OpenHands¶
OpenHands configures its LLM via a config.toml file
(or the Settings UI). Point it at Mango Inference's
Anthropic-compatible endpoint with a custom base
URL, model, and API key.
Prerequisites¶
- A Mango Inference API key.
- OpenHands installed (CLI or Docker).
1. Configure the LLM¶
[llm]
model = "anthropic/zai-org/GLM-5.3"
base_url = "https://api.mangoboost.io/v1"
api_key = "<your-api-key>"
The anthropic/ model prefix routes the request through OpenHands' Anthropic
client, which sends the API key in the x-api-key header, matching how
Mango Inference authenticates /v1/messages requests. See
Authentication.
In the Settings UI, the equivalent fields are Custom Model, Base URL, and API Key under a custom LLM provider.
The anthropic/ prefix is LiteLLM's, and it moves
OpenHands routes models through LiteLLM, so the prefix is what selects the
client, and the naming convention is LiteLLM's to change. Check
all-hands.dev for your version if the model is not
picked up. What matters on the Mango Inference side is unchanged: the
request must reach /v1/messages with the key in x-api-key.
OpenHands is an agent, so it leans on tool calling throughout. That depends on
the model you configure being trained for it rather than on the platform. See
Tool calling before choosing one. Note also that
/v1/messages/count_tokens is served, so context and cost estimates that
depend on it work normally.
2. Start OpenHands¶
openhands
Confirm the agent responds using the configured Mango Inference model.