diff --git a/internal/agent/prompts.go b/internal/agent/prompts.go index fd1eaabd..06960a69 100644 --- a/internal/agent/prompts.go +++ b/internal/agent/prompts.go @@ -105,79 +105,74 @@ func AvailablePlaceholders() []PlaceholderDefinition { return result } -// formatKnowledgeBaseList formats knowledge base information for the prompt +// formatKnowledgeBaseList formats knowledge base information as XML for the prompt func formatKnowledgeBaseList(kbInfos []*KnowledgeBaseInfo) string { if len(kbInfos) == 0 { - return "None" + return "" } - var builder strings.Builder - builder.WriteString("\nThe following knowledge bases have been selected by the user for this conversation. ") - builder.WriteString("You should search within these knowledge bases to find relevant information.\n\n") - for i, kb := range kbInfos { - // Display knowledge base name and ID - builder.WriteString(fmt.Sprintf("%d. **%s** (knowledge_base_id: `%s`)\n", i+1, kb.Name, kb.ID)) - - // Display knowledge base type + var b strings.Builder + b.WriteString("\n") + for _, kb := range kbInfos { kbType := kb.Type if kbType == "" { - kbType = "document" // Default type + kbType = "document" } - builder.WriteString(fmt.Sprintf(" - Type: %s\n", kbType)) - + b.WriteString(fmt.Sprintf("\n", + kb.ID, kb.Name, kbType, kb.DocCount)) if kb.Description != "" { - builder.WriteString(fmt.Sprintf(" - Description: %s\n", kb.Description)) + b.WriteString(fmt.Sprintf("%s\n", kb.Description)) } - builder.WriteString(fmt.Sprintf(" - Document count: %d\n", kb.DocCount)) - // Display recent documents if available - // For FAQ type knowledge bases, adjust the display format if len(kb.RecentDocs) > 0 { if kbType == "faq" { - // FAQ knowledge base: show Q&A pairs in a more compact format - builder.WriteString(" - Recent FAQ entries:\n\n") - builder.WriteString(" | # | Question | Answers | Chunk ID | Knowledge ID | Created At |\n") - builder.WriteString(" |---|-------------------|---------|----------|--------------|------------|\n") + b.WriteString("\n") for j, doc := range kb.RecentDocs { - if j >= 10 { // Limit to 10 documents + if j >= 10 { break } question := doc.FAQStandardQuestion if question == "" { question = doc.FileName } - answers := "-" + b.WriteString(fmt.Sprintf("\n", + doc.ChunkID, doc.KnowledgeID, doc.CreatedAt)) + b.WriteString(fmt.Sprintf("%s\n", question)) if len(doc.FAQAnswers) > 0 { - answers = strings.Join(doc.FAQAnswers, " | ") + for _, ans := range doc.FAQAnswers { + b.WriteString(fmt.Sprintf("%s\n", ans)) + } } - builder.WriteString(fmt.Sprintf(" | %d | %s | %s | `%s` | `%s` | %s |\n", - j+1, question, answers, doc.ChunkID, doc.KnowledgeID, doc.CreatedAt)) + b.WriteString("\n") } + b.WriteString("\n") } else { - // Document knowledge base: show documents in standard format - builder.WriteString(" - Recently added documents:\n\n") - builder.WriteString(" | # | Document Name | Type | Created At | Knowledge ID | File Size | Summary |\n") - builder.WriteString(" |---|---------------|------|------------|--------------|----------|---------|\n") + b.WriteString("\n") for j, doc := range kb.RecentDocs { - if j >= 10 { // Limit to 10 documents + if j >= 10 { break } docName := doc.Title if docName == "" { docName = doc.FileName } - // Format file size fileSize := formatFileSize(doc.FileSize) - summary := formatDocSummary(doc.Description, 120) - builder.WriteString(fmt.Sprintf(" | %d | %s | %s | %s | `%s` | %s | %s |\n", - j+1, docName, doc.Type, doc.CreatedAt, doc.KnowledgeID, fileSize, summary)) + b.WriteString(fmt.Sprintf("\n", + doc.KnowledgeID, doc.Type, fileSize, doc.CreatedAt)) + b.WriteString(fmt.Sprintf("%s\n", docName)) + if doc.Description != "" { + summary := formatDocSummary(doc.Description, 120) + b.WriteString(fmt.Sprintf("%s\n", summary)) + } + b.WriteString("\n") } + b.WriteString("\n") } - builder.WriteString("\n") } - builder.WriteString("\n") + b.WriteString("\n") } - return builder.String() + b.WriteString("") + return b.String() } // renderPromptPlaceholders renders placeholders in the prompt template diff --git a/internal/agent/tools/grep_chunks.go b/internal/agent/tools/grep_chunks.go index 8fd2fda2..4637d7f5 100644 --- a/internal/agent/tools/grep_chunks.go +++ b/internal/agent/tools/grep_chunks.go @@ -374,7 +374,7 @@ func (t *GrepChunksTool) searchChunks( return results, int64(len(results)), nil } -// formatOutput formats the search results for display (grep-style output) +// formatOutput formats the search results as XML func (t *GrepChunksTool) formatOutput( ctx context.Context, results []knowledgeAggregation, @@ -382,46 +382,37 @@ func (t *GrepChunksTool) formatOutput( patterns []string, countOnly bool, ) string { - var output strings.Builder + var b strings.Builder - // If count_only mode, just return the count if countOnly { - output.WriteString(fmt.Sprintf("%d\n", totalCount)) - return output.String() + b.WriteString(fmt.Sprintf("\n", totalCount)) + return b.String() } - // Show search info - if len(patterns) == 1 { - output.WriteString(fmt.Sprintf("Pattern: '%s' (case-insensitive)\n", patterns[0])) - } else { - output.WriteString(fmt.Sprintf("Patterns (%d): %v (case-insensitive, OR logic)\n", len(patterns), patterns)) + b.WriteString(fmt.Sprintf("\n", len(results))) + for _, p := range patterns { + b.WriteString(fmt.Sprintf("%s\n", p)) } - output.WriteString(fmt.Sprintf("Matches: %d knowledge item(s)\n\n", len(results))) if len(results) == 0 { - output.WriteString("No matches found.\n") - return output.String() + b.WriteString("") + return b.String() } - for idx, result := range results { - var patternSummaries []string + for _, result := range results { + b.WriteString(fmt.Sprintf("\n", + result.KnowledgeID, result.KnowledgeTitle, result.ChunkHitCount, result.TotalChunkCount)) for _, pattern := range patterns { count := result.PatternCounts[pattern] - patternSummaries = append(patternSummaries, fmt.Sprintf("%s=%d", pattern, count)) + if count > 0 { + b.WriteString(fmt.Sprintf("\n", pattern, count)) + } } - - output.WriteString( - fmt.Sprintf("%d) knowledge_id=%s | title=%s | chunk_hits=%d | chunk_total=%d | pattern_hits=[%s]\n", - idx+1, - result.KnowledgeID, - result.KnowledgeTitle, - result.ChunkHitCount, - result.TotalChunkCount, - strings.Join(patternSummaries, ", "), - ), - ) + b.WriteString("\n") } - return output.String() + + b.WriteString("") + return b.String() } type knowledgeAggregation struct { diff --git a/internal/agent/tools/knowledge_search.go b/internal/agent/tools/knowledge_search.go index 22c657c7..412d9490 100644 --- a/internal/agent/tools/knowledge_search.go +++ b/internal/agent/tools/knowledge_search.go @@ -395,6 +395,21 @@ func (t *KnowledgeSearchTool) Execute(ctx context.Context, args json.RawMessage) } } + // Enrich image info for search results (lazy-loaded from child image chunks) + if