Compare · Waterr vs Recall.ai
Waterr vs Recall.ai
Waterr is an AI meeting agent API: it hosts the call by join link, invite, or website embed and supplies the agent — a persona with a script and goals that talks, sees, and scores each participant, then returns transcript, recording, and scores. Recall.ai is a meeting-bot API that captures recordings and transcripts; a talking agent is a webpage and model pipeline you build yourself. To ship an AI-run meeting, pick Waterr.
vsAt a glance
Waterr and Recall.ai, side by side
| Waterr | Recall.ai | |
|---|---|---|
| What it is | Hosted AI agent API — scenario in, scored meeting out | Meeting-bot API for recordings, transcripts, and meeting data, plus Desktop and Mobile Recording SDKs |
| Talking agent | Built in — persona, voice, meeting script, avatar | You build it: a webpage the bot streams via Output Media, wired to your own speech model |
| Where participants join | Waterr-hosted video room by join link, scheduled invite, or website embed | Calls on Zoom, Google Meet, Teams, Webex, or GoTo Meeting (beta); the bot joins as a participant |
| What you write | API calls and webhooks — or your own OpenAI-compatible agent endpoint (Enterprise) | Bot orchestration, plus the agent webpage and model integration if the bot should talk |
| Goal scoring | Yes — each participant scored 0–5 per goal, with written feedback | Not listed in its docs — analysis is your build on top of the data |
| Vision | Persona reads camera and screen share in real time | Per-participant video streams delivered to you; interpreting them is your build |
| Mid-call control | Meeting Actions — make the agent say, know, or show something | Endpoints to output audio, video, screenshare, and chat from the bot |
| Recordings & transcripts | Transcript every meeting, recording when enabled, returned with the analysis | Mixed or per-participant audio and video, async or real time |
| Transcription | Built in — part of every call’s per-minute stack | Built-in, or bring Deepgram, AssemblyAI, ElevenLabs, AWS Transcribe, Rev, Speechmatics, or platform captions |
| Pricing shape | Per-minute credits for the whole agent stack, billed to the second; plans at $29, $99, and $299 a month | $0.50 per recording hour pay-as-you-go; built-in transcription $0.15/hr; 4-core bots +$0.10/hr, GPU bots +$1/hr; agent models billed separately |
| Best for | Calls an AI runs — interviews, sales calls, onboarding, support | Recordings and transcripts for notetaker and conversation-intelligence products |
Competitor claims verified against docs.recall.ai (bot overview, meeting platforms, Output Media, output audio, regions, Desktop Recording SDK) and recall.ai/pricing on 28 October 2026. "Not listed in its docs" means absent from docs.recall.ai as of that date.
The honest split
Which one should you pick?
Choose Waterr when…
- The AI is meant to run the conversation — an interview, a sales call, an onboarding walkthrough — not only record it.
- Persona, voice, script, and goals should be configuration, not a webpage and model pipeline you build and host.
- You need each participant scored against goals, with written feedback, delivered with the transcript and recording.
- Participants should join from a link, a scheduled invite, or a widget on your site, with no meeting platform to integrate.
Choose Recall.ai when…
- You only need recordings and transcripts of calls your team already runs.
- You plan to build and host the agent’s model pipeline yourself.
The details
An agent that runs the call vs a bot that records it
Both products answer "put something into a meeting with an API call," which is why they show up in the same searches. What arrives is different. A Waterr agent is a participant with a job: a scenario defines who it is, how it runs the conversation, and what the call should measure. A participant opens a join link, a time-bound invite, or a widget on your site, and the agent is there. It talks, sees the camera and shared screen, and when the call ends the API returns the transcript, the recording, and a score for each goal.
A Recall bot is a capture device. By default it joins as a participant named "Meeting Notetaker" with its mic and camera off, and its job is to hand your application the meeting — mixed or per-participant audio and video, async or in real time, transcripts from Recall’s own engine or a third-party provider, participant events, and metadata. Turning that bot into an agent is a separate build.
What it takes to make a Recall bot talk
Recall’s Output Media API has the bot load a webpage you host and stream that page’s audio and video into the call as its camera or a screenshare. The page receives meeting audio through getUserMedia and transcripts over a websocket, and Recall recommends pairing it with a speech-to-speech model rather than a speech-to-text, LLM, text-to-speech chain.
The documented constraints: video output is 1280×720 at 15 fps, the page is always the bot’s video (there is no audio-only mode), and the bot’s own video cannot be included in the recording. Heavier pages need 4-core bots at an extra $0.10 an hour or GPU bots at an extra $1 an hour, and the models behind the agent are billed by their own vendors. Everything that makes the bot an agent — the persona, the prompt, the model contract, tool calls, and any evaluation of the conversation — is yours to build and operate.
What Waterr gives you instead
On Waterr the agent is the product. Persona, meeting script, capabilities such as vision, knowledge base, and custom functions, and goals live on one scenario object. Goals score the participant 0–5 with written feedback after every call, and signed, retried webhooks tell your system when the transcript, recording, and analysis are ready.
Meeting Actions let your server reach into a live call — have the agent say a sentence verbatim, hand it a fact from your CRM, or put a screen app in front of the participant. Teams that already have an agent can point an OpenAI-compatible endpoint at Waterr on the Enterprise plan and let Waterr supply the room, speech, turn-taking, transcripts, and analysis. The work that sits on your side of a Recall integration — webpage, model wiring, scoring — is already done.
FAQ
Common questions
Waterr, if you want the agent finished. Waterr supplies the persona, voice, script, vision, and goal scoring, hosts the call, and returns the transcript, recording, and scores by API. With Recall you get a bot and a media path through Output Media; the speech model, persona, turn-taking, tools, and evaluation are yours to build.
Run your first AI meeting today.
One POST creates the scenario. One link runs the meeting. The API hands back the transcript, goal scores, and recording.
Keep reading
