Yesterday, OpenBot launched on Product Hunt and finished as the #2 Product of the Day with 350 upvotes.
Honestly, we didn t know what to expect. Seeing so many people support OpenBot, test the product, ask questions, and share honest feedback meant a lot to our entire team.
Thank you to everyone who:
- upvoted and commented,
- shared our launch,
- downloaded and tested OpenBot,
- reported bugs,
- or told us what we should build next.
The ranking is exciting, but the most valuable part was seeing that other people share our vision: AI should move beyond isolated chats toward teams of specialized agents that can work together.
We re now going through every comment and piece of feedback from the launch. Some of it has already influenced what we want to improve next.
This launch was not the finish line. It was the beginning of the next stage for OpenBot.
Thank you again for making this launch special. We re just getting started.
- Kamil and the OpenBot team
@norbertbodziony Congratulations. And happy product launch.
the multiplayer angle is the part that'd make me cautious before rolling this out to a team - if Claude, ChatGPT and Grok are all sitting in the same live workspace reading the same shared context, has anyone tested what happens when one agent's output contains something the next agent reading it would treat as an instruction? that's less of an edge case once you have multiple vendors with different safety tuning looking at the same shared board instead of isolated sessions
@galdayan Agents keep separate conversations, and shared channels let you choose which agents collaborate. This helps limit shared context, but it isn’t a hard security boundary or a complete safeguard against this kind of risk.
@kamil_cyrek makes sense, "soft boundary" is probably the honest way to describe it then. we ended up putting the actual guard at the output layer instead of trusting isolation - anything coming back from one agent gets treated as untrusted data before another agent sees it, same as we'd treat text pulled off a random webpage. slower to build than just letting them talk but it means the multiplayer feature doesn't have to also be a security feature
An open source workspace where you can mix Claude, ChatGPT, Grok and custom agents, give them roles and let them delegate, covers a gap that single model tools leave open. Letting your team collaborate live with the same agents is a neat idea. Is the delegation between agents configurable or does it happen automatically based on the roles?
@karimbenkeroum You can assign roles and choose which agents join a channel. The lead agent then delegates tasks to the relevant teammates.