t.chunkService != nil && len(deduplicatedResults) > 0 { + byTenant := make(map[uint64][]*types.SearchResult) + for _, r := range deduplicatedResults { + tid := t.searchTargets.GetTenantIDForKB(r.KnowledgeBaseID) + if tid == 0 { + continue + } + byTenant[tid] = append(byTenant[tid], r.SearchResult) + } + for tid, batch := range byTenant { + searchutil.EnrichSearchResultsImageInfo(ctx, t.chunkService.GetRepository(), tid, batch) + } + } + // Build output logger.Infof(ctx, "[Tool][KnowledgeSearch] Formatting output with %d final results", len(deduplicatedResults)) result, err := t.formatOutput(ctx, deduplicatedResults, kbIDs, queries) @@ -1065,144 +1080,106 @@ func (t *KnowledgeSearchTool) formatOutput( }, nil } - // Build output header - output := "=== Search Results ===\n" - output += fmt.Sprintf("Found %d relevant results", len(results)) - output += "\n\n" - // Count results by KB kbCounts := make(map[string]int) for _, r := range results { kbCounts[r.KnowledgeID]++ } - output += "Knowledge Base Coverage:\n" - for kbID, count := range kbCounts { - output += fmt.Sprintf(" - %s: %d results\n", kbID, count) - } - output += "\n=== Detailed Results ===\n\n" + // Format individual results as XML + var ob strings.Builder + ob.WriteString(fmt.Sprintf("\n", len(results))) - // Format individual results formattedResults := make([]map[string]interface{}, 0, len(results)) - currentKB := "" faqMetadataCache := make(map[string]*types.FAQChunkMetadata) - // Track chunks per knowledge for statistics - knowledgeChunkMap := make(map[string]map[int]bool) // knowledge_id -> set of chunk_index - knowledgeTotalMap := make(map[string]int64) // knowledge_id -> total chunks - knowledgeTitleMap := make(map[string]string) // knowledge_id -> title + knowledgeChunkMap := make(map[string]map[int]bool) + knowledgeTotalMap := make(map[string]int64) + knowledgeTitleMap := make(map[string]string) for i, result := range results { var faqMeta *types.FAQChunkMetadata if result.KnowledgeBaseType == types.KnowledgeBaseTypeFAQ { meta, err := t.getFAQMetadata(ctx, result.ID, faqMetadataCache) if err != nil { - logger.Warnf( - ctx, - "[Tool][KnowledgeSearch] Failed to load FAQ metadata for chunk %s: %v", - result.ID, - err, - ) + logger.Warnf(ctx, "[Tool][KnowledgeSearch] Failed to load FAQ metadata for chunk %s: %v", result.ID, err) } else { faqMeta = meta } } - // Track chunk indices per knowledge if knowledgeChunkMap[result.KnowledgeID] == nil { knowledgeChunkMap[result.KnowledgeID] = make(map[int]bool) } knowledgeChunkMap[result.KnowledgeID][result.ChunkIndex] = true knowledgeTitleMap[result.KnowledgeID] = result.KnowledgeTitle - // Group by knowledge base - if result.KnowledgeID != currentKB { - currentKB = result.KnowledgeID - if i > 0 { - output += "\n" - } - output += fmt.Sprintf("[Source Document: %s]\n", result.KnowledgeTitle) - - // Get total chunk count for this knowledge (cache it) - // Use KB's tenant_id from searchTargets to support cross-tenant shared KB - if _, exists := knowledgeTotalMap[result.KnowledgeID]; !exists { - // Get tenant_id from searchTargets using the KB ID from the result - effectiveTenantID := t.searchTargets.GetTenantIDForKB(result.KnowledgeBaseID) - if effectiveTenantID == 0 { - logger.Warnf(ctx, "[Tool][KnowledgeSearch] KB %s not found in searchTargets, skipping chunk count", result.KnowledgeBaseID) + // Cache total chunk count per knowledge + if _, exists := knowledgeTotalMap[result.KnowledgeID]; !exists { + effectiveTenantID := t.searchTargets.GetTenantIDForKB(result.KnowledgeBaseID) + if effectiveTenantID == 0 { + logger.Warnf(ctx, "[Tool][KnowledgeSearch] KB %s not found in searchTargets, skipping chunk count", result.KnowledgeBaseID) + knowledgeTotalMap[result.KnowledgeID] = 0 + } else { + _, total, err := t.chunkService.GetRepository().ListPagedChunksByKnowledgeID(ctx, + effectiveTenantID, result.KnowledgeID, + &types.Pagination{Page: 1, PageSize: 1}, + []types.ChunkType{types.ChunkTypeText}, "", "", "", "", "", + ) + if err != nil { + logger.Warnf(ctx, "[Tool][KnowledgeSearch] Failed to get total chunks for knowledge %s: %v", result.KnowledgeID, err) knowledgeTotalMap[result.KnowledgeID] = 0 } else { - _, total, err := t.chunkService.GetRepository().ListPagedChunksByKnowledgeID(ctx, - effectiveTenantID, result.KnowledgeID, - &types.Pagination{Page: 1, PageSize: 1}, - []types.ChunkType{types.ChunkTypeText}, "", "", "", "", "", - ) - if err != nil { - logger.Warnf( - ctx, - "[Tool][KnowledgeSearch] Failed to get total chunks for knowledge %s: %v", - result.KnowledgeID, - err, - ) - knowledgeTotalMap[result.KnowledgeID] = 0 - } else { - knowledgeTotalMap[result.KnowledgeID] = total - } + knowledgeTotalMap[result.KnowledgeID] = total } } } - // relevanceLevel := GetRelevanceLevel(result.Score) - output += fmt.Sprintf("\nResult #%d:\n", i+1) - output += fmt.Sprintf( - " [chunk_id: %s][chunk_index: %d]\nContent: %s\n", - result.ID, - result.ChunkIndex, - result.Content, - ) + ob.WriteString(fmt.Sprintf("\n", + i+1, result.ID, result.ChunkIndex, result.KnowledgeID, result.KnowledgeTitle)) + ob.WriteString(fmt.Sprintf("%s\n", result.Content)) - // 解析并输出关联的图片信息 if result.ImageInfo != "" { var imageInfos []types.ImageInfo if err := json.Unmarshal([]byte(result.ImageInfo), &imageInfos); err == nil && len(imageInfos) > 0 { - output += fmt.Sprintf(" Related Images (%d):\n", len(imageInfos)) - for imgIdx, img := range imageInfos { - output += fmt.Sprintf(" Image %d:\n", imgIdx+1) - if img.URL != "" { - output += fmt.Sprintf(" URL: %s\n", img.URL) - } + for _, img := range imageInfos { + ob.WriteString(fmt.Sprintf("\n", img.URL)) if img.Caption != "" { - output += fmt.Sprintf(" Caption: %s\n", img.Caption) + ob.WriteString(fmt.Sprintf("%s\n", img.Caption)) } if img.OCRText != "" { - output += fmt.Sprintf(" OCR Text: %s\n", img.OCRText) + ob.WriteString(fmt.Sprintf("%s\n", img.OCRText)) } + ob.WriteString("\n") } } } if faqMeta != nil { + ob.WriteString("\n") if faqMeta.StandardQuestion != "" { - output += fmt.Sprintf(" FAQ Standard Question: %s\n", faqMeta.StandardQuestion) + ob.WriteString(fmt.Sprintf("%s\n", faqMeta.StandardQuestion)) } if len(faqMeta.SimilarQuestions) > 0 { - output += fmt.Sprintf(" FAQ Similar Questions: %s\n", strings.Join(faqMeta.SimilarQuestions, "; ")) - } - if len(faqMeta.Answers) > 0 { - output += " FAQ Answers:\n" - for ansIdx, ans := range faqMeta.Answers { - output += fmt.Sprintf(" Answer Choice %d: %s\n", ansIdx+1, ans) + for _, sq := range faqMeta.SimilarQuestions { + ob.WriteString(fmt.Sprintf("%s\n", sq)) } } + if len(faqMeta.Answers) > 0 { + for _, ans := range faqMeta.Answers { + ob.WriteString(fmt.Sprintf("%s\n", ans)) + } + } + ob.WriteString("\n") } + ob.WriteString("\n") + formattedResults = append(formattedResults, map[string]interface{}{ - "result_index": i + 1, - "chunk_id": result.ID, - "content": result.Content, - // "score": result.Score, - // "relevance_level": relevanceLevel, + "result_index": i + 1, + "chunk_id": result.ID, + "content": result.Content, "knowledge_id": result.KnowledgeID, "knowledge_title": result.KnowledgeTitle, "match_type": result.MatchType, @@ -1213,11 +1190,9 @@ func (t *KnowledgeSearchTool) formatOutput( last := formattedResults[len(formattedResults)-1] - // 添加图片信息到结构化数据 if result.ImageInfo != "" { var imageInfos []types.ImageInfo if err := json.Unmarshal([]byte(result.ImageInfo), &imageInfos); err == nil && len(imageInfos) > 0 { - // 构建简化的图片信息列表 imageList := make([]map[string]string, 0, len(imageInfos)) for _, img := range imageInfos { imgData := make(map[string]string) @@ -1253,36 +1228,23 @@ func (t *KnowledgeSearchTool) formatOutput( } } - // Add statistics and recommendations for each knowledge - output += "\n=== Retrieval Statistics ===\n\n" + // Retrieval statistics + ob.WriteString("\n") for knowledgeID, retrievedChunks := range knowledgeChunkMap { totalChunks := knowledgeTotalMap[knowledgeID] retrievedCount := len(retrievedChunks) title := knowledgeTitleMap[knowledgeID] - if totalChunks > 0 { - percentage := float64(retrievedCount) / float64(totalChunks) * 100 remaining := totalChunks - int64(retrievedCount) - - output += fmt.Sprintf("Document: %s (%s)\n", title, knowledgeID) - output += fmt.Sprintf(" Total Chunks: %d\n", totalChunks) - output += fmt.Sprintf(" Retrieved: %d (%.1f%%)\n", retrievedCount, percentage) - output += fmt.Sprintf(" Remaining: %d\n", remaining) - + percentage := float64(retrievedCount) / float64(totalChunks) * 100 + ob.WriteString(fmt.Sprintf("\n", + knowledgeID, title, totalChunks, retrievedCount, remaining, percentage)) } } + ob.WriteString("\n") + ob.WriteString("") - // // Add usage guidance - // output += "\n\n=== Usage Guidelines ===\n" - // output += "- High relevance (>=0.8): directly usable for answering\n" - // output += "- Medium relevance (0.6-0.8): use as supplementary reference\n" - // output += "- Low relevance (<0.6): use with caution, may not be accurate\n" - // if totalBeforeFilter > len(results) { - // output += "- Results below threshold have been automatically filtered\n" - // } - // output += "- Full content is already included in search results above\n" - // output += "- Results are deduplicated across knowledge bases and sorted by relevance\n" - // output += "- Use list_knowledge_chunks to expand context if needed\n" + output := ob.String() data := map[string]interface{}{ "knowledge_base_ids": kbsToSearch, diff --git a/internal/agent/tools/list_knowledge_chunks.go b/internal/agent/tools/list_knowledge_chunks.go index 02e95ad5..13633ea2 100644 --- a/internal/agent/tools/list_knowledge_chunks.go +++ b/internal/agent/tools/list_knowledge_chunks.go @@ -6,6 +6,7 @@ import ( "fmt" "strings" + "github.com/Tencent/WeKnora/internal/searchutil" "github.com/Tencent/WeKnora/internal/types" "github.com/Tencent/WeKnora/internal/types/interfaces" ) @@ -160,6 +161,22 @@ func (t *ListKnowledgeChunksTool) Execute(ctx context.Context, args json.RawMess totalChunks := total fetched := len(chunks) + // Enrich image info from child image chunks (lazy loading) + if fetched > 0 { + chunkIDs := make([]string, 0, fetched) + for _, c := range chunks { + chunkIDs = append(chunkIDs, c.ID) + } + infoMap := searchutil.CollectImageInfoByChunkIDs(ctx, t.chunkService.GetRepository(), effectiveTenantID, chunkIDs) + for _, c := range chunks { + if c.ImageInfo == "" { + if merged, ok := infoMap[c.ID]; ok { + c.ImageInfo = merged + } + } + } + } + knowledgeTitle := t.lookupKnowledgeTitle(ctx, knowledgeID) output := t.buildOutput(knowledgeID, knowledgeTitle, totalChunks, fetched, chunks) @@ -236,7 +253,7 @@ func (t *ListKnowledgeChunksTool) lookupKnowledgeTitle(ctx context.Context, know return strings.TrimSpace(knowledge.Title) } -// buildOutput builds the output for the list knowledge chunks tool +// buildOutput builds the output as XML for the list knowledge chunks tool func (t *ListKnowledgeChunksTool) buildOutput( knowledgeID string, knowledgeTitle string, @@ -244,65 +261,54 @@ func (t *ListKnowledgeChunksTool) buildOutput( fetched int, chunks []*types.Chunk, ) string { - builder := &strings.Builder{} - builder.WriteString("=== Knowledge Document Chunks ===\n\n") + var b strings.Builder + titleAttr := "" if knowledgeTitle != "" { - fmt.Fprintf(builder, "Document: %s (%s)\n", knowledgeTitle, knowledgeID) - } else { - fmt.Fprintf(builder, "Document ID: %s\n", knowledgeID) + titleAttr = fmt.Sprintf(" title=\"%s\"", knowledgeTitle) } - fmt.Fprintf(builder, "Total chunks: %d\n", total) + fmt.Fprintf(&b, "\n", + knowledgeID, titleAttr, total, fetched) if fetched == 0 { - builder.WriteString("No chunks found. Please confirm the document has been parsed.\n") - if total > 0 { - builder.WriteString("Document exists but the current page is empty. Please check pagination parameters.\n") - } - return builder.String() + b.WriteString("") + return b.String() } - fmt.Fprintf( - builder, - "Fetched: %d chunks, range: %d - %d\n\n", - fetched, - chunks[0].ChunkIndex, - chunks[len(chunks)-1].ChunkIndex, - ) - builder.WriteString("=== Chunk Content Preview ===\n\n") - for idx, c := range chunks { - fmt.Fprintf(builder, "Chunk #%d (Index %d)\n", idx+1, c.ChunkIndex+1) - fmt.Fprintf(builder, " chunk_id: %s\n", c.ID) - fmt.Fprintf(builder, " Type: %s\n", c.ChunkType) - fmt.Fprintf(builder, " Content: %s\n", summarizeContent(c.Content)) + for _, c := range chunks { + fmt.Fprintf(&b, "\n", + c.ID, c.ChunkIndex, c.ChunkType) + fmt.Fprintf(&b, "%s\n", summarizeContent(c.Content)) - // Output associated image information if c.ImageInfo != "" { var imageInfos []types.ImageInfo if err := json.Unmarshal([]byte(c.ImageInfo), &imageInfos); err == nil && len(imageInfos) > 0 { - fmt.Fprintf(builder, " Associated images (%d):\n", len(imageInfos)) - for imgIdx, img := range imageInfos { - fmt.Fprintf(builder, " Image %d:\n", imgIdx+1) + for _, img := range imageInfos { if img.URL != "" { - fmt.Fprintf(builder, " URL: %s\n", img.URL) + fmt.Fprintf(&b, "\n", img.URL) + } else { + b.WriteString("\n") } if img.Caption != "" { - fmt.Fprintf(builder, " Caption: %s\n", img.Caption) + fmt.Fprintf(&b, "%s\n", img.Caption) } if img.OCRText != "" { - fmt.Fprintf(builder, " OCR Text: %s\n", img.OCRText) + fmt.Fprintf(&b, "%s\n", img.OCRText) } + b.WriteString("\n") } } } - builder.WriteString("\n") + + b.WriteString("\n") } if int64(fetched) < total { - builder.WriteString("Note: The document has more chunks. Adjust offset or make multiple calls to retrieve all content.\n") + fmt.Fprintf(&b, "\n", int64(total)-int64(fetched)) } - return builder.String() + b.WriteString("") + return b.String() } // summarizeContent summarizes the content of a chunk