Module context 

Context assembly — token-budgeted context building for LLM prompts.

The context assembler retrieves facts via hybrid search, ranks them by tier priority, fills a token budget greedily, and formats the output for injection into system prompts, messages, or raw JSON.

Structs

CharTokenEstimator

Character-based token estimator (~4 chars per token). Accurate to within ~10% for English text across GPT/Claude models.

ContextBlock

The assembled context block returned by ContextBuilder::build().

ContextBuilder

Token-budgeted context assembler.

ContextConfig

Configuration for context assembly.

Enums

OutputFormat

Format for the assembled context block.

Traits

TokenEstimator

Pluggable token estimation. Implementations convert text to an approximate token count without requiring a full tokenizer dependency.

Functions

format_markdown

Format facts as human-readable Markdown.

format_raw

Format facts as raw JSON (array of objects with text, tier, category, confidence).

format_system_prompt

Format facts as an XML-tagged block for system prompt injection.

sort_by_tier_priority

Sort facts by tier priority (Working first, then Conversation, then Knowledge). Within the same tier, preserve the original order (which is by retrieval score).