Crates

The workspace holds 16 packages at version 0.20.0. None is published to a registry: a project depends on them from Git at a release tag. A package name is its directory name; its library name uses underscores, so llm-core is use llm_core. This page is generated from cargo metadata by llm-docs; the overview says which crate to depend on for what.

Package	Library	Directory	What it is
llm-conformance	binary llm-conformance	checks/conformance	—
llm-bedrock	llm_bedrock	crates/llm-bedrock	Amazon Bedrock Runtime behind the model port: Converse and Anthropic InvokeModel, event-stream decoding and hand-written SigV4.
llm-blocking	llm_blocking	crates/llm-blocking	A blocking turn adapter over any asynchronous model port, for a synchronous agent loop.
llm-chat	llm_chat	crates/llm-chat	Chat Completions protocol projection and gateway ingress for the declared supported subset.
llm-cli	llm_cli	crates/llm-cli	The llm command: lists the models an llm.routes/1 configuration file declares and runs one streamed turn on any of them.
llm-core	llm_core	crates/llm-core	Neutral model requests, results, streaming, capabilities and typed failures; no I/O or execution authority.
llm-cost	llm_cost	crates/llm-cost	Usage accounting, versioned price estimates and explicit admission limits.
llm-credentials	llm_credentials	crates/llm-credentials	Injected secret resolution and caller-managed credential refresh; optional backend adapters.
llm-docs	binary llm-docs	crates/llm-docs	Generates and drift-checks the derived pages of the llm documentation site, and holds the programs its guides quote.
llm-http	llm_http	crates/llm-http	Shared bounded HTTP and SSE transport; no vendor fields or credential acquisition.
llm-messages	llm_messages	crates/llm-messages	Messages protocol projection and gateway ingress for the declared supported subset.
llm-models	llm_models	crates/llm-models	Builds the Chat Completions, Responses or Messages port a catalog serving model declares, with the caller's resolver for its account, and reads llm.routes/1 model routes into a catalog and resolvers.
llm-providers	llm_providers	crates/llm-providers	Provider/account and authentication bindings independent of protocol selection.
llm-responses	llm_responses	crates/llm-responses	Responses protocol projection, gateway ingress and a single-attempt client for the declared supported subset.
llm-routing	llm_routing	crates/llm-routing	TOML catalogs, capability-aware resolution and explicit ordered fallback.
llm-tool-call	llm_tool_call	crates/llm-tool-call	One forced tool call on any model, returning the tool's JSON input, and a Responses model preset over the operator's Codex login.

Optional features

Every feature is off by default unless the package lists it as a default. A feature adds the optional dependencies named beside it and nothing else.

Package	Feature	Turns on	Adds dependencies
llm-bedrock	process	tokio/process	—
llm-cost	sqlite	—	rusqlite
llm-credentials	codex-auth-file	—	base64, rustix, serde_json, sha2
llm-credentials	codex-renewal	codex-auth-file	llm-http, tempfile
llm-credentials	environment	—	sha2
llm-credentials	file	—	rustix, sha2
llm-credentials	json-pointer	—	serde_json, sha2
llm-credentials	keychain	—	keyring-core, sha2
llm-credentials	native-keychain	keychain	zbus-secret-service-keyring-store, apple-native-keyring-store, windows-native-keyring-store
llm-credentials	secrets	—	secrets-core, secrets-keychain, keyring-core, sha2
llm-models	environment	llm-credentials/environment	—
llm-models	file	llm-credentials/file	—
llm-models	keychain	llm-credentials/keychain	keyring-core
llm-models	process	llm-bedrock/process	—

No package turns on a feature by default.