Expand description
A scriptable fake OpenAI-compatible server for tests (and loadmaster __fake-runtime).
By default it behaves like a correct runtime: tools in the request produce tool_calls
whose arguments are built from the tool’s parameter schema, response_format produces
JSON built from the schema, a tool message produces a final answer, and streaming
splits everything across deltas. Each Misbehavior breaks one of those the way real
runtimes do, so every failure class the probe knows can be exercised.
Structs§
Enums§
- Impersonate
- Which runtime’s identifying endpoints and headers to imitate.
- Misbehavior