Changeset 0.18.0 (#79)

server/handlers/memory_inject.go

@@ -0,0 +1,100 @@
+package handlers
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"strings"
+
+	"git.gobha.me/xcaliber/chat-switchboard/knowledge"
+	"git.gobha.me/xcaliber/chat-switchboard/models"
+	"git.gobha.me/xcaliber/chat-switchboard/store"
+)
+
+// maxMemoryChars is the approximate character budget for injected memories.
+const maxMemoryChars = 6000
+
+// BuildMemoryHint loads active memories for the user (and persona if active)
+// and formats them as a system prompt injection.
+//
+// Phase 2: if an embedder is available and the user's latest message
+// is provided, it generates a query vector for semantic recall. Otherwise
+// falls back to keyword/confidence ranking.
+func BuildMemoryHint(ctx context.Context, stores store.Stores, embedder *knowledge.Embedder, userID, personaID, lastUserMessage string) string {
+	if stores.Memories == nil {
+		return ""
+	}
+
+	var pID *string
+	if personaID != "" {
+		pID = &personaID
+	}
+
+	var memories []models.Memory
+	var err error
+
+	// Try hybrid recall with embedding if available
+	if embedder != nil && embedder.IsConfigured(ctx) && lastUserMessage != "" {
+		memories, err = hybridRecallForInjection(ctx, stores, embedder, userID, pID, lastUserMessage)
+	}
+
+	// Fall back to keyword recall
+	if err != nil || len(memories) == 0 {
+		memories, err = stores.Memories.Recall(ctx, userID, pID, "", 30)
+	}
+
+	if err != nil {
+		log.Printf("⚠ memory injection failed: %v", err)
+		return ""
+	}
+	if len(memories) == 0 {
+		return ""
+	}
+
+	// Format memories as bullet points, respecting token budget
+	var b strings.Builder
+	b.WriteString("Known facts about this user (from previous conversations):\n")
+
+	totalChars := b.Len()
+	included := 0
+	for _, m := range memories {
+		line := fmt.Sprintf("• %s: %s\n", m.Key, m.Value)
+		if totalChars+len(line) > maxMemoryChars {
+			break
+		}
+		b.WriteString(line)
+		totalChars += len(line)
+		included++
+	}
+
+	if included == 0 {
+		return ""
+	}
+
+	log.Printf("🧠 Injected %d memories for user %s (persona: %s)", included, userID, personaID)
+	return b.String()
+}
+
+// hybridRecallForInjection generates a query vector from the user's message
+// and performs hybrid semantic + keyword recall.
+func hybridRecallForInjection(ctx context.Context, stores store.Stores, embedder *knowledge.Embedder, userID string, personaID *string, message string) ([]models.Memory, error) {
+	text := message
+	if len(text) > 2000 {
+		text = text[:2000]
+	}
+
+	var teamID *string
+	if stores.Teams != nil {
+		ids, _ := stores.Teams.GetUserTeamIDs(ctx, userID)
+		if len(ids) > 0 {
+			teamID = &ids[0]
+		}
+	}
+
+	result, err := embedder.EmbedChunks(ctx, userID, teamID, []string{text})
+	if err != nil || len(result.Vectors) == 0 {
+		return nil, err
+	}
+
+	return stores.Memories.RecallHybrid(ctx, userID, personaID, "", result.Vectors[0], 30)
+}