What the Agora Startup Program Gives You
Agora offers up to $5,000 in credits toward its real-time engagement platform, which covers the meter that actually runs the bill: minutes of voice, video and low-latency messaging carried across Agora's network.
AI Perks tracks it alongside $7.7M in credits across 194 companies.
The boundary matters more here than in most categories. Agora sells transport - moving audio and video between people fast enough that they can interrupt each other. The credit covers that invoice and nothing else: not the model tokens behind a voice agent, not the speech-to-text and text-to-speech wrapped around it, not the storage and CDN egress your recordings generate. Eligibility depends on your stage and funding, and the current terms are listed on getaiperks.com.
What Agora Is Actually For
Agora is a real-time communications network with SDKs for voice, video, interactive live streaming and low-latency messaging, sold so that you never have to operate WebRTC infrastructure yourself.
The decision that matters is not Agora versus another RTC vendor. It is hosted real-time versus two cheaper things founders skip past.
The first is whether you need real-time at all. One-to-many delivery where nobody talks back - a webinar replay, a recorded lesson - is a job for HLS and a CDN, and the cost gap is not close. Real-time pricing buys sub-second latency. If nobody is interrupting anyone, you are paying for a property your product never uses.
The second is plain WebRTC. A two-person browser call is free and built in, and for a long stretch of a product's life that is genuinely enough. The cliff arrives on four fronts at once:
- Three or more participants. Peer-to-peer means every device uploads to every other device. At four people that is three uplinks from a phone on hotel wifi, and it falls apart.
- Restrictive networks. A meaningful share of real sessions cannot connect directly and must be relayed. Running that relay yourself is a bandwidth bill plus an on-call rotation.
- Distance. Cross-border calls are where a managed global network stops being a convenience and becomes the product.
- Mobile. Surviving a 4G-to-wifi handoff, backgrounding and heavy packet loss is years of SDK work unrelated to your product.
The honest heuristic: if real-time is a feature, a two-party peer connection or a self-hosted media server carries you further than you expect. If real-time is the product - telehealth, live tutoring, social audio, voice agents - the reliability difference shows up in retention. Spend the credit there.
How Agora Pricing Behaves at Scale
Agora meters minutes, but the minutes it counts are per stream per user, not per session. That distinction is the largest single source of surprise on a real-time invoice, because group sessions scale quadratically while the wall clock does not move.
| Meter | What drives it | What makes it spike |
|---|---|---|
| Audio minutes | Each user's audio streams, counted individually | Always-on channels nobody ever leaves |
| Video minutes | Aggregate resolution across streams | Publishing at capture resolution, not display resolution |
| Resolution tier | Total pixel area being carried | 1080p streams rendered into 180px thumbnails |
| Cloud recording | A separate recording stream per session | Recording every session by default |
| Storage and egress | Your own bucket and CDN | Indefinite retention, no lifecycle policy |
| Real-time messaging | Signaling, presence, chat | Presence heartbeats on very large channels |
| Speech and model calls | Billed by other vendors entirely | Assuming RTC credits cover the agent |
Rates, free allowances and tier boundaries change. Verify current figures against Agora's own pricing page before modelling anything.
The arithmetic that decides your bill is participant fan-out. Every user in a channel receives a stream from every other user, so billable minutes rise with the square of headcount:
| Participants | Streams each user receives | Billable minutes, one 30-minute call |
|---|---|---|
| 2 | 1 | 60 |
| 4 | 3 | 360 |
| 6 | 5 | 900 |
| 8 | 7 | 1,680 |
| 12 | 11 | 3,960 |
Every row is the same thirty minutes on a clock. The twelve-person call bills 66x the two-person call. The exact formula varies by product and plan, so check Agora's usage documentation, but the shape holds across every vendor in this category.
Put a free monthly allowance against that table and it stops being abstract. An allowance around 10,000 minutes - the figure commonly cited for Agora's free tier, worth verifying - is roughly 166 two-person calls or about 2.5 twelve-person calls. Same allowance, two wildly different products.
The second lever is resolution. Video is priced in tiers set by the pixel area you carry, so 1080p costs meaningfully more per minute than 720p, and 720p costs multiples of audio only. The tier follows what you publish, not what your UI renders. A grid showing eight people at 180px wide while every client publishes 1080p is paying the top tier to throw pixels away. AI Perks lists the credit amount, the multipliers are yours to control.
