Struct ContextPipeline 

pub struct ContextPipeline { /* private fields */ }

Runtime context pipeline that composes context from multiple sources.

The pipeline:

1. Loads session history from the store
2. Applies compaction filter (reorder post-compaction messages)
3. Invokes registered context adapters (system prompt, RAG, etc.)
4. Applies token-budget trimming as a safety net
5. Produces a final ChatRequest or ContextOutput

Implementations

impl ContextPipeline

pub fn new() -> Self

Creates a new context pipeline with default settings.

pub fn with_factory(factory: ContextFactory) -> Self

Creates a context pipeline with an existing context factory.

pub fn with_max_history(self, max: usize) -> Self

Sets the maximum number of history messages to include as a fallback limit.

This limit applies before token-based trimming. Default: 50.

pub fn with_max_history_tokens(self, max: usize) -> Self

Sets the maximum token budget for historical messages before trimming.

Older messages are dropped first when the budget is exceeded. Default: 64,000 tokens.

pub fn with_compaction_filter(self, enable: bool) -> Self

Enables or disables the compaction message filter.

When enabled, compacted message pairs are replaced with a synthetic checkpoint system message. Enabled by default.

pub fn with_cache_strategy(self, config: PromptCacheConfig) -> Self

Sets the prompt cache strategy used to auto-place cache breakpoints.

When cache_strategy.enabled is true and cache_strategy.auto_breakpoints is true, the pipeline places up to three cache markers on each ChatRequest:

1. The end of the last system message (if it has at least min_system_tokens tokens).
2. The end of the conversation tail (the last message before the current user turn).
3. Each tool definition in the request.

pub fn register<A>(&mut self, adapter: A)

where A: ContextAdapter + 'static,

Registers a context adapter.

pub fn register_arc(&mut self, adapter: Arc<dyn ContextAdapter>)

Registers an already shared context adapter.

pub fn adapter_names(&self) -> impl Iterator<Item = &str>

Returns adapter names in registration order.

pub async fn build( &self, store: &RuntimeStore, session_id: Uuid, model: ModelName, user_message: Option<&str>, tools: Option<&[ToolSpec]>, ) -> RuntimeResult<ChatRequest>

Builds a ChatRequest from session history, registered adapters, and the current user message.

The pipeline loads session messages, applies compaction filtering, token-budget trimming, runs registered adapters, and assembles the final request with optional tool specs.

Errors

Returns RuntimeError on adapter failure or storage errors.

pub async fn build_context( &self, store: &RuntimeStore, session_id: Uuid, user_message: Option<&str>, ) -> RuntimeResult<ContextOutput>

Builds context output without creating a chat request.

Equivalent to build but returns the raw ContextOutput instead of wrapping it in a ChatRequest. Useful when the caller needs access to the composed message list without binding to a specific model or tools.

Errors

Returns RuntimeError on adapter failure or storage errors.

Trait Implementations

impl Default for ContextPipeline

fn default() -> Self

Returns the “default value” for a type.