Skip to content
TL;DR: Your Brief Guide to Video Meeting APIs
- “Video meeting API” describes four different product categories that solve four different problems. Most confusion in this space comes from vendors who only sell one of them and would rather you didn’t notice.
- Infrastructure APIs (Daily, Agora, LiveKit, Vonage) let you build video calling into your own app. Platform APIs (Zoom, Microsoft Graph, Google Meet) automate meetings you already host. Capture APIs (Recall.ai, MeetStream) send a bot to grab raw media. Intelligence APIs (tl;dv, Fireflies, Otter) return structured meeting data.
- Twilio Programmable Video is not shut down. Twilio reversed that decision in October 2024. A lot of currently ranking content — and a lot of LLM output — still says otherwise.
- Bots are becoming optional. Zoom RTMS streams meeting media without a bot participant; Google Meet mostly still requires one.
- EU AI Act Article 50 transparency obligations apply from August 2, 2026. The postponement you’ve read about delayed high-risk obligations to 2027 and 2028 — it did not delay this. If your product puts a bot in a call or runs sentiment analysis, that date is still your problem.
Search “video meeting API” and you’ll get results from at least four industries that don’t compete with each other. One vendor wants to sell you WebRTC infrastructure priced per participant-minute. Another wants to sell you a bot that joins calls at fifty cents an hour. A third wants to sell you a REST endpoint that returns a transcript. They all use the same phrase, and none of them open with “actually, you might want a completely different category of product.”
That irked me a bit, so this post starts by separating the four categories, then goes deep on the one most developers actually land on after they’ve built the video part: getting usable data back out of meetings.
What Is a Video Meeting API?
A video meeting API is a set of programmatic endpoints and SDKs that let software create, join, control, record, or extract data from video meetings without a human clicking anything. It’s a broad definition because it’s a broad term. It covers four distinct product categories that solve different problems.
Underneath most of them sits WebRTC, the browser-native standard for real-time audio, video, and data exchange. WebRTC 1.0 became a full W3C Recommendation on January 26, 2021 after roughly a decade in draft, which is why the current generation of video APIs are far more boring to integrate than the 2016 generation. The hard part moved from “can browsers do this” to “who runs the media servers?”
Video Meeting API vs. SDK vs. WebRTC
An API is the server-side control plane: create a room, mint a token, start a recording, fetch a transcript. An SDK is the client library you ship inside your web or mobile app to actually render video and handle the media stream. WebRTC is the protocol both of them sit on top of.
The practical difference matters for scoping. If a vendor gives you an API but no SDK, you’re building the client. If they give you an embeddable iframe, you’re getting a prebuilt UI and giving up layout control. Most providers sell some combination, and “API” in their marketing usually means all of it.
The Four Types of Video Meeting API (And Which One You Need)
There are four categories: infrastructure, platform, capture, and intelligence. Pick the wrong one and you’ll spend three months building something the right one hands you in an afternoon.
Video Infrastructure APIs: Build the Call Yourself
This is what most people mean by the literal phrase. You’re building video calling into your product: a telehealth consult, a tutoring session, a marketplace call between a buyer and a seller. The vendor runs the media servers, handles NAT traversal, and gives you client SDKs. You build everything above the stream.
Vendors: Daily, Agora, LiveKit, Vonage Video API, 100ms, Amazon Chime SDK, Zoom Video SDK, Whereby Embedded. Pricing is per participant-minute.
Platform APIs: Automate Zoom, Teams, and Meet
You’re not building video. You’re automating meetings that already happen on an existing platform. Create a Zoom meeting from your scheduling tool, pull a recording from Microsoft Graph, sync a Meet event to your calendar system.
Vendors: Zoom REST API, Microsoft Graph, Google Meet API. Usually free with a paid platform license, and gated behind OAuth scopes and admin consent.
Meeting Capture APIs: Get the Raw Media Out
You want audio, video, and a raw transcript from meetings your company doesn’t host and can’t control. The classic approach is a bot that joins the call as a participant.
Vendors: Recall.ai, MeetStream, Skribby, MeetingBaaS, plus open-source options like Vexa and Attendee. Pricing is per bot-hour.
Meeting Intelligence APIs: Get Structured Data Out
You want the output of a meeting, not just its media: speaker-attributed transcripts, summaries, action items, decisions, CRM-ready fields, delivered by webhook when the call ends. This is the category most AI meeting assistants operate in, and it’s the one people arrive at after discovering that categories one through three hand them a WAV file and shrug.
