Hi everyone! The contribution guide recommends posting larger UI proposals in

#agent-canvas

, but I can’t see that channel, so I’m sharing this here. We’re exploring self-hosted OpenHands for our platform at Zoey, and one feature we’d find very useful is an interactive human terminal directly inside the conversation UI. We simply want a human to be able to access and interact with the same machine and workspace directly. This would be useful when the agent is taking time, when we want to inspect the current state, run a quick command, review files or test something ourselves. At the moment, we can see the terminal commands executed by the agent, but we cannot type commands ourselves. I found the older request https://github.com/OpenHands/OpenHands/issues/2643, but it’s closed. Is this already supported, planned or tracked somewhere else?

Hey, welcome around! That channel is #C06QT0AGY4W :) but this channel is perfectly fine!

I see where you are going. I also believe an agent is much stronger than any human working with the terminal. But in our context, speed matters, and sometimes asking an agent for dummy stuff is too slow compared to doing the little things yourself. For instance, checking your git history, git atomicity, or running a little test yourself is much faster. I don't know about you guys, but I'm not close to giving it away. I'm still faster without talking to an agent. Agentic UX is great, but as engineers, we still need close interaction with the software, since agents are too slow and sometimes wrong or dumb about the little details. Do you experience the same? Or am I alone?

What I often do is use the VSCode button and then I use its terminal window to interact with the container. Also works to upload and download files.

Yeah that's a workaround that works when you dev locally. But in my case, we want to integrate openhands into our platform with this feature without having a 10 minutes setup. It needs to be a first class feature.

Hi Romain--we do have an API endpoint that allows command execution

Hi Robert, I guess that command execution api endpoint is used by the agent to run commands? You don't encourage it in the UI, but does this mean that a UI for an interactive terminal exists? I already embedded OpenHands UI in a self managed cloud setup with modal. I did not use your cloud setup because I needed more freedom for LLM keys setup for our clients.

we do not have an interactive terminal UI no

Hello, everyone! Looking time lurking, first time poster here. We're building an AIOps CyberSec SRE platform for a hackathon in Argentina (and will continue developing it later). Our AI agent helps investigating production incidents and finding root cause focused on CyberSec (more info in the repo: github.com/annatchijova/zaynor). We'll provide a (web) chat interface to interract with agent. My question is similar to the TS's: I want to chat with my Agent about evidence it gathered and the reasons behind, usually after it finished its initial investigation. In rare cases I'd like to guide and help the agent with it.

You can always do that: chat with your Agent about what it did. One way is directly in the same conversation; another way is using

/btw

for a side-conversation. There are other alternatives too: you could “fork” a conversation, which makes a new conversation with the full history of the first; while the agent can continue on the older one as well. These are all in `agent-canvas`` web UI.