Anyone used it?

[Killea]

Has anyone used it?

[bertheto]

Hi @Killea !
Yes, I've been using AgentChatBus intensively — and honestly, thank you so much for building this. It's a genuinely brilliant concept and the implementation is really solid. I'm thrilled to be collaborating with you and super excited about all the future features we'll be putting in place together!
Here's a quick overview of what I've set up on my end inside Cursor:
I built a full orchestration layer that uses AgentChatBus as the communication backbone for structured multi-agent reviews. The core idea is agent pairs — two specialized subagents with opposing or complementary perspectives (e.g. an advocate vs. an auditor) — that debate code changes through ACB threads before anything gets committed.
Concretely:
I defined 5 domain-specific review pairs (SCSS/design tokens, test coverage, AG-Grid patterns, Signals migration, i18n) — each pair is registered and communicates via its own ACB thread
A Cursor slash command (/acb quick-review [key]) bootstraps the whole flow: registers both agents, opens a thread, seeds the context, and lets them debate
The agents exchange structured arguments through ACB's msg_post / msg_wait cycle, with a final synthesis at the end
I also use context variables per thread (introduced in v0.1.2) to pass shared metadata like the diff, file paths, and review scope — which made the whole thing much cleaner
The result is that before any significant commit, I get a genuine two-sided review from domain experts that surface trade-offs I wouldn't have caught alone. It's become a core part of my daily workflow.
Can't wait to see where this project goes. Keep up the amazing work!

[Killea]

Wow, you are an expert!

[Killea]

Thank you so much for the kind words and your work (PRs) — I really appreciate it. It’s awesome to hear how deeply you’ve integrated AgentChatBus and your agent-pair review workflow in Cursor, which sounds genuinely impressive (especially the structured debate + synthesis and the per-thread context variables). I’m excited to keep collaborating and to build the next set of features together!
By the way, not sure if you have time to share your workflow in detail, but if you have some time, of course, like adding a section in Read me, or post some blog article or video. I'm sure people are curious about how an expert will use this MCP. 😂

[Killea]

I use this myself as well, in my real job. I'm happy that an agent can correct the other one.😂
Like today, an agent said he found an edge case in the chat after he read the previous agent's comments.
This feels great.

[bertheto]

Thank you so much for the kind words — it really means a lot coming from the maintainer! 😊

I'm genuinely excited about this collaboration too. AgentChatBus fills a real gap for structured multi-agent workflows, and I still have quite a few ideas brewing (thread tags & filtering, minimap navigation, export improvements, and more on the orchestration side). Looking forward to building them together!

As for sharing the workflow — yes, absolutely, that's something I'd love to do. The agent-pair review pattern (one agent proposes, the other challenges, synthesis emerges from the debate) is something I've found surprisingly effective in practice, and I think the community would find it valuable. A dedicated section in the README, a blog post, or a short demo video all feel like great options. I'll think about the best format and get back to you on that.

Thanks again for building something that makes this kind of experimentation possible!

[Killea]

Great, thank you.
I added bus_connect and changed some logic to make the agent chat quicker.
Also, I think there should be ablity in the system to disable/enable the detailed features that have an impact on LLM's attention_mechanisms. Perhaps I'll add later.
Getting the correct agent status is also very hard.

[Killea]

My copilot has almost used up, 67%, it's only the 2nd day of this month.

[bertheto]

Same on my side 😁