👋 Hey everyone! I'm looking for feedback from folks who are interested in, or are already using, OpenHands with their team! We’re exploring how teams use OpenHands Agent Canvas today, including interest in managing multiple Canvas backends and role-based access controls (RBAC). ➡️ If you use OpenHands with colleagues—or are considering it—please take this short survey: forms.gle/xMHWJuPVCpSpuiDo6 If you’re open to a follow-up conversation, you can opt in through the form. Your feedback will directly help shape our team-focused roadmap. Thank you!

Hi @calm-artist-7206 and @modern-fireman-99755 — could you help triage three of my long-open software-agent-sdk PRs? - #4413 fixes encrypted model profiles not being decrypted for subagents. The bug has now been independently confirmed by another user. CI is green, there are no unresolved review threads, and the latest review found no material issues. It overlaps with the older #4183 — could you help choose the preferred implementation? #4413 is smaller and was updated more recently. - #4434 adds opt-in Laminar instrument selection, reducing deferred initialization latency from 5.971s to 0.370s. It is mergeable, CI-green, approved, low-risk, and has no unresolved findings. Could you help merge it? - #4402 adds explicit MCP tool refresh for long-lived conversations after backend deployments. The earlier concerns around stateless MCP, recovery, and multiple clients have been addressed with real stateful and stateless deployment tests. It still needs a human maintainer to confirm the lifecycle and API boundary. Once that is confirmed, I can rebase it and rerun CI immediately. Thanks for your help!

Hii could someone help unblock my OpenHands PR? • #16913 fixes stale conversation history merging after switching conversations. • All tests pass. The only failing check is the new ready-for-dev gate. The linked issue #16875 has the required reproduction steps and acceptance criteria, but I cannot edit its body to move the reproduction screenshot into Actual Behavior. Could someone with access help update the issue body? Thank you!

Hey Folks! We have our monthly community call in just a few minutes. We've had an active month of contributions across the open source repos and I'll be highlighting some the updates to OpenHands Agent Canvas and in the SDK. If you can't make it check out the August Product release blog post we published yesterday on the openhands blog: openhands.dev/blog/new-in-agent-canvas-august-2026 OpenHands Community Meeting Thursday, September 3 · 12:00 – 1:00pm Time zone: America/New_York Google Meet joining info Video call link: meet.google.com/xvv-bdgs-snt 👋 See y'all there!

Could someone review #4838 and #4784? Both are small PRs with just a few lines of changes. The MCP reconnect issue in #4837 is affecting both us and another user in production. Thanks!

Wow, I didn’t even notice but OpenHands downloads have doubled for 4 straight months! x.com/ClickHouseDB/status/2097037214454587572

I've been using opencode go as my LLM provider and tried to pick up where I left off several days ago and found this new issue waiting for me. Aparently opencode go now requires special headers. Is there a way for me to configure this in the advanced section of the LLM config? github.com/aaif-goose/goose/issues/11900 and github.com/vastsa/PI-Desktop/issues/48

image.png

Everything was working fine a few days ago.

I'm on the latest OH

Error code: 401 - {'error': {'message': 'Authentication Error, Invalid proxy server token passed. Received API Key = sk-...Al2A, Key Hash (Token) =f8cae4dd484ac6602844a6c82ac8772e595a01ea5c979568309864f4ef15b3ea. Unable to find token in cache or `LiteLLM_VerificationTokenTable`', 'type': 'token_not_found_in_db', 'param': 'key', 'code': '401'}} Why am I hitting this error!

How do we get rid of this annoying "gettng started" box. I don't have or need some of the things it wants me to setup. There should be a dismiss button on this.

I just updated to 1.17 fyi

I don't want to see the "what do you want to build after I've already started a project" It shows this after every entry.

I keep getting this message about ever hour or so, using opencode as the llm provider and deepseek 4 flash.

litellm.BadRequestError: OpenAIException - Error from provider (Console Go): Upstream request failed: [invalid_request_error] The

reasoning_content

in the thinking mode must be passed back to the API.

I tell it to continue and it does but it will happen again.

This message was deleted.

👋 For anyone who has joined our slack recently there are a few channels you may be interested in joining for specific areas of interest: • #C06QT0AGY4W - OpenHands/OpenHands primary repo Agent Canvas. If you have any PRs you are working on, interested in working on, or just looking to chat about Agent Canvas this is where conversations are usually happening • #C06R25BT5B2 - OpenHands agent related things including the software-agent-sdk • #C0B4Z32L064 - Automation focus for OpenHands/automation which powers the automation server inside of Agent Canvas • #C0A6U0HTD7X - Conversations around the extensions we bundle with agent canvas including the plugins, automations, skills, and mcp catalogue • #C086ARSNMGA - General OpenHands feedback which is always great to here • #C0AB0FTCECA - If you are looking to share an open role or you are sharing when you are open to work this is the spot • #C06PB3T5ZK6 - Good spot for ~Everything else There are a handful of other channels as well but those are great spots to get a bit more focused engagement with folks!