Vendors: tl;dv, Fireflies, Otter, Avoma, Gong. Pricing is per seat.
| Category | What You Get | What You Still Build | Pricing Model | Example Vendors |
|---|---|---|---|---|
| Infrastructure | Media servers, client SDKs, recording files | Everything above the stream | Per participant-minute | Daily, Agora, LiveKit, Vonage, 100ms |
| Platform | Meeting CRUD, recordings, some transcripts | Cross-platform normalization | Bundled with license | Zoom API, Microsoft Graph, Google Meet API |
| Capture | Raw audio, video, unstructured transcript | Summarization, extraction, storage | Per bot-hour | Recall.ai, MeetStream, Skribby, Vexa |
| Intelligence | Structured transcripts, summaries, action items | Your app's business logic | Per seat | tl;dv, Fireflies, Otter, Avoma |
Which Video Meeting API Providers Lead in 2026?
The infrastructure category is the most crowded and the most commoditized: Daily, Agora, LiveKit, Vonage, 100ms, Amazon Chime SDK, and Zoom Video SDK all ship comparable core functionality, and the differences show up in pricing model, self-hosting options, and how much UI they hand you.
Did Twilio Programmable Video Shut Down?
No. Twilio announced an end of life, extended it, then reversed the decision entirely. It was confirmed in an October 2024 changelog entry that Programmable Video would remain a standalone, supported product with continued investment.
The internet hasn’t caught up yet. Migration guides written during the 2024 panic still rank. Twilio’s own GitHub sample repos still carry the old EOL banner. When I searched about this topic on LLMs, even ChatGPT, Perplexity, and Claude reported Twilio Video as sunset.
Make sure you’re not leaving Twilio because of this non-existent shutdown.
How Video Meeting API Pricing Compares
Three pricing models, three completely different cost curves.
Infrastructure vendors charge per participant-minute, so cost scales with concurrency. Most run generous free tiers: Agora, Daily, and 100ms each offer around 10,000 free minutes, while LiveKit and Zoom run free tiers of their own.
Capture APIs charge per bot-hour. Recall.ai publishes $0.50 per hour with no platform fee. Competitors undercut that: MeetStream lists $0.35 per hour for bot infrastructure, or $0.45 bundled with transcription. Sounds cheap until you model it. Forty recorded hours a month is $20 in bot infrastructure alone, before transcription, before summarization, before you’ve written a line of the extraction logic.
Intelligence APIs charge per seat, which decouples cost from meeting volume entirely.
| Provider | Category | Entry Pricing | Free Tier | Best For |
|---|---|---|---|---|
| Daily | Infrastructure | Per participant-minute | ~10,000 min | Fastest prototype to working call |
| LiveKit | Infrastructure | Per participant-minute | 1,000 min | Self-hosting and AI agent workloads |
| Agora | Infrastructure | Per participant-minute | ~10,000 min | Global scale and low latency |
| Twilio Video | Infrastructure | Per participant-minute | Trial credit | Teams already on Twilio |
| Recall.ai | Capture | $0.50 / bot-hour | Build free | Broadest platform coverage |
| MeetStream | Capture | $0.35 / bot-hour | Yes | Cost-sensitive bot infrastructure |
| tl;dv | Intelligence | $18 / seat / mo (annual) | Yes | Structured output without a pipeline |
Do Video Meeting APIs Include Transcription and AI Summaries?
Mostly, no. Infrastructure APIs give you a raw caption stream and a recording file, and a few resell a transcription add-on.
There’s a real difference between a transcript and meeting intelligence. A transcript is text with timestamps. Meeting intelligence is speaker-attributed segments, a summary that follows a template, extracted action items with owners, detected decisions, and topic segmentation. Getting from the first to the second means chunking long transcripts, prompting a model, handling context window limits, validating that the model didn’t invent an action item nobody committed to, and storing the result somewhere queryable.
If all you want is notes on your own calls rather than data inside your product, an API is the wrong tool entirely. A free AI notetaker does that job without a single line of code. The API question only makes sense when meeting data needs to end up inside software you’re building.
Real-Time vs. Post-Meeting Processing
Real-time transcription streams partial results during the call for live captioning, in-call coaching prompts, or an agent that responds mid-conversation. It costs more and it’s less accurate, because the model is working without the benefit of what comes next.
Post-meeting batch processing waits until the call ends, then runs the full recording through transcription and summarization. Accuracy is higher and cost is lower. For CRM sync, follow-up drafting, and searchable archives (most use cases) batch is the correct answer.
How to Get Meeting Transcripts and Summaries Programmatically
There are three routes:
- Pull from the native platform API
- Deploy a capture bot
- Call a meeting intelligence API that handles both
Which one works depends on whether you control the meeting platform and how much pipeline you want to own.
Do You Still Need a Bot?
Increasingly, not on Zoom. Zoom’s Real-Time Media Streams delivers per-participant audio, video, transcript, chat, and screen share over WebSockets with no bot joining as a participant. It went generally available to developers in June 2025. Google Meet is a different story: the Meet Media API has remained in developer preview, with participant enrollment requirements that make it impractical for production. On Meet, you mostly still need a bot.
