CoAgentor
AI Agents that participate live in meetings
141 followers
AI Agents that participate live in meetings
141 followers
Most AI Meeting agents just take notes. CoAgentor brings AI meeting agents into your live calls, to answer live. They listen, 'raise their hand', and speak up - answering questions, surfacing data, and keeping every meeting on point, informed, and productive.
Free Options
Launch Team / Built With
Render: Get up to $100K in startup creditsCloud infrastructure for founders who want to move fast and NOT break things
Promoted
The "meeting around the information instead of with it" framing is sharp. We're running something parallel in a different setting, live AI to AI conversations in an arena, and the presence problem is similar but inverted. Humans interrupt by stopping, AIs interrupt by layering. Curious if your trigger rules had to handle that differently when agents are talking to agents vs agents listening to humans. Have you had to tune turn-taking differently for those two cases?
@ichiba That's interesting, and not sure I've run into it much yet!
When you have multiple agents or multiple human participants in a meeting, CoAgentor does diarized transcription, meaning it can realize where the words are coming from, individual person or agent.
A lot of my meetings, I try to really only have one agent who is a fairly active participant and the rest are back channel insight givers - or you only call on one a time:
"Stats Agent, what were the trends in performance over the last few days and weeks"
"Github architect, what kind of unseen complexities do you see with feature X, in the Y repo"
I'd love to hear more about your project, and if you get the chance to test CoAgentor any feedback you have about this kind of scenario!!
@coagentor The "one active, rest as back channel" philosophy is smart. We tried the opposite (multiple actives with overlap resolution) and it falls apart fast. The diarization angle is underrated too, most multi-agent tools treat the conversation as an undifferentiated stream. I'll take a look at CoAgentor with a meeting-heavy use case in mind and report back.
@ichiba Thanks! Looking forward to it!
Wow so its like a partner you have while presenting or listening. One question if we setup multiple agents will there be conflict like both trying to speak at the same time or is there some priority system?
@prateek_kumar28
Yea, that was one of the trickier things to get right at first, figuring out when to stop the agents from talking. It's actually what led to the Hand Raise and silent back channel modes, where overlap matters a lot less.
On the top tier plan, you can create as many or as few agents as you want, each scoped with its own data source contexts, and within each agent you can configure different trigger settings. So there are a few levels of control:
Live (speaks when triggered)
Hand Raise (waits to be called on)
Silent back channel (writes to Slack / Teams / dashboard)
Different agents with different purposes, governed by their data source scoping and their trigger settings.
Obviously you can create a real audio mess with ten agents all speaking at once 😅 so there are some overlap safeguards, but back-to-back speech and cutting real participants off is still a bit of a work in progress.
Meetings are usually messy anyway, and humans typically stop speaking when they get interrupted.
If you get to test it out Prateek, would love your feedback!!
Very cool idea, what made you want to build it?
And where is data held, i saw you are US based, but do you have gdpr / SOC2 certification?
@michal_chmielowski1 Thanks! It was a lot of fun to build!!
No SOC 2 atm, that sort of cert / audit is on the roadmap once we start moving upmarket. Right now the product is pitched at individuals, small teams and SMBs rather than regulated enterprises, so the compliance lift follows the customer.
The technical security basics are in place, RLS on tables, encryption at rest, server side auth checks on all routes, ownership checks on resource IDs to prevent IDOR, and every OAuth integration is platform-approved by the vendor except for slack, which is in progress, but takes a while to get marketplace approved (Google, Microsoft, HubSpot, Github, JIRA, Confluence all passed, Notion in motion.)
If you have a real use case, would be happy to explore options that would make it more suitable for you!
"Meeting around the information instead of with it" is the exact pain. The hand-raise + silent back-channel split is clever — interrupts break flow, silent breaks trust. Curious how CoAgentor handles hallucination risk when the agent is about to speak and the source data is ambiguous; does it defer to silence by default?
@mcarmonas There are levels of confidence depending on the agent config you are using, or the call you're on, so you can change between talk / hand raise / send silent back-channel, and across each of the three, if you're even a bit confident trigger, or make it so that the agent would "bet it's life on it" before it would trigger.
If you get the chance to check it out, would love to hear any feedback or suggestions you have!!