MediaSFU infrastructure
Media transport, agent runtime, recording output, and the managed SFU services shown in our published rates.
Published rates, explained
Know what you pay before you scale.
Start with the MediaSFU unit your workload actually uses, add only the providers you choose, and see when automatic volume discounts change the rate.
$0supported provider markup
2.5M+minutes unlock 20% off automatically
7.5M+minutes unlock 40% off automatically
10 SDKsone runtime across web, mobile, native, and Unity
Bill anatomy
Separate the platform from the providers.
A useful estimate names every layer. MediaSFU publishes its layer and lets you keep control of the rest.
Your chosen providers
AI, SIP, phone-number, storage, and optional provider charges stay direct. MediaSFU does not add a provider markup.
Your workload
Minutes, media quality, agent usage, recordings, and concurrency determine which published units apply.
Two meters. You pay only for media that actually flows.
Every call, meeting, agent and recording reduces to streams moving for minutes. Where a stream goes decides which meter it runs on.
Media between people
Media routed between participants: each producer × the people consuming it. Nobody consumes their own stream.
- Audio
- $0.10/1K min
- SD video
- $0.75/1K min
- HD video
- $1.50/1K min
Media into MediaSFU servers
Any stream that ends at, or is processed on, our servers: recording, an AI agent listening, noise cancellation, transcription feeds.
- Audio
- $2/1K min
- SD video
- $6/1K min
- HD video
- $12/1K min
No flow, no charge.Muted cameras, tiles off screen and idle rooms move no media, so they bill nothing.
AI voice agent call
One caller, one agent, 10 minutes of audio
- Caller's voice, consumed by the agent server1 × 1 × 10 min × $0.002$0.02
- Agent's voice, consumed by the caller1 × 1 × 10 min × $0.0001$0.001
Per call$0.0211,000 calls a month$21/moYour STT, LLM and TTS providers bill you directly.
1:1 video consult
Two people, HD camera both ways, 30 minutes
- 2 producers × 1 other person each2 × 1 × 30 min × $0.0015$0.09
Per consult$0.09500 consults a month$45/moNothing reaches the server as a sink, so only meeting minutes apply.
Team standup
5 people on SD video, everyone on camera, 15 minutes
- 5 producers × 4 others each5 × 4 × 15 min × $0.00075$0.225
Per standup$0.225Every workday$4.95/moA camera that is off is not a producer, so it adds nothing.
Recorded podcast
Two hosts, audio, 60 minutes, each track recorded
- 2 producers × 1 other host each2 × 1 × 60 min × $0.0001$0.012
- Both tracks, consumed by the recorder2 × 1 × 60 min × $0.002$0.24
Per episode$0.252Weekly$1.01/moThe first composed recording is included.
Rates at a glance
What 10,000 minutes costs on each path.
One volume, three paths, so the differences stay concrete. Taxes and external-provider costs are not included.
Plain phone or browser audio
10,000 audio transport minutes
$1.00 MediaSFU media total
Use this path for a second number, SIP calling, or browser audio without an AI agent.
- $0.0001 per minute
- Carrier or number charges remain direct
- No AI-provider cost
Audio AI agent
10,000 AI-ready audio minutes
$21.00 Media + agent infrastructure
MediaSFU handles the real-time audio and agent runtime. Bring your own STT, LLM, and TTS providers.
- $0.0021 per minute
- $1 media + $20 agent infrastructure
- AI-provider usage remains direct
Multimodal vision agent
10,000 video/vision minutes
$33.75 Media + vision agent infrastructure
Use this model for a video agent that can see, hear, reason, and respond in real time, streaming vision at QnHD (320×180), the agent-room default.
- $0.003375 per minute
- $3.75 QnHD media + $30 vision runtime
- Model-provider usage remains direct
Published base rates
Start with the rate that matches the job.
