ask_llm
Layer for llm requests, generic over models and providers
Usage
Provides 2 simple primitives:
oneshot and conversation functions, which follow standard logic for llm interactions, that most providers share.
Then the model is automatically chosen based on whether we care about cost/speed/quality. Currently this is expressed by choosing Model::{Fast/Medium/Slow}, from which we pick a model as hardcoded in current implementation.
Semver
Note that due to specifics of implementation, minor version bumps can change effective behavior by changing what model processes the request. Only boundary API changes will be marked with major versions.