For hosted meeting bots, start with Recall.ai, Meeting BaaS, Skribby or MeetStream. Attendee and Vexa are open source, so you can self-host them. Nylas Notetaker API bundles a bot with email and calendar APIs. Zoom Realtime Media Streams skips the bot on Zoom calls, and the Fireflies.ai API can send its bot to a live meeting.
Six of the nine options below publish an hourly rate for a bot in a meeting, and every one of those rates is under a dollar an hour. So the bot itself is cheap. What costs you is everything around it: the hours lost when a bot sits in a waiting room nobody admits, the transcription you might pay for on top, and the rewrite if you pick the wrong way of capturing the call in the first place. That's why I'd test before I compared.
An afternoon's test that saves a rewrite
Most of these give you free hours or starter credit, so this costs you a few hours and nothing else. Pick two hosted APIs and one open-source project, then send each a bot into three real meetings: one on Zoom, one on Google Meet and one on Teams, ideally from accounts outside your own company.
Watch for five things. How long it takes from your API call to the bot appearing in the room. What happens in a waiting room or lobby, and whether your code hears about it. Whether the transcript streams during the call or only arrives afterwards. Whether you get speaker names, or just Speaker 1 and Speaker 2. And whether the bot leaves on its own when the humans do, because a bot that hangs on for an hour after everyone's gone is billing you for silence.
Write the results in a shared doc. It's the cheapest bit of due diligence in the whole build, and it'll tell you more than any feature table.
Hosted bot APIs you don't have to run
Recall.ai
A hosted API that sends a bot into video calls for recordings, transcripts and metadata, with a separate desktop SDK for recording without a bot.
Meeting BaaS
Meeting bots as a service for Zoom, Google Meet and Teams, sold as a developer API with subscription tiers and an option to self-host.
Skribby
Small, usage-billed bot API for Zoom, Teams and Meet, aimed at developers who want a bot deployed in minutes rather than a platform.
MeetStream
Bot infrastructure for AI apps and agents that need to act during the call, with bot APIs for the three big meeting platforms.
Nylas Notetaker API
If your app already touches users' inboxes and calendars, this managed meeting bot sits in the same API as email, calendar and scheduling.
Open-source, bot-free and add-on routes
Attendee
Open-source meeting bot API you can run on your own servers or use hosted, covering Zoom, Google Meet and Teams.
Vexa
Apache-licensed and self-hostable, it streams live transcripts over WebSocket and exposes meetings through MCP, which suits agent builders.
Zoom Realtime Media Streams (RTMS)
Zoom's first-party route to live audio, video and transcript data with no participant bot at all, for apps built on Zoom alone.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Fireflies.ai API
Teams already on Fireflies can send its note-taking bot to a meeting link through a GraphQL mutation, without building bot infrastructure.
The four choices that shape your build
| Choice | One way | The other way | Ask yourself |
|---|---|---|---|
| Who runs the bots | Hosted API, billed by usage | Open source on your own servers | Who fixes it when a meeting platform changes? |
| How you capture | A bot joins as a participant | No bot: first-party streams or on-device recording | Will your users accept a visible bot? |
| When you get data | Recording and transcript after the call | Live stream during the call | Does anything need to happen mid-meeting? |
| What you buy | A bot API on its own | Bots bundled with email and calendar APIs | Do you need the other APIs anyway? |
Hosted or self-run. Meeting platforms update their clients, and a bot that joined fine last month can need fixing this month. With a hosted API that's the vendor's on-call problem. Self-hosted, it's yours, along with scaling the bots when usage spikes. What you get back is control over where recordings live, which matters if your customers are in regulated industries. Two of the nine are open source, and a third offers self-hosting next to its hosted service.
Bot or no bot. A bot in the participant list is honest and easy to explain, but some users hate it and some hosts remove it. The alternatives are a first-party stream from the meeting platform itself or recording on the user's own computer. The catch with the first-party stream on this list is that it only covers Zoom, so a product that also needs Meet and Teams still needs bots for those.
After the call or during it. A notetaker can wait for the recording. A voice agent that answers questions in the meeting can't. If you're building the second kind, test streaming latency in that afternoon test, and pair the bot with a speech-to-text API that's fast on live audio, not just accurate on files.
Stand-alone or bundled. If you're already integrating calendars to know which meetings to join, a provider that sells both saves one vendor and one set of OAuth scopes. If you aren't, the bundle is just more surface area.
What changed in the last year
Two things stand out. Real-time access is now the headline, not a footnote: two of the nine advertise MCP access so an AI agent can read or act on a live meeting, and one markets itself directly at real-time agents. And bot-free capture went from workaround to product, with a first-party streaming API on one side and a desktop recording SDK on the other. Both are worth testing if a visible bot is a problem for your users.
Do you make meeting bot APIs or meeting data tools? Tool makers can ask to be considered for a featured place on this list.
How the bill adds up
The headline rate is per recording hour on six of the nine, and all six put it under a dollar. Five of the nine give you free hours or starter credit to begin with. One bundles bot hours into monthly plans with a free tier. One sells monthly tiers that cap bots per day, next to prepaid pay-as-you-go. The first-party Zoom stream runs on developer credits, and the Fireflies docs read for this guide don't show a separate hourly bot rate.
The things that move your real cost are what the headline hides. Is transcription included, or billed as its own hourly line? At least one hosted plan prices it separately. Are you charged while a bot waits in a lobby? Is there a daily or concurrent bot cap you'll hit on a Monday morning when everyone's calls start at once? And if you self-host, the bill becomes servers, storage and an engineer's time, and whether that beats an hourly rate depends on your volume and on who maintains it.
Which route I'd take
- You're prototyping an AI notetaker this month: a hosted per-hour API. Get the product right first; you can move later if volume justifies it.
- Recordings must stay on your own infrastructure: an open-source bot API you self-host, and budget for the maintenance.
- You're building an agent that speaks or acts in the meeting: a provider with live streaming over WebSocket or MCP, tested for latency, not only for accuracy.
- Your product only lives on Zoom and users dislike bots: the platform's own media stream.
- You also need email and calendar data: a bundled provider, so meeting joins and calendar sync share one integration.
- You just want notes for your own team: you don't need an API at all. Look at the AI note takers built for that and skip the engineering.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.


