Struct SamplingLlmClient 

pub struct SamplingLlmClient { /* private fields */ }

LlmClient impl whose complete() calls back via the connected MCP client’s sampling capability.

Construct via SamplingLlmClient::new (production path: wraps a real Peer<RoleServer>) or SamplingLlmClient::with_sampling_client (test path: takes the abstracted SamplingClient trait object directly so [super::super::test_support::fake_mcp_client:: FakeMcpClient] can drive it).

Cheap to clone — every field is Arc-shared.

Implementations

impl SamplingLlmClient

pub fn new( peer: Peer<RoleServer>, write_handle: WriteHandle, audit_principal: Option<String>, ) -> Self

Build a client wrapping a real Peer<RoleServer>. Production path — called from crate::mcp::SoloMcpServer::populate_sampling_steward when an MCP session reaches initialize with a sampling-capable peer.

pub fn with_sampling_client( sampling_client: Arc<dyn SamplingClient>, write_handle: WriteHandle, audit_principal: Option<String>, ) -> Self

Test-friendly constructor accepting any SamplingClient implementation. Pair with [super::super::test_support::fake_mcp_client::FakeMcpClient] in tests.

pub fn with_max_tokens(self, n: u32) -> Self

Override the per-call max_tokens cap.

pub fn with_timeout(self, t: Duration) -> Self

Override the per-call timeout.

Trait Implementations

impl Clone for SamplingLlmClient

fn clone(&self) -> SamplingLlmClient

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl LlmClient for SamplingLlmClient

fn name(&self) -> &str

Backend identifier — “qwen3-coder-30b-local”, “claude-sonnet-4-6”, “gpt-5o”, etc. Used in dev-log entries and consolidation provenance (Provenance::by).

fn complete<'life0, 'life1, 'async_trait>( &'life0 self, messages: &'life1 [Message], ) -> Pin<Box<dyn Future<Output = CoreResult<Message>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Run a single completion turn. Implementations handle their own retries, rate limits, and context-window management.

fn is_real_llm(&self) -> bool

True iff this client talks to a real LLM backend (HTTP, FFI, local model), false for deterministic test stubs whose complete() returns canned data.