Struct MessageRequestBuilder 

pub struct MessageRequestBuilder<'a> { /* private fields */ }

Fluent builder for Claude API message requests.

Created via ClaudeClient::messages(). Chain methods to configure, then call .send() for the full response or .stream() for SSE events.

Implementations

impl<'a> MessageRequestBuilder<'a>

pub fn model(self, model: &str) -> Self

Set the model by ID (e.g. “claude-opus-4-6”)

pub fn opus(self) -> Self

Use Claude Opus 4.6 - most intelligent, best for agents and coding

pub fn sonnet(self) -> Self

Use Claude Sonnet 4.6 - best speed/intelligence balance

pub fn haiku(self) -> Self

Use Claude Haiku 4.5 - fastest and cheapest

pub fn system(self, text: &str) -> Self

Set the system prompt (plain text)

pub fn system_cached(self, text: &str) -> Self

Set the system prompt with prompt caching enabled

pub fn user(self, text: &str) -> Self

Add a user text message

pub fn user_with_image_base64( self, text: &str, media_type: &str, base64_data: &str, ) -> Self

Add a user message with an image (base64 encoded)

pub fn user_with_image_url(self, text: &str, url: &str) -> Self

Add a user message with an image URL

pub fn assistant(self, text: &str) -> Self

Add an assistant text message (for multi-turn conversations)

pub fn messages(self, msgs: Vec<Message>) -> Self

Set the full message list (for agentic loops where you manage messages)

pub fn tool_results(self, results: Vec<ContentBlock>) -> Self

Add tool results as a user message (for the agentic tool loop)

pub fn thinking_adaptive(self) -> Self

Enable adaptive thinking (recommended for Opus 4.6 / Sonnet 4.6). Claude dynamically decides when and how much to think.

pub fn thinking_enabled(self, budget_tokens: usize) -> Self

Enable thinking with a fixed token budget (older models only). budget_tokens must be less than max_tokens (minimum 1024).

pub fn thinking_disabled(self) -> Self

Explicitly disable thinking

pub fn effort(self, level: &str) -> Self

Set the effort level: “low”, “medium”, “high” (default), or “max” (Opus only). Lower effort = cheaper/faster. Higher effort = deeper reasoning.

pub fn tools(self, tools: Vec<Tool>) -> Self

Set the available tools for Claude to call

pub fn tool_choice_auto(self) -> Self

Let Claude decide whether to use tools (default)

pub fn tool_choice_any(self) -> Self

Force Claude to use at least one tool

pub fn tool_choice_specific(self, name: &str) -> Self

Force Claude to use a specific tool by name

pub fn tool_choice_none(self) -> Self

Prevent Claude from using any tools

pub fn json_schema(self, schema: Value) -> Self

Constrain the response to match a JSON schema

pub fn max_tokens(self, n: usize) -> Self

Maximum tokens to generate (default: 4096)

pub fn temperature(self, t: f32) -> Self

Sampling temperature (0.0 = deterministic, 1.0 = creative)

pub fn top_p(self, p: f32) -> Self

Nucleus sampling parameter

pub fn top_k(self, k: usize) -> Self

Top-k sampling parameter

pub fn stop_sequences(self, seqs: Vec<String>) -> Self

Custom stop sequences

pub fn user_id(self, id: &str) -> Self

Set a user ID for tracking/billing

pub fn beta(self, header: &str) -> Self

Enable a beta feature by header string (e.g. “compact-2026-01-12”)

pub fn build(self) -> MessagesRequest

Build the request without sending (useful for inspection/testing)

pub async fn send(&self) -> Result<MessagesResponse, ClaudeApiError>

Send the request and return the full response (non-streaming).

pub async fn stream(&self) -> Result<SseParser, ClaudeApiError>

Send the request with streaming, returning an SSE event parser. Use parser.next_event() to consume events one at a time.