What Agora Credits Stack With
Real-time credits stack unusually cleanly right now, because the highest-value thing being built on RTC - the live voice agent - is a four-vendor pipeline, and every vendor in that pipeline runs a startup program.
One turn of a voice agent is: microphone, transport, speech-to-text, model, text-to-speech, transport, speaker. Each hop is a separate invoice from a separate company:
- Real-time transport credits cover the Agora leg, the $5,000 on this page
- Model credits cover the LLM, usually the largest line item in the pipeline
- Speech credits cover transcription and synthesis, both metered per second of audio
- Cloud credits cover orchestration, recording storage and egress
- Observability credits cover the session analytics you need to debug latency
A team holding three of those five has most of that pipeline funded by grants. Which programs combine cleanly and which quietly exclude each other is why AI Perks is maintained as a live list, not a folder of bookmarks.
What Founders Get Wrong About Real-Time Credits
The most expensive mistake is budgeting wall-clock minutes instead of stream minutes, because the number in your model and the number on the invoice can differ by an order of magnitude for identical product usage.
Five failure patterns, in rough order of what they cost:
Budgeting session time. "500 users, one 30-minute call a week" sounds like 65,000 minutes a month. If those calls average six participants, it is closer to 325,000. The product did not change, only the unit you counted in.
Publishing at capture resolution. Cameras default high, layouts render small. Matching publish resolution to the largest tile your UI displays is a config change that can move you down a pricing tier permanently.
Never closing the channel. People close tabs, phones sleep, reconnect loops leave ghost participants. Without leave handling and idle timeouts, a channel bills at full rate with nobody in it. One stuck session can outspend a month of real usage.
Recording by default. Cloud recording is a second meter, and the file it produces then costs storage and egress on your own cloud bill for as long as you keep it. Record on an explicit trigger, and set a lifecycle policy the same day.
Assuming the credit covers the agent. In a voice agent, transport is usually the smallest of four costs. Funding transport and not inference solves the cheap third of the problem.
On the way out: the SDK is proprietary, but the concepts under it - channel, publish, subscribe, track - map onto WebRTC and every competitor. Keep the integration behind one module of your own, and decide at 70% of credit consumed what your unsubsidised bill looks like, not at 100%.
Where Agora Credit Terms Are Tracked
Agora sits in the communications category at getaiperks.com, beside the messaging, telephony and voice infrastructure programs. Amounts and conditions are set by the vendor and revised without notice, so the listing is the thing to read rather than a figure quoted here.
Two questions decide whether a real-time grant is worth much. Is the whole pipeline covered, or only the transport leg? And what does the same usage cost at list price, once your own participant count and publish resolution go through the tables above?
Frequently Asked Questions
How much is the Agora startup program worth?
Up to $5,000 in credits toward Agora real-time voice, video and messaging. How long that lasts depends almost entirely on session shape: audio-only two-party sessions stretch it for a very long time, while multi-party 1080p video burns it quickly. Current amounts and eligibility are tracked at getaiperks.com.
Why is my Agora minute count so much higher than my call duration?
Because minutes are counted per stream per user, not per session. In a six-person call each participant receives five streams, so thirty minutes on the clock bills as roughly 900 minutes. The multiplier grows with the square of participant count, which is why group products need a different model than one-to-one products.
Do I need Agora, or is plain WebRTC enough?
For two-party browser calls in good network conditions, WebRTC alone is genuinely enough and costs nothing. You hit the cliff at three or more participants, on restrictive networks needing relay, on cross-border calls, and on mobile. Those four are what a managed real-time network is actually selling you.
Do Agora credits cover my AI voice agent costs?
No. Agora bills for transport only. The transcription, the model tokens and the voice synthesis are three separate vendors, and together they usually cost more than the transport does. Plan to hold model and speech grants alongside the real-time grant. Compatible programs are listed at getaiperks.com.
What is the fastest way to cut an Agora bill?
Three changes, in order of impact: publish video at the resolution your UI actually renders rather than at capture resolution, close channels on disconnect with an idle timeout, and drop to audio-only in any view where video is not visible on screen. Each is configuration, not architecture.
Can I combine Agora credits with other startup credits?
Yes, and communications credits stack well because the invoices do not overlap. Cloud credits cover orchestration and recording storage, model credits cover the LLM, speech credits cover transcription and synthesis. AI Perks tracks $7.7M in credits across 194 companies, including which of them conflict.
Ship the calls. Let someone else pay for the first million minutes.