPS) ```
Circuit breaker behavior:
``` coturn crashes → supervisor restarts (up to 5 times/hour) After 5 crashes in 1 hour → circuit opens → Emits: realtime.coturn.circuit_opened → Clients fall back to STUN-only → Admin alert sent
Recovery: manual coturnctl restart OR auto after 1-hour cooldown (half-open test restart; returns to normal once stable 5+ min) ```
Cloudflare Calls fallback (optional):
bash
CF_TURN_KEY_ID=...
CF_TURN_API_TOKEN=...
When coturn is unavailable (cooldown period), Aeion can fall back to Cloudflare Calls TURN credentials. Short-lived credentials (1-hour TTL), no per-minute cost.
Per-tenant TURN override: any tenant with their own TURN infrastructure can configure it directly — once set, that tenant's sessions use your TURN server instead of the platform default, with no code required.
BYO LiveKit cluster: any tenant can point Realtime at their own externally hosted LiveKit instance instead of the embedded one — just supply your instance's API URL, WebSocket URL, and credentials in the admin panel.
Transcription — Whisper.cpp vs Cloud Providers
| Provider | Accuracy (clean) | Accuracy (noisy) | Diarization | Cost |
|---|---|---|---|---|
| Whisper.cpp | 85-90% | 70-80% | No | Free |
| Cloud Whisper | 92-95% | 85-90% | Yes (+$0.002/min) | $0.006/min |
| OpenAI Whisper | 90-93% | 82-87% | No | $0.006/min |
| AWS Transcribe | 88-92% | 80-85% | Yes | $0.024/min |
Compliance — HIPAA, PCI, GDPR-strict
Twilio HIPAA: BAA available at $5,000/month extra. Shared infrastructure with non-HIPAA tenants.
Daily.co HIPAA: $2,500/month add-on. Shared infrastructure.
Aeion Realtime HIPAA: set a session's compliance profile to "hipaa" and it's self-contained isolation from the start — encrypted, isolated recording storage; TURN traffic restricted to your own self-hosted relay (no third-party TURN); only BAA-covered AI providers permitted; 7-year (2555-day) retention; consent required before recording; every event audit-logged; end-to-end encryption available.
Key differences:
- Aeion Realtime: Self-hosted, per-tenant isolation, BYO storage backend
- Twilio/Daily.co: Shared infrastructure, vendor-controlled storage, per-tenant namespace only
GDPR-strict profile: EU-only recording storage with no US transfers; TURN traffic restricted to your self-hosted relay; only EU-data-resident AI providers permitted; 30-day default retention; consent required; every event audit-logged; all data stays in the EU; erasure requests processed immediately.
HMAC-SHA256 audit signing: every compliance event — including retention enforcement, like "HIPAA retention window expired: 1 recording deleted, 3 transcripts deleted" — is signed with HMAC-SHA256. The signing key rotates every 90 days (checked daily at 03:00 UTC), and rotated keys are kept for 7 years so old signatures remain verifiable.
Twilio/Daily.co don't offer this:
- HMAC-SHA256 signed audit logs
- 7-year key retention
- Per-compliance-profile enforcement
- GDPR-strict profile (EU residency, no US transfers)
- PCI compliance profile
The True Cost Comparison
| Cost Component | Aeion Realtime | Twilio Video | Daily.co | Vonage |
|---|---|---|---|---|
| **Infrastructure** | $89/mo (existing VPS) | $0 | $0 | $0 |
| **WebRTC (video)** | $0 | $3,750/mo | $3,000/mo | $4,000/mo |
| **TURN relay** | $0 | $36/mo | $36/mo | $50/mo |
| **Recording** | $0 | $900/mo | $1,200/mo | $1,500/mo |
| **Transcription** | $0 (Whisper.cpp) | $6,000/mo | $6,000/mo | $7,200/mo |
| **HIPAA BAA** | $0 | $5,000/mo | $2,500/mo | $3,000/mo |
| **100 users, 50 hrs/mo** | **$89/mo** | **$18,686/mo** | **$12,736/mo** | **$15,750/mo** |
| **Annual cost** | **$1,068/yr** | **$224,232/yr** | **$152,832/yr** | **$189,000/yr** |
| **AI participants** | **89 MCP tools** | ❌ None | ❌ None | ❌ None |
| **Pre-built agents** | **5 agents** | ❌ None | ❌ None | ❌ None |
When Twilio/Daily.co Make Sense
Use Twilio or Daily.co when:
- You need WebRTC in a pure client-side app with no backend
- You don't have access to a VPS or server infrastructure
- You need very fast time-to-market for a proof-of-concept
- Your volume is low (< 100 participant-hours/month)
- You need global TURN infrastructure without managing servers
Use Aeion Realtime when:
- Your volume is > 500 participant-hours/month (cost savings kick in hard)
- You need AI participants (Review Producer, Helpdesk Triage, etc.)
- You have compliance requirements (HIPAA, PCI, GDPR-strict)
- You need self-hosted infrastructure (data residency, sovereignty)
- You want fixed costs that don't scale with growth
- You want full control over the stack
The tipping point: At ~2,000 participant-hours/month, Aeion Realtime's fixed VPS cost ($89/month) becomes cheaper than Twilio's per-minute fees. For most SaaS businesses with 100+ active users, the tipping point is reached in the first month.
Migration from Twilio/Daily.co:
1. Export session history from Twilio/Daily
2. Configure coturn on your Hetzner VPS
3. Configure LiveKit (embedded or external)
4. Update client SDK: HybridTransport → Aeion Realtime
5. Migrate existing sessions (if needed)
The SDK migration is the only code change. Everything else is configuration.
Frequently Asked Questions
Yes for the common case — WebRTC under the hood, same TURN/STUN infrastructure available, same codecs (VP8/VP9/H264/AV1). Adaptive bitrate, simulcast, and SFU-based forwarding. Twilio / Daily have more globally-distributed POPs out of the box; you can deploy multiple Aeion Realtime nodes regionally to match if your audience is multi-region.
An AI agent joins the call as a participant — listens to audio, transcribes in real time, executes actions in Aeion (file a ticket, look up an order, generate a summary at end-of-call). Native to the platform; not a bot bridged through a webhook. Useful for sales-call note-takers, customer-support QA, meeting assistants, virtual concierge.
Aeion Bridge can host coturn or your provider's TURN; the SFU runs on your VPS (Hetzner, Contabo, AWS EC2, etc.). The orchestration is fully managed by Aeion — you provision a VPS, the platform configures + monitors it. No raw WebRTC ops needed.
SFU-based architecture supports 1000+ participants per room with subscribers receiving only the streams they care about. For 10K+ events / town halls, the architecture scales out across multiple SFU instances. Compared to Twilio's $0.00167/min/participant — at scale, your AWS bill for SFU CPU is the new cost floor.
Yes — SFU-side recording (per-participant or composed), real-time transcription via your chosen STT provider (Whisper, AssemblyAI, Deepgram, Azure Speech), post-call AI summarization via AeionClaw. Recording storage in your BYOB S3. Transcription cost optimized via provider routing.