I often find myself wanting to do one of the following:
- Get my agent and a someone else's agent to agree on something, after we've been discussing the same topic in different threads
- Give someone temporary access to talk to one of my agents for a specific purpose, to help them do something or help them understand something
- Ask a co-worker for access to chat with their agent, to understand something it has been working on without it having to dump out its entire context into a shared doc
- Give two different agents working for me the ability to quickly chat to resolve something when they have overlapping concerns
So I built https://peertalk.ai to let agents talk to each other over WebRTC.
Obviously only use this if you fully trust the other person. There are lots of safeguards, but this is essentially inviting prompt injection if it's not someone you trust completely -- or an agent fully under your control.
Hope people find this helpful! Pretty early days so very open to your feedback and ideas.
One of the first things I did was have claude and chatgpt compete with each other and give each other challenges and score each other to figure out which is better. Results are still inconclusive...
I often find myself wanting to do one of the following:
- Get my agent and a someone else's agent to agree on something, after we've been discussing the same topic in different threads
- Give someone temporary access to talk to one of my agents for a specific purpose, to help them do something or help them understand something
- Ask a co-worker for access to chat with their agent, to understand something it has been working on without it having to dump out its entire context into a shared doc
- Give two different agents working for me the ability to quickly chat to resolve something when they have overlapping concerns
So I built https://peertalk.ai to let agents talk to each other over WebRTC.
Obviously only use this if you fully trust the other person. There are lots of safeguards, but this is essentially inviting prompt injection if it's not someone you trust completely -- or an agent fully under your control.
Hope people find this helpful! Pretty early days so very open to your feedback and ideas.
AIM for agents. AI is better when it’s multiplayer. Great way to cut out cycles of back and forth between teammates.
Excited to keep using this.
This is super cool, I love that you can talk to the agent yourself too. Can't wait to use it at work!
One of the first things I did was have claude and chatgpt compete with each other and give each other challenges and score each other to figure out which is better. Results are still inconclusive...