Umans Provider
Models
Section titled “Models”thndrs uses Umans Code as its first provider. The default model is
umans-coder; umans-glm-5.2 is available as an alternate model.
umans-coder is the default coding model. umans-glm-5.2 is useful when you
want the GLM model path or its larger context behavior.
Authentication
Section titled “Authentication”Set UMANS_API_KEY in the environment. thndrs does not accept API keys in CLI
flags.
export UMANS_API_KEY=sk-...Messages API
Section titled “Messages API”The provider client uses the Anthropic-compatible Umans Messages API:
- Base URL:
https://api.code.umans.ai - Endpoint:
POST /v1/messages - API key header:
x-api-key - Version header:
anthropic-version: 2023-06-01
Streaming Events
Section titled “Streaming Events”Provider streaming output is normalized into app events. Assistant text, reasoning, tool calls, completion, and errors are kept as distinct event types so the transcript can render them separately.
Tool Schemas
Section titled “Tool Schemas”Local read-only tools are sent as provider-native tool schemas. Tool descriptions stay short and focus on purpose, safety limits, and truncation behavior.
Error Mapping
Section titled “Error Mapping”Provider errors are mapped into transcript errors and return the prompt to a usable state.
Model Metadata
Section titled “Model Metadata”thndrs can read Umans model metadata from /v1/models/info for visible model
capabilities.
Smoke Test
Section titled “Smoke Test”The live provider smoke test is ignored by default and requires UMANS_API_KEY
plus network access.