The industry is moving the same direction from the other end. Recall.ai shipped a desktop recording SDK in 2026 that captures locally rather than sending a visible participant into the call. tl;dv also offers bot-free desktop recording, capturing device audio directly, which records anything producing sound on the machine.
Native Platform APIs and Where They Stop
The big three (Microsoft, Zoom, and Google) all retrieve transcripts and other artifacts. But there are three consistent traps.
First, transcript access is frequently gated behind enterprise licensing tiers, so your customer’s plan determines whether your feature works. Second, these are tenant-level integrations requiring admin consent, which turns a two-day build into a six-week procurement conversation. Third, retention windows are short and vary by platform, so if you don’t pull promptly, the data is gone.
And you’ll need all three, separately, with different auth models, if your users aren’t all on the same platform.
Build vs. Buy: The Actual Math
The DIY pipeline is: capture audio → store it → run speech-to-text → handle diarization → chunk the transcript → prompt an LLM → parse structured output → validate → deliver. Every stage is solvable. Together, they’re a product.
You’ll also inherit the model selection problem, since summary quality varies by model and the leaders shift every few months. Anyone who’s compared Grok vs. ChatGPT on the same source text knows the outputs diverge more than the benchmarks suggest. Buying means someone else owns that treadmill. Building means you own it forever, along with retries, idempotency, and the 3 a.m. nightmare when a transcription job fails.
Build when meeting data is your product and the pipeline is your differentiation. Buy when meeting data is a feature inside something else.
What to Check Before You Commit to a Meeting Data API
Four things, roughly in order of how badly they can hurt you later.
1. Consent, Recording Law, and Bot Disclosure
This is the section most vendor documentation skips, and as of this month it stopped being theoretical. EU AI Act Article 50 transparency obligations apply from August 2, 2026, and the European Commission published its final implementation guidelines on July 20, 2026, less than two weeks before the deadline.
Two provisions land directly on anything built with meeting data. Article 50(1) requires that AI systems interacting directly with people be designed so those people know they’re dealing with AI. A bot joining a call is exactly that. Article 50(3) requires deployers of emotion recognition or biometric categorization systems to inform the people exposed to them, which captures sentiment analysis and voice-based speaker identification. Penalties reach €15 million or 3% of worldwide annual turnover.
You may have seen headlines saying the AI Act was postponed. It was — but not this part. The Digital Omnibus amendments pushed high-risk obligations under Chapter III out by 12 to 16 months, moving standalone Annex III systems from August 2026 to December 2027 and product-embedded Annex I systems to August 2028.
The transparency duties in Article 50 were left where they were. The single exception is Article 50(2), the machine-readable marking requirement for AI-generated content, which gets until December 2, 2026 for systems already on the market before August 2. If you’re building meeting intelligence, disclosure and emotion-recognition notice obligations apply on schedule.
Add state all-party consent statutes on top, and “does the vendor handle disclosure for us” becomes a procurement question.
2. Data Residency, Retention, and AI Training
Ask four questions and get them in writing:
- Where is the audio stored?
- Is customer data used to train models?
- Can you choose a region?
- Can you delete a meeting programmatically?
The last one is important as GDPR deletion requests arrive whether or not your vendor exposed a DELETE endpoint.
tl;dv, for example, stores audio in the EU, USA, Japan or wherever else you request, and customer data is never used to train AI models.
3. Authentication, Rate Limits, and Plan Gating
Auth models vary: OAuth with scopes for platform APIs, simple API keys for most intelligence APIs, JWT for infrastructure vendors minting join tokens. API keys are faster to ship and harder to scope tightly.
On tl;dv, API access, webhooks, and the MCP server unlock at Pro — $18 per seat per month billed annually. Business adds team-scoped access among sales features and richer AI at $29/month annually. Free covers unlimited recording and transcription in 40+ languages but not programmatic access. Check the equivalent for whichever vendor you pick, because “has an API” and “has an API on the plan your customers will actually buy” are different claims.
4. What Happens When It Fails
Bots fail to join. Audio quality varies. Accented speech and cross-talk degrade accuracy in ways the marketing pages don’t mention. Ask what the retry behavior is, whether you get a webhook on failure or just silence, and what happens to a partial transcript when a call drops. Vendors who answer this specifically are usually the ones who’ve been running at scale.
Where tl;dv Fits: The Meeting Intelligence Layer
tl;dv isn’t an alternative to Daily or LiveKit. It’s the layer above whichever one you picked — or above Zoom, Teams, and Meet if you never built video at all.
