Crate llmkit_core 

Core traits, types, and errors shared by every llmkit crate.

This crate has no I/O and no async runtime dependency. Provider adapters (llmkit-openai, llmkit-anthropic, llmkit-ollama) implement LlmProvider against these types; middleware and the facade crate build on top of them.

Re-exports

pub use error::LlmError;

pub use error::LlmResult;

pub use provider::LlmProvider;

pub use stream::ChatStream;

pub use stream::StreamDelta;

pub use tools::Tool;

pub use tools::ToolCall;

pub use tools::ToolChoice;

pub use tools::ToolResult;

pub use tools::ToolSchema;

pub use types::ChatRequest;

pub use types::ChatRequestBuilder;

pub use types::ChatResponse;

pub use types::ContentPart;

pub use types::EmbedRequest;

pub use types::EmbedResponse;

pub use types::FinishReason;

pub use types::Message;

pub use types::MessageContent;

pub use types::Role;

pub use usage::pricing;

pub use usage::CostEstimate;

pub use usage::ModelPricing;

pub use usage::TokenUsage;

Modules

error

Unified error type. Every fallible llmkit operation returns LlmResult<T>.

provider

The LlmProvider trait — one call site regardless of backend.

stream

Streaming response types.

tools

Tool/function-calling types, normalised across providers.

types

Request and response types, normalised across providers.

usage

Token usage and per-model cost estimation.