Documentation
¶
Overview ¶
Package assembler builds context strings from retrieved chunks for use in LLM prompts, with token budgeting and citation tracking.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func AssembleSimple ¶
AssembleSimple creates a simple context string by joining chunk contents.
Types ¶
type AssembleResult ¶
type AssembleResult struct {
// Context is the assembled context string for the LLM prompt.
Context string `json:"context"`
// Citations tracks which chunks were included.
Citations []Citation `json:"citations,omitempty"`
// TotalTokens is the estimated total token count.
TotalTokens int `json:"total_tokens"`
// TruncatedCount is the number of chunks that were dropped due to budget.
TruncatedCount int `json:"truncated_count"`
}
AssembleResult contains the assembled context and metadata.
type Assembler ¶
type Assembler struct {
// contains filtered or unexported fields
}
Assembler builds context strings from retrieved chunks.
type BudgetManager ¶
type BudgetManager struct {
// contains filtered or unexported fields
}
BudgetManager tracks token consumption against a budget.
func NewBudgetManager ¶
func NewBudgetManager(counter TokenCounter, maxTokens int) *BudgetManager
NewBudgetManager creates a new budget manager.
func (*BudgetManager) CanFit ¶
func (b *BudgetManager) CanFit(tokens int) bool
CanFit returns true if the given number of tokens fits in the remaining budget.
func (*BudgetManager) Consume ¶
func (b *BudgetManager) Consume(tokens int)
Consume adds tokens to the used count.
func (*BudgetManager) EstimateTokens ¶
func (b *BudgetManager) EstimateTokens(text string) int
EstimateTokens returns the token count for the given text.
func (*BudgetManager) Remaining ¶
func (b *BudgetManager) Remaining() int
Remaining returns the tokens still available.
func (*BudgetManager) Used ¶
func (b *BudgetManager) Used() int
Used returns the total tokens consumed.
type Citation ¶
type Citation struct {
// ChunkIndex is the position in the original retrieval results.
ChunkIndex int `json:"chunk_index"`
// Content is the chunk text.
Content string `json:"content"`
// Score is the retrieval relevance score.
Score float64 `json:"score"`
// Metadata from the source chunk.
Metadata map[string]string `json:"metadata,omitempty"`
}
Citation tracks the provenance of a chunk included in the assembled context.
type CitationTracker ¶
type CitationTracker struct {
// contains filtered or unexported fields
}
CitationTracker collects citations as chunks are included.
func (*CitationTracker) Citations ¶
func (t *CitationTracker) Citations() []Citation
Citations returns all recorded citations.
func (*CitationTracker) Count ¶
func (t *CitationTracker) Count() int
Count returns the number of citations recorded.
type Option ¶
type Option func(*Assembler)
Option configures the Assembler.
func WithMaxTokens ¶
WithMaxTokens sets the maximum token budget for assembled context.
func WithTemplate ¶
WithTemplate sets a custom context template.
func WithTokenCounter ¶
func WithTokenCounter(tc TokenCounter) Option
WithTokenCounter sets a custom token counter.
type SimpleTokenCounter ¶
type SimpleTokenCounter struct{}
SimpleTokenCounter estimates tokens by dividing character count by 4.
func (*SimpleTokenCounter) CountTokens ¶
func (c *SimpleTokenCounter) CountTokens(text string) int
CountTokens returns an approximate token count (~4 chars per token).
type Template ¶
type Template struct {
// Header is prepended to the assembled context.
Header string
Footer string
// Separator is placed between chunks.
Separator string
// ChunkFormat formats each chunk. Use %d for index and %s for content.
ChunkFormat string
}
Template defines how retrieved chunks are formatted into context.
func DefaultTemplate ¶
func DefaultTemplate() *Template
DefaultTemplate returns a sensible default template.
type TokenCounter ¶
type TokenCounter interface {
// CountTokens returns the estimated number of tokens in the text.
CountTokens(text string) int
}
TokenCounter estimates token counts for text.
Source Files