Skip to main content Crate llmshim Copy item path Source pub use fallback::completion_with_fallback ;pub use fallback::completion_with_fallback_and_policy ;pub use fallback::FallbackConfig ;pub use llmshim_catalog as catalog;breaker Provider circuit breaking. cache Caller-owned stability annotations, provider-owned placement mechanics. client config Configuration file management for llmshim. cost USD cost accounting for a completed response. credentials Where a provider key comes from when the process environment does not carry one. env Shared environment/config loading. error fallback log models Curated, verified discovery. Dynamic metadata lives in crate::catalog .
Unverified facts stay Support::Unknown / None; never guess a number. policy Trusted, out-of-band policy hooks for individual provider attempts. provider providers reasoning Typed, lossless reasoning and a single fail-closed replay policy.
Opaque data is never inspected to decide where a block may be sent. router schema One schema walker with per-transport policy. External references are never
fetched; unresolved/cyclic schemas fall back per tool rather than per request. shim Capability-based request plans. Internal protocols are decoded and validated
before a response, stream event, or log entry can reach the caller. streaming Per-response stream normalization, shared by HTTP dispatch and manual SSE
consumers. Tool arguments, ids and signatures are assembled before exposure. toolcall Owned tool-call identities, persisted wire mappings, and conversation checks. usage Shared cache accounting. Native counters are read once at the transport
boundary; consumers never need to recognize a provider’s usage dialect. vision Vision/image content block translation between providers. SHARED_CLIENT Shared HTTP client — reuses connection pool across all requests.
Shared HTTP client with connection pooling. completion Top-level entry point. Resolves the provider from the model string and fires the request. completion_with_logger Completion with optional logging. completion_with_logger_and_policy completion_with_policy stream Streaming entry point. Returns an SSE stream of OpenAI-format chunks. stream_with_policy warmup Pre-establish TCP+TLS connections to all configured provider endpoints.
Call once after creating the Router to eliminate cold-start latency on first request.