Hi, just started using openhands today (shifting from Cursor + Claude Code extension) and had a couple of quick questions: • Is it possible to tune the effort level in Open Hands while using a Claude subscription? • What's a quick and easy way to connect Open Hands to a HPC server? I usually ssh to my server and run tmux there. Thank you!!

Hello, what would be the best model to run with 48gb vram. Qwen3.6 35B A3B is substantially worse than even cheapest openai models.

Hello, is there any way to add a conversation from the dev_conversations to the conversations overview in the interface again? I have some conversations from a previous installation and want to add them to the overview.

The organization's budget or the budget associated with your user account has been reached. Please contact your OpenHands Enterprise administrator to request an override. what does that mean if im working with the free LLM profile: OpenHands/Deepseek-v4-flash

Whenever I send a message, OpenHands completely freezes for about 30 seconds before recovering, consistently throwing this error:

WARNING Unexpected error retrieving secret 'OPENHANDS_AUTOMATION_API_KEY': ReadTimeout: timed out secret_registry.py:291

This happens on every single prompt submission and affects other secrets as well. It looks like the process is getting blocked/deadlocked on internal secret retrieval until the 30-second HTTP timeout expires. Has anyone encountered this or know what might be causing it on Windows?

Hi everybody! I am new to openhands, and I am experimenting it for local development. So far I was able to succeed creating a couple of sample apps, but I also found a few issues that are hard for me to understand. My setup is: 2 hosts running ollama, one is running qwen3.5:9B and the other is running qwen3.8:27B (heavily quantized, due to memory constraints; but still capable of developing the 2 apps I mentioned above), one host running openhands canvas UI through docker on Debian 13. Unfortunately from time to time I get stuck and I am not able to understand what is happening. 1. when running condenser, on large context (>100k), I see in ollama that the task is reset after 50s from the first request, and then is tried again after a while. I tried setting 900s timeout in the LLM configuration, but it does not seem to affect results. Moreover I wanted to try to use qwen3.5:9B for context condensation, but I was not able to convince the condenser to use this model, neither by setting it as default nor by switching conversation model; 2. I have a couple of conversations that, instead of trying to contact the external hosts, openhands tries to connect to localhost and so I always get errors like

litellm.APIConnectionError: Ollama_chatException - Cannot connect to host localhost:11434 ssl:<ssl.SSLContext object at 0x7f99c22d0f50> [Connect call failed ('127.0.0.1', 11434)]

. What can cause these issues?

Hi, What idea about an error:

Disconnected (check URL or network). Check that the backend URL is correct and the backend server is reachable. If the backend is on another origin, check that it allows this frontend origin.

on Canvas site when be back from another tab (selfhost k8s)?

Error: 'PromptTokensDetailsWrapper' object has no attribute 'cache_creation_tokens'. How to resolve this? This occurs both in the chat interface and when saving configuration file changes. I installed the latest agent-canvas version via npm.

Is it normal that sometimes we need to pause and restart the conversation in order for the agent to process our latest message immediately? Otherwise, the agent continues with its unnecessary internal reasoning and struggles because it doesn’t take your message into account, even though your message clarifies everything. Is it simply finishing the last inference request, or is some kind of loop happening? Other agents interrupt the current inference request or provide a

/steer

command.

I'm designing the architecture of an app but want to create the front end before I do all the technical stuff to make it work. Is there a front-end UI designer. I'm familiar with Stitch and others and know that I can ask AI (and I have), but I would prefer some human insight. The application will be SaaS on Google Cloud. I don't have a software dev background; my background is wildly varied.

Hi everyone! Is there any integration with Linear agents for OpenHands?

Hi! Does anyone know who I should contact regarding a private security advisory submitted to OpenHands on GitHub? I'd really appreciate any guidance. Thanks!

Agent Canvas: child response returned once, but not on repeated attempts Hi everyone. I’m troubleshooting child responses in OpenHands Agent Canvas 1.25. My setup: • Canvas running locally in Docker on Windows 11. • A self-hosted

qwen3-coder-30b-fp8

model served by vLLM on a separate NVIDIA DGX Spark, connected through an OpenAI-compatible endpoint. • vLLM configured with

--tool-call-parser qwen3_coder

and

--enable-auto-tool-choice

. A simple parent–child smoke test worked once: the parent launched a child conversation, I watched the child run in Canvas, and its response returned to the parent. However, that success has not been repeatable despite many attempts. The problem I’m investigating is the child’s response not returning to the parent. Has anyone encountered this with Canvas child conversations? Which logs or events should I check to determine whether the child’s final response was delivered to the parent, versus the parent model failing to process it? Could the tool-call parser configuration affect this handoff? I’m asking specifically about Canvas child conversations, not SDK sub-agent delegation.

Hi all. I opened a draft PR adding a new backend that uses rootless podman,

PodmanWorkspace

, to the SDK. The closest alternative before this was

DockerWorkspace

pointed at podman, which starts the container, but the agent can't write to mounted project directories. This backend fixes that with

--userns keep-id

. PR: github.com/OpenHands/software-agent-sdk/pull/5665 Issue: github.com/OpenHands/software-agent-sdk/issues/5664 I'd appreciate feedback on the interface before I go further. There are three short questions under Notes in the PR. Thanks :)

Do we now get ratelimited?

