Hacker Newsnew | past | comments | ask | show | jobs | submit | swiftlyTyped's commentslogin

Introducing AG-UI; the Agent-User Interaction Protocol.

AG-UI is an open, lightweight, event-based protocol that standardizes how AI agents connect to fronted-end applications.

After a year of one-off, CopilotKit <> Agent frameworks collaborations, we decided to ship a standard, and open way for agents to connect to apps.

We are launching with day 0 integrations with LangChain, Mastra, CrewAI and AG2 with more partnerships on the way.

How AG-UI works: - Creates a unified, two-way bridge between backend agents and frontend apps - Works with any event transport (SSE, WebSockets, webhooks) - Supports 16 standardized event types for common agent-user interactions

Where AG-UI fits in the agent protocol landscape: - MCP: for agent←→tools communication - A2A/ACP: for agent←→agent communication - AG-UI: for agent←→human interaction in applications

We launched the protocol yesterday, and have been blown away by the reception. We have the first working group this Friday, to help expand and steer the direction of the protocol.

Check out the repo to get started!


Looks cool!


This is awesome!


Thanks :)


Awesome! Congrats on the launch


I've been actively using Potiz since it's been in Beta and really enjoy it. Great to see how far it's come along.


yay


Thanks!

Co-agents help the user intervene at any step in the agent process / steer the agent. So the agent doesn’t need to be perfect to be useful.

Similar to the UX that Devin has but for any product.

We’re starting with LangChain / LangGraph integrations.


So basically a human feedback in the middle to correct the response in case desired.


Yes exactly


Really great stuff.

People use the same model / server for all queries not because it's sensible, but because it's simple. This brings the same simplicity to the far more optimal solution.

And great startup play too, by definition no incumbent can fill this role.


Thanks - glad to hear the idea resonates!


Hi, CopiotKit CEO here (I wrote the original viral tweet). This article is great! Thanks for posting.

I'd also written an analysis of the code - including announcing a $1000 prize for the best alternative code: https://ai88.substack.com/p/ceiling-has-been-raised-analyzin...

We were going to announce the winner this week but if we get a few more submissions we will definitely consider them.

Just submit a PR to https://github.com/CopilotKit/CopilotKit


Could you help me understand which Copilots are involved in this?

Your tweet at https://twitter.com/ataiiam/status/1765089261374914957 mentions "Cursor's copilot".

The blog post at https://rtpg.co/2024/03/07/parsing-copilots-type-spaghetti/ talks about GitHub Copilot - did they make a mistake there?

And your product is CopilotKit - is that related to the GitHub and Cursor Copilots in some way or is it something different?


Hah, too many copilots… let me try to clarify:

We are building CopilotKit = a framework + platform for building context-aware AI assistants into any application (not necessarily coding related applications).

Part of CopilotKit is about giving the Copilot / AI agents access to the application through a typed "inline realtime API". And to make ergonomics great for _our_ users, CopilotKit ships with hardcore type programming.

Cursor's Copilot (unrelated to CopilotKit) helped us write this type code (that's what went viral). And yes, GitHub Copilot is a mis-attribution.


I'm going to make a completely fresh alternative written from the ground up. When I'm done you can all find it easily because it's called CoPilot, shouldn't be hard to find.


The people behind Google's chat product naming could really learn a lot from this situation


Indeed. For example, they could learn that Copilot is a winning name so lets go with that.


You love swift ??


That's free training/fine-tuning material, right?


By the way - we practice what we preach- Copilots raise the bar (or the ceiling...) on human productivity - in every domain. Which is why we're building infrastructure to make building copilots easier...

What ideas do you think are still missing from today's Copilots?

I.e. suppose we were looking back at today's Copilots 5 years from now- besides better models, what else has changed?


I wrote the article (not the code - though I was part of it).

The code is is 30 lines of conditional, recursive, partially-type-inferred Typescript type expression. It's actually not too bad when you break it down.

What it enables: Zod-like safety + convenience, over plain JSONs.

Hope it's an interesting read


Three experiments looking into how GPT4, Gemini & LLaVA perform when getting conflicting visual and textual information. Some pretty interesting results.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: