Connect OpenHands to Anoman AI
OpenHands (formerly OpenDevin) routes its LLM calls through LiteLLM, so it accepts any OpenAI-compatible base URL. Point it at Anoman to add PII masking, tool-call policy enforcement, and full observability to every agent step, plus prompt-injection detection when capacity allows.
Prerequisites
- An Anoman AI account with an active API key (
anm-sk-...) - OpenHands installed (docs.all-hands.dev)
Enable streaming for long agent runs
Set stream: true on requests. Autonomous, multi-step, and long-thinking tasks can exceed the 100-second edge timeout and fail with a 524 error; streaming returns tokens as they are generated and keeps the connection alive, so the request completes. Most tools stream by default — confirm it is on for agentic use.
More detail in the streaming guide.
Step 1: Set the LLM environment variables
OpenHands reads the model, base URL, and key from environment variables. Point the base URL at Anoman's OpenAI-compatible endpoint and use your Anoman API key.
Step 2: Or configure it in the UI
In the OpenHands Settings panel, open Advanced options and set the Base URL and API Key fields to the values above. Choose any model available on Anoman.
Step 3: Start a task
Run an OpenHands session. Each planning and tool-execution call now routes through Anoman — guardrails run before the provider call, and every request is metered and traced in your dashboard.
Verify the guardrails are active
Every Anoman response carries guardrail result headers. Inspect them in your dashboard traces to confirm the pipeline is running on OpenHands requests — PII masking and content moderation apply automatically, and prompt injection is scored when capacity allows.
Need help with your integration?
Reach the Anoman team directly with questions about your setup.