Crate atomr_infer 

atomr-infer

Multi-runtime GPU + remote inference as a supervised actor system on top of atomr and the backend-agnostic atomr-accel compute substrate. See docs/architecture.md for the design.

This crate is a rollup: it re-exports the public surface of the workspace’s per-runtime crates behind feature flags so downstream consumers depend on a single crate (inference) and pick the backends they need at compile time.

Pure-remote builds

cargo build -p inference --features remote-only

produces a binary that compiles no GPU dependencies at all — useful for pure-remote routers (a deployment that fronts OpenAI / Anthropic / Gemini / LiteLLM with rate limiting, fallback and observability but owns no hardware).

Re-exports

pub use atomr_infer_core as core;

pub use atomr_infer_runtime as runtime;

Modules

prelude

Re-export the most commonly used types so callers can use atomr_infer::prelude::*; and have everything they need to declare Deployments and write actors.