The best LLM for OpenHands
OpenHands takes a task and runs with it — writing code, executing it in a sandbox, reading the output, and iterating, often for many minutes unattended. That's the single workload most likely to expose a model backend's weak spots: provider rate limits and empty tool turns. CodeBurst gives OpenHands an OpenAI-compatible endpoint that fails over across providers and repairs tool turns, so a long run reaches the finish line.
Long unattended runs are where backends fail
A quick prompt rarely trips anything. A 20-minute OpenHands task firing call after call is a different story:
- Rate limits mid-task. Sustained call volume hits a provider's cap, and the whole task aborts with no graceful recovery.
- Empty tool-synthesis turns. A reasoning model returns blank on the turn it should fold a command's output back into a plan — and the agent loop stalls.
Because the run is unattended, you often don't notice until you come back to a half-finished task.
What CodeBurst adds
| Failure | CodeBurst |
|---|---|
| Provider rate-limits mid-task | Reroutes to a healthy provider in the same request — the run continues. |
| Empty tool-synthesis turn | codeburst-agent retries with a corrected format. |
| Context grows past the window | Route a compaction step through codeburst-compress (size-aware). |
Configure OpenHands
OpenHands uses LiteLLM, so reference the OpenAI-compatible endpoint with the openai/ prefix. In config.toml:
[llm]
model = "openai/codeburst-agent"
base_url = "https://codeburst.ai/api/v1"
api_key = "YOUR_CODEBURST_KEY"Or set them as environment variables instead:
export LLM_MODEL="openai/codeburst-agent"
export LLM_BASE_URL="https://codeburst.ai/api/v1"
export LLM_API_KEY="YOUR_CODEBURST_KEY"Nothing else about your OpenHands setup changes — it just stops depending on a single provider being up for the whole run.
Get started
Get an API key Best LLM for AI agents →FAQ
How do I set CodeBurst as the OpenHands LLM?
In [llm]: model="openai/codeburst-agent", base_url="https://codeburst.ai/api/v1", your key — or the LLM_* env vars.
Why a router for OpenHands?
Long unattended runs hit rate limits and empty tool turns; CodeBurst fails over and repairs them so the task finishes.
Does the openai/ prefix matter?
Yes — OpenHands uses LiteLLM slugs, so it's openai/codeburst-agent.