Module providers 

Structs

ChannelRef

Reference to a configured [channels.<type>.<alias>] entry.

ModelCostRatesByProvider

[cost.rates.providers.models.<type>.<model>] — token-cost rates per (provider type, model). One slot per provider type; each slot is a HashMap<model_id, ModelCostRates>. The slot list matches ModelProviders byte-for-byte (same source macro).

ModelProviderRef

Reference to a configured [providers.models.<type>.<alias>] entry.

ModelProviders

Typed model provider container — one slot per canonical model_provider type.

Providers

Top-level wrapper for every provider category. TOML root sees a single [providers] table with one sub-key per category:

TranscriptionCostRatesByProvider

cost.rates.providers.transcription.<type>.<model> — per-(provider type, resource) cost rates.

TranscriptionProviderRef

Reference to a configured [providers.transcription.<type>.<alias>] entry.

TranscriptionProviders

Typed transcription-provider container — one slot per STT family. Mirrors ModelProviders / TtsProviders. Closed set of 6 families: groq, openai, deepgram, assemblyai, google, local_whisper.

TtsCostRatesByProvider

cost.rates.providers.tts.<type>.<voice> — per-(provider type, resource) cost rates.

TtsProviderRef

Reference to a configured [providers.tts.<type>.<alias>] entry.

TtsProviders

Typed TTS-provider container — one slot per TTS family. Mirrors ModelProviders but smaller (TTS has a closed set of 5 families: openai, elevenlabs, google, edge, piper). No catch-all needed.

Constants

MAX_FALLBACK_DEPTH

Hard ceiling on providers.models.<alias>.fallback chain depth. The cycle guard only bounds chains that loop; a long acyclic chain would otherwise recurse one stack frame per alias at config-load and build time, turning a pathological config into a startup stack overflow. Both the validation walk and the runtime build walk stop descending past this depth and prune the rest of the branch.