Expand description
Wraps an LLM call with Wardline input and output pipelines.
The reference path stays synchronous end to end: guard the prompt, make a
blocking model call (reqwest::blocking, ureq, a local function),
guard the response. This crate does not start an async runtime and does
not host a model.
prompt ──▶ input pipeline ──▶ llm_call ──▶ output pipeline ──▶ reply
│ │
└── Block: never call the model └── Block: never release the replyStructs§
- Blocked
- A pipeline refused the prompt or the reply.
Enums§
- Guarded
Error - Failure of
guarded_prompt: a guard said no, or the model call failed. - Stage
- Which side of the model call refused the work.
Functions§
- guarded_
prompt - Guards
prompt, callsllmonly if allowed, then guards the reply.