Module completion 

Provider-agnostic completion requests and responses. A provider’s wire translates CompletionRequest into its own request and normalizes the reply as CompletionResponse; a Model sends it.

use rig_core::{Model, completion::CompletionRequest, providers::openai::OpenAIConfig};

let model = OpenAIConfig::from_env()?.connect(http).completion("gpt-4o");
let request = CompletionRequest::new("What is Rig?");
let response = model.call(request).await?;
println!("{:?}", response.choice);

Re-exports

pub use cache_cost::CacheCost;

pub use cache_cost::CacheRates;

pub use handle::ModelRef;

pub use history::Accepts;

pub use history::LaterSystem;

pub use history::Media;

pub use history::Pairing;

pub use history::Place;

pub use history::Replay;

pub use history::ReplayTarget;

pub use history::adapt;

pub use message::AssistantContent;

pub use message::AssistantMessage;

pub use message::Message;

pub use message::MessageError;

pub use options::CacheRetention;

pub use options::CheckError;

pub use options::Effort;

pub use options::GenerationOptions;

pub use options::OnUnsupported;

pub use options::Reasoning;

pub use options::ServiceTier;

pub use options::UnsupportedOption;

pub use options::Verbosity;

pub use prompt::Section;

pub use prompt::SystemPrompt;

pub use provider_options::ExtensionOptions;

pub use provider_options::OptionsError;

pub use provider_options::ProviderExtension;

pub use provider_options::ProviderOptions;

pub use provider_options::ReplyExtras;

pub use provider_options::SHARED;

pub use request::*;

Modules

cache_cost

What prompt caching cost and saved, priced the same way on every provider.

handle

Serializable model labels resolved to live handles by a runtime.

history

Shaping a conversation for the model it is sent to. adapt is the one place history meets its target: a turn the same model produced keeps its provider items, any other turn replays from its canonical fields, failed turns are skipped, and every tool call gets an answer.

message

options

Portable generation options: provider-neutral knobs a request sets once and every completion wire maps to its own JSON, or refuses. They hold only what CompletionRequest has no field for, so temperature, max_tokens, tool_choice and output_schema stay on the request.

prompt

A system prompt as data: a role and named sections, rendered into one system message only when the request is sent, so a layer in between can read, change or move a section by its tag.

provider_options

Typed per-provider request options and reply extras. A provider’s ProviderExtension marker names its key, its serialize-only Options and its Extras view of a reply. ProviderOptions holds entries for several providers at once, and each wire reads only the entry named by its own provider.

request

Completion requests, normalized responses, and provider model contracts.