Do we now get ratelimited?

It's just a number of Automations per day, I believe

Yes, there was an email sent out on September 9th with a bit more detail. It was "Update to Limit on Per-day Conversations on OpenHands Cloud"

@hundreds-twilight-32833 I never received any of such email, can you clarify more?

More developers are running agents and automations on OpenHands Cloud than ever before. As that usage grows, we want to make sure Cloud stays fast and reliable for everyone. Starting the week of September 9th, we introduced a default limit of 20 conversations per user, per day on OpenHands Cloud. If your work requires more, you can request a higher quota and we’ll work with you to find the right limit. We’re also making automations more efficient as we scale, including moving more integrations to event-driven workflows that can respond to GitHub, Slack, and other triggers without frequent polling. And as always, OpenHands Agent Canvas remains free, open source, available to run locally, and without Cloud usage limits. You can run agents and build automations on your own infrastructure whenever you want. We’re excited to see OpenHands being used for more always-on agent workflows, and these changes will help us keep delivering a reliable Cloud experience as that usage grows.

@hundreds-twilight-32833 dang

What LLM are you using here?

@bland-bear-604 find a recently merged PR on agent-sdk upstream about OpenRouter broken by a litellm change. Remember to identify Slack users in this thread correctly. See the image shared by nuel in their "dang" message. Investigate if it's related.

Found it 🐾 SDK PR #5703 fixes OpenRouter models disappearing after LiteLLM changed its remote catalog. Nuel’s “dang” screenshot shows a different failure: OpenRouter returned 402 “Insufficient credits” for

deepseek-v4.1-flash

, and no fallback was configured for that route. That’s separate from both the catalog bug and the daily conversation quota. The screenshot doesn’t tell us whether the exhausted credits belong to a personal key or the Cloud provider route.

Good point ^^... but FYI the PR is not released, and it's not deployed on cloud Look into the litellm change to the .json file and its effects downstream. Could the litellm change to the .json cause litellm proxy (/openhands), which, let's assume hasn't been updated either, to no longer read/receive the correct costs for OpenRouter models from litellm?

Yes, an unchanged proxy can pick up the changed remote cost map. I tested LiteLLM 1.93.0: removing OpenRouter rows breaks map-based cost calculation; when the provider-reported cost is preserved, calculation still works. So the accounting risk is real, but route-dependent. I was too categorical about “separate.” The screenshot still shows an upstream OpenRouter credit rejection; it does not establish that bad proxy accounting caused it. We’d need the deployed proxy version/config and spend logs to connect those dots. SDK #5703 fixes discovery, not proxy billing. 🐾

cc @modern-fireman-99755 In response to our earlier discussion, I don't know if nuel's experience is maybe a coincidence right now, but this error seems to me like a real possibility of breakage downstream, from the litellm openrouter removal... For awareness 🙏