Struct Context 

pub struct Context { /* private fields */ }

Manages the conversation state sent to an LLM provider.

A Context bundles together three things:

1. Messages – the conversation history as a Vec<ChatMessage>
2. Tools – the tools available to the model as a Vec<ToolDefinition>
3. Configuration – optional ReasoningEffort and prompt cache key

Pass a Context to StreamingModelProvider::stream_response to generate a response.

Construction

use llm::{Context, ChatMessage, ContentBlock, ToolDefinition, types::IsoString};

let context = Context::new(
    vec![
        ChatMessage::System {
            content: "You are a helpful assistant.".into(),
            timestamp: IsoString::now(),
        },
        ChatMessage::User {
            content: vec![ContentBlock::text("What is Rust?")],
            timestamp: IsoString::now(),
        },
    ],
    vec![], // no tools
);

Managing messages

• add_message – Append a single message.
• push_assistant_turn – Append an assistant response together with its tool call results in one step.
• clear_conversation – Remove all non-system messages.

Conversation compaction

Long conversations can be compacted to stay within context limits:

• messages_for_summary – Get non-system messages to feed to a summarizer.
• with_compacted_summary – Create a new Context where conversation messages are replaced by a ChatMessage::Summary. The system prompt, tools, and configuration are preserved.

Token estimation

estimated_token_count provides a rough pre-flight estimate using a ~4 bytes/token heuristic. This is intentionally approximate – use it to detect obvious overflow before calling the provider, not for precise accounting.

Encrypted reasoning

filter_encrypted_reasoning creates a copy where encrypted reasoning content is kept only for messages from a matching model. This is necessary because encrypted reasoning tokens are model-specific and cannot be replayed to a different model.

Implementations

impl Context

pub fn new(messages: Vec<ChatMessage>, tools: Vec<ToolDefinition>) -> Self

pub fn prompt_cache_key(&self) -> Option<&str>

pub fn set_prompt_cache_key(&mut self, key: Option<String>)

pub fn reasoning_effort(&self) -> Option<ReasoningEffort>

pub fn set_reasoning_effort(&mut self, effort: Option<ReasoningEffort>)

pub fn add_message(&mut self, message: ChatMessage)

pub fn set_tools(&mut self, tools: Vec<ToolDefinition>)

pub fn messages(&self) -> &Vec<ChatMessage>

pub fn tools(&self) -> &Vec<ToolDefinition>

pub fn message_count(&self) -> usize

Returns the number of messages in the context

pub fn estimated_token_count(&self) -> u32

Estimate total token count using the ~4 bytes/token heuristic. Includes messages and tool definitions. Used for pre-flight overflow detection.

pub fn push_assistant_turn( &mut self, content: &str, reasoning: AssistantReasoning, completed_tools: Vec<Result<ToolCallResult, ToolCallError>>, )

Build an assistant turn and its tool call results and append them to messages.

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

Return a copy with encrypted reasoning filtered for the given model. Encrypted content is kept only when its source model matches.

pub fn clear_conversation(&mut self)

Clear all non-system messages, retaining only system prompts.

pub fn messages_for_summary(&self) -> Vec<&ChatMessage>

Get all non-system messages for summarization

pub fn with_compacted_summary(&self, summary: &str) -> Context

Create a new context with all messages replaced by a summary. Preserves the system prompt and tools.

Trait Implementations

impl Clone for Context

fn clone(&self) -> Context

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Context

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for Context

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for Context

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.