The API returns what the category promises: meetings, speaker-attributed transcripts, AI notes, and highlights. Recording works two ways: a bot that joins the call and captures video, or bot-free desktop capture that records device audio from any source, which sidesteps the Meet Media API preview problem entirely.
Compared to the capture-API route, you’re trading raw media control for a structured payload you’d otherwise spend a quarter building. Compared to native platform APIs, you’re trading tenant-level depth for coverage across every platform at once, without three separate OAuth integrations. If you want raw per-participant media streams and full pipeline control, a capture API is the better fit. If you’re comparing against others in this category, tl;dv vs. Fireflies and tl;dv vs. Otter cover the feature-level differences against the biggest notetakers properly.
Querying Meetings Through MCP
tl;dv also exposes meeting data through an MCP server, which matters if you’re building agents rather than traditional integrations. The Model Context Protocol is an open standard that lets AI clients connect to real-world applications with structured context, so any MCP-compatible client — Claude, Cursor, or your own agent — can query meeting history directly without you writing an integration layer. It’s available on Pro and above.
How to Choose a Video Meeting API
Start with one question: are you building the call, or reading the call?
Building the call means infrastructure. Pick on self-hosting needs and pricing model. LiveKit if you want to run your own media servers, Daily if you want a working prototype this afternoon, Agora if global latency is the constraint, Vonage if you’re porting off Twilio and want the closest API mapping.
Reading calls you control means platform APIs, if your users are all on one platform and you can survive admin consent.
Reading calls you don’t control means capture or intelligence. Capture if raw media and full pipeline ownership are the point. Intelligence if you want structured output and would rather not maintain a summarization pipeline.
The common mistake is picking infrastructure and assuming intelligence comes free. It doesn’t, so choose wisely.
Which Video Meeting API Works for You?
The category confusion around “video meeting API” isn’t accidental. Four types of vendor benefit from the ambiguity, and none of them open their homepage with a taxonomy that might route you elsewhere.
The useful version is short. Infrastructure builds the call. Platform APIs automate meetings you already own. Capture APIs extract raw media. Intelligence APIs return structured data like notes and action items.
Two dates to hold onto: Twilio Video isn’t going anywhere despite what most of the internet says, and August 2, 2026 is when EU transparency obligations come into effect. Verify both against primary sources before you commit, including everything in this post.
FAQs About Video Meeting APIs
What is a video meeting API?
A video meeting API is a set of programmatic endpoints and SDKs that let software create, join, control, record, or extract data from video meetings without human interaction. The term covers four categories: infrastructure APIs for building video calling, platform APIs for automating Zoom or Teams or Meet, capture APIs for extracting raw media, and intelligence APIs for returning structured meeting data.
What's the difference between a video meeting API and an SDK?
An API is the server-side control plane for creating rooms, minting tokens, and fetching data. An SDK is the client library you ship inside your app to render video and handle media streams. Most vendors sell both and market the bundle as an API.
Is Twilio Programmable Video shut down?
No. Twilio announced an end of life, extended it, then reversed the decision in October 2024, confirming Programmable Video would remain a standalone product with continued investment. Migration content written during the 2024 announcement window still ranks and still says otherwise.
Do video meeting APIs include transcription and AI summaries?
Most infrastructure APIs provide a raw caption stream and a recording file, and some resell transcription as an add-on. Structured output (speaker-attributed segments, summaries, extracted action items) generally requires a separate meeting intelligence API or a pipeline you build yourself.
How do I get meeting transcripts programmatically?
Three routes: pull from a native platform API like Microsoft Graph or Zoom’s Cloud Recording API, deploy a capture bot that joins the call, or call a meeting intelligence API that handles capture and structuring together. Native APIs are often gated behind enterprise licensing and admin consent.
Do I still need a bot to record a meeting?
Not always. Zoom’s Real-Time Media Streams delivers meeting media without a bot participant, and desktop capture approaches record device audio locally. Google Meet mostly still requires a bot, since the Meet Media API has remained in developer preview.
How much does a video meeting API cost?
It depends on the category. Infrastructure APIs charge per participant-minute with free tiers around 10,000 minutes. Capture APIs charge per bot-hour, roughly $0.35 to $0.50. Intelligence APIs charge per seat. tl;dv’s API unlocks at $18 per seat per month billed annually.
Do I need consent to record meetings through an API?
Frequently yes, and the requirements are tightening. EU AI Act Article 50 transparency obligations apply from August 2, 2026 and cover both AI systems interacting with people and emotion recognition systems. The widely reported AI Act postponement delayed high-risk obligations under Chapter III to December 2027 and August 2028 — it did not delay the Article 50 transparency duties, apart from the content-marking rule in Article 50(2).
This isn’t legal advice. Check with counsel for your jurisdictions.