Per-minute infrastructure rates before automatic volume discounts.
| Service | Media only / min | AI-ready total / min | When to use it |
|---|---|---|---|
| Phone call (telephony) | $0.0001 | $0.0021 | Use the media-only rate for human calls; add agent infrastructure only when AI is active. |
| Web/mobile audio only | $0.0001 | $0.0021 | The same transparent audio unit applies to browser and mobile voice workloads. |
| Web/mobile video or multimodal, from QnHD | $0.000375 | $0.003375 | QnHD (320×180) media plus the vision-capable agent runtime; SD ($0.00075) and higher qualities use the matrix below. |
Automatic volume rates
The threshold and the resulting rate belong together.
Published base rateFrom the first billable minute
2.5M+ min/month: 20% offAudio media $0.00008 - audio AI-ready $0.00168
7.5M+ min/month: 40% offAudio media $0.00006 - audio AI-ready $0.00126
Output quality
Live media and retained recordings are different units.
Rates below are per 1,000 minutes and make resolution tradeoffs visible.
| Output | Live media / 1K min | Recording / 1K min |
|---|---|---|
| Audio only | $0.10 | $2.00 |
| QnHD (320x180) | $0.375 | $3.00 |
| SD (640x360) | $0.75 | $6.00 |
| HD (1280x720) | $1.50 | $12.00 |
| FHD (1920x1080) | $2.25 | $15.00 |
| QHD (2560x1440) | $3.00 | $18.00 |
Procurement checklist
Compare like with like.
Use the same workload and scope on every vendor page.
Before accepting a savings claim
- Compare the same minute definition: participant media, agent runtime, or recorded output.
- Add carrier, SIP, phone-number, AI-provider, storage, and optional-service costs where they apply.
- Check whether bandwidth, egress, or transcoding is included or billed separately.
Before committing engineering time
- Model your actual quality, concurrency, and monthly volume instead of comparing headline rates alone.
- Verify automatic volume thresholds and whether discounts require a sales contract.
- Price the operating surface too: ready-made apps, widgets, SDKs, APIs, and required engineering effort.
Clear answers
Pricing questions buyers ask before launch.
What counts as meeting minutes, and what counts as recording minutes?
Meeting minutes are media routed between people: one participant's stream delivered to another, billed at the live media rates. Recording minutes are any stream that ends at, or is processed on, MediaSFU servers: a recording, an AI agent listening to a caller, noise cancellation, a transcription feed. Those bill at the recording rates.
What does billed per actual media flow mean?
Routing is counted as each media producer × the people consuming it, per minute. Nobody consumes their own stream, so a 1:1 video call is 2 × 1, four people with one camera on is 1 × 3, and a five-person meeting with every camera on is 5 × 4. A camera that is off is not a producer, and an idle room moves no media, so neither is billed. Larger rooms also get automatic room-size tiers.
Is $0.0001 the complete cost of a phone call?
It is the MediaSFU audio transport rate. Carrier, SIP-provider, and phone-number charges are paid directly to the providers you choose. If an AI agent handles the call, the agent-infrastructure rate also applies.
Do you mark up AI or SIP providers?
No. Bring your own supported STT, LLM, TTS, SIP, and number providers and keep their billing relationship direct. This makes the MediaSFU portion of the bill easier to audit.
When are volume discounts applied?
Published automatic discounts begin at 2.5 million billable minutes per month for 20% off and 7.5 million minutes per month for 40% off. The discounted rate depends on the service unit being used.
Why are recording rates higher than live media rates?
Recording produces retained output and uses a separate processing pipeline. The recording matrix is priced per 1,000 output minutes by quality; storage and optional post-processing may be separate.
Replace assumptions with your workload
Model the minutes, quality, and output you expect.
The calculator shows the MediaSFU infrastructure portion so provider choices remain explicit.
Published rates are planning inputs, not a tax-inclusive quote. External provider, carrier, phone-number, storage, and optional-service charges vary by configuration.