voxora-traits

Canonical traits and value types for the voxora model-agnostic ASR bridge.

This crate is the canonical home of the public API surface. It has zero runtime dependencies beyond async-trait and thiserror (no tokio, no reqwest, no http) so it builds offline and stays small.

Removed predecessor: `voxora-core`

voxora-core was a thin re-export shim from voxora 0.3.0 to 0.3.1 that allowed downstream code to import the trait surface via use voxora_core::*; while the canonical home was this crate. The shim was deprecated in 0.3.1 and removed in 0.4.0. New code has always depended, and continues to depend, on voxora-traits directly.

Upgrade guide — `voxora-core` (deprecated) → `voxora-traits` (canonical)

// before — voxora-core shim (0.3.0 / 0.3.1)
use voxora_core::AsrEngine;
use voxora_core::TranscribeOptions;

// after — voxora-traits directly (0.4.0+)
use voxora_traits::AsrEngine;
use voxora_traits::TranscribeOptions;

In your Cargo.toml, swap voxora-core = "0.3" for voxora-traits = "0.4". If you used the voxora-core/serde Cargo feature, switch to voxora-traits/serde.

Surface

Item	Kind	Description
`AsrEngine`	trait	Synchronous, `Send + Sync` inference contract
`ModelSource`	trait	Asynchronous, `Send + Sync` model acquisition contract
`AsrError`	enum	Every error a voxora operation may return
`ModelCapabilities` / `TranscribeOptions` / `TranscriptionResult` / `TranscriptionSegment`	struct	Inference value types
`ModelDescriptor` / `ModelDir` / `ModelSourceKind` / `Quantization` / `QuantizationPreference` / `ResolveOptions`	struct / enum	Source value types

ASR-specific

The design is ASR-specific per operator direction. The AsrEngine trait is purpose-built for automatic speech recognition. New domains (LLM, vision, multimodal) are out of scope per the architecture handoff.

License

Apache-2.0.

voxora-traits 0.4.0