Hi all. I thought I'd come here before raising an issue. My OpenHands container (latest, SDK v1.54.0) went down earlier today without any interaction on my end, and with nothing happening (no automations, no agents working, nothing). Now it won't come up - the following error appears in the container logs: > Error [ERR_MODULE_NOT_FOUND]: Cannot find module '/opt/agent-canvas/mars-web-bridge.mjs' imported from /opt/agent-canvas/static-server.mjs > at finalizeResolution (nodeinternal/modules/esm/resolve272:11) > at moduleResolve (nodeinternal/modules/esm/resolve879:10) > at defaultResolve (nodeinternal/modules/esm/resolve1006:11) > at #cachedDefaultResolve (nodeinternal/modules/esm/loader705:20) > at #resolveAndMaybeBlockOnLoaderThread (nodeinternal/modules/esm/loader725:38) > at ModuleLoader.resolveSync (nodeinternal/modules/esm/loader763:56) > at #resolve (nodeinternal/modules/esm/loader687:17) > at ModuleLoader.getOrCreateModuleJob (nodeinternal/modules/esm/loader607:35) > at ModuleJob.syncLink (nodeinternal/modules/esm/module job276:33) > at ModuleJob.link (nodeinternal/modules/esm/module job381:17) { > code: 'ERR_MODULE_NOT_FOUND', > url: 'file:///opt/agent-canvas/mars-web-bridge.mjs' > } > Node.js v24.21.0 I'm baffled on what might have happened. Unfortunately I messed with my docker compose, and I think I accidentally wiped out earlier container logs. I was wondering if anyone can tell me how to diagnose what happened? I could try a fresh container, but I don't want to throw away diagnostic data if I can help it.

Hi, I am struggling with hooks. I have a docker container running the backend (1.27.1), within the workspace directory I have the hooks.json and its related scripts (as explained docs.openhands.dev/openhands/usage/customization/hooks). The frontend is running externally (same version). I can successfully initiate conversations, but hooks seems to be never fired : • There are no logs from the scripts • The output of the container does not mention any hooks • In the "Show available hooks" menu from the conversation there is nothing I see that the SDK offers another implementation, but before exploring it, how can I investigate or fix the current setup? I really appreciate your help.

Hello everyone, How are you doing? I have an idea for a marketing collaboration with OpenHands. Who would be the best person to reach out to?

Greetings: I am seeing a behavior from OpenHands that I don't understand. OpenHands is grinding away developing and testing some code. I am using a local NVIDIA DGX spark as the backend. I see OpenHands "Thinking" for long periods while there are no requests to the spark. I would say that roughly 90% of the time while OpenHands is working on a task, there are no requests to the spark. Is this normal? What is it doing during the time when there are no requests to the spark?

Hi everyone, is there any way to use models from the Qwen Cloud token plan on OpenHands? I tried setting this up in the advanced configuration using the OpenAI-compatible URL, but I don't know the correct format for the model ID, with or without the prefix, like in LiteLLM. Anyway, using only the model ID from the provider's documentation is not working. Thanks for any help! 😄

Hi everyone, I have a question about OpenHands. Is it possible to self-host OpenHands and rebrand it as our own platform, while retaining the same functionality as the official hosted version? The idea is to provide users with a similar experience under our own branding, including the ability to configure a global AI API key that users can access without needing to provide their own keys, alongside the other features available in the hosted version. I’d also like to know whether this is technically feasible with the current codebase, whether there are any limitations compared to the official hosted version, and whether any commercial licensing fees or special permissions are required for this kind of deployment. Has anyone done something similar or can the OpenHands team clarify what would be required to achieve this?

Hey folks, I just reinstalled Open Hands after about a year of not using it due to work banning "shadow AI". What is the best way to disable any automatic

git

action on the part of the agent? I like to review everything and make the commits and pushes myself.

Has anyone tried computer use? Perhaps via an MCP server running in a docker container? I don't have any real reason to do this at the moment, I just thought it would be interesting. I see this: github.com/domdomegg/computer-use-mcp

Hi Team, I tried installing

vulnerability

Fixer. After cloning the repository with

git clone

, the

npm install

process took a very long time. I let it run for about 3 hours, but it still hadn't finished, so I had to stop it. Does anyone know what might be causing this? Is there a faster way to install the dependencies?

I'm noticing in 1.33 on Chrome that the browser tab isn't very reliable. It says

No page loaded yet. Ask OpenHands to open a URL. Example: "Open <https://example.com>"

even though the agent has browsed to a page. Anyone else run into this?

I've started playing around with a kanban style coding factory app for openhands. It's only integrated with GitHub so far, pulling issues from there and flagging when something needs attention. I'm just wondering if I'm wasting my time and something like this already exists?

My local agent keeps complaining about bot blocks when searching the web. I have searxng running locally on the AI machine and I've confirmed it works using Open WebUI. What is the best way to tell Open Hands about it? New skill? MCP server?

Hey hey all - we have an issue where we are trying to spin up openhands against a CloudSQL postgres DB but we are hitting a few issuess running the migrations against a blank DB. I've created this PR to fix those, can you please let me know if this makes sense?

I came across the ThetaCog extension, which is used in Visual Studio, say, with Claude Code. I'm using the /steer command. It seems to keep the specification in check. thetadriven.com Interested in what others think, as my background is not in software development.

Hi! I tried couple days now to setup this locally and i got ollama server running with gemma4 and qwen3.5, i have the openhands server running and i can chat with the agent. But… Whenever I say something it is so really bad, I must have done something wrong somewhere..? Bad LLM choice? I dont know… I got github connection status “OK” when testing connection for its mcp. I have enabled skills for github in profile.. But look at this prompt and answer … and when I answer the agents questions it’s like it dont remember asking me…