📢 Hi all, we're aware of some users in OpenHands Cloud receiving LLM errors when attempting to use the free DeepSeek V4.1 model, or switching back to use the free DeepSeek V4.1 model. We are actively investigating this issue and will provide updates as we make progress. We apologize for the inconvenience here!

What causes this? This was one conversation, now it's split into a few and I'm getting duplicate messages?

What I've noticed is on long projects OH slows down a lot especially trying to type information in the box. It's slow and trails..reminds me of dial up. Then, after sometime of it coding I'll get the lost connection to the backend screen but a refresh has me back to the normal screen and now typing is fast again. No, nothing on the host is a issue, ram and cpu are not over consumed.

This is the docker image, one behind the latest I believe.

Hi everyone! I’m looking to speak with people who use Automations in Agent Canvas. I’d love to learn what you’re automating, how you’ve configured it, and what’s working—or not working—well. If you’re interested, please book a 30-minute conversation here or DM me directly. I can share how we use automations to power the open source repos handling code reviews, issue triage, automatic documentation, and memory curation for automation self-improvement.

hand

For anyone planning to join the community meeting today we've updated the video call link. It should be updated through Luma but if you have any issues here is the event link information: OpenHands Community Meeting Thursday, October 1 · 12:00 – 1:00pm Time zone: America/New_York Google Meet joining info Video call link: meet.google.com/sca-wzay-ena I'll make another channel ping a few minutes before just to make sure no one pops into the old link. Just getting this fixed up so we can auto record going forward. Thanks everyone!

👋 We are live on the community meeting for anyone who wants to join. We'll be looking back on releases in September and doing some demos!

Hi all, work done during my internship with Openhands (in collaboration with @wide-doctor-37830, Greg Durrett, Isil Dillig and @limited-student-10747). We worked on refining existing agent skills using traces already collected by the agent. This saves cost 1.4x-42x over other SoTA techniques while matching or beating performance on various tasks such as spreadsheet formula authoring, math reasoning, and PR review. Paper: alphaxiv.org/abs/2610.skillrefiner-offline-skill-refinement X: x.com/AnirudhKhatry/status/2106071467259543893 LinkedIn:linkedin.com/posts/anirudh-khatry_if-youve-deployed-an-agent-youre-already…?rcm=…&…

<!here> 🚀 OpenHands Agent Canvas 1.25.0 is out! This release makes Agent Canvas easier to configure, customize, and use day to day. Here are some highlights: • Bulk Add LLM profiles - Bulk-add provider models as LLM profiles, select the tools available to an agent profile, and give local profiles custom instructions or a custom persona. • Introducing the Model Router - Configure direct prompts, choose whether routing runs when a conversation starts, and use improved defaults for a smoother first-run experience. • Voice dictation in chat - An optional mic button lets you dictate into the composer, review or edit the text, and then send it when ready. It supports browser speech recognition or an OpenAI-compatible transcription endpoint. • Better handling for large local workspaces - Configure Files-panel discovery settings, including exclusions, symbolic-link support, and the file-count limit. Agent Canvas now warns when the displayed tree is incomplete. • Canvas Apps are easier to update - Update installed Canvas Apps from within Agent Canvas, and see app icons declared in an extension’s manifest. • Agent Profiles get more control - Select tools from the server catalog and give a profile its own persona or extra instructions. You can even override the OpenHands built-in persona or add on to it. • Polish, accessibility, and reliability - This release also includes 33 bug fixes and other improvements across onboarding, mobile/narrow layouts, settings, automations, rate-limit handling, and WebSocket reconnection behavior. See the full release notes for everything included or check the github tag to look through the code changes Thank you to everyone who contributed to this release: awss1i, enyst, errmakov, FOWEPJF255, georgeglarson, Harsh23Kashyap, henriquehirako, hieptl, jabrailkhalil, juanmichelini, KXHXK, kr1shna-exe, lilagrc, neubig, nicolasdmolina, qingshungLI, shinji00222, simonrosenberg, smolpaws, tofarr, VascoSch92, and wcy12378. A special welcome to first-time contributors: FOWEPJF255, shinji00222, jabrailkhalil, henriquehirako, nicolasdmolina, qingshungLI, kr1shna-exe, wcy12378, and awss1i