Enha (RAG): raw text as primary key in vector db

9 files changed, 21 insertions, 23 deletions

diff --git a/README.md b/README.md
index f158bcb..31bfab9 100644
--- a/README.md
+++ b/README.md
@@ -33,6 +33,7 @@
 - lets say we have two (or more) agents with the same name across multiple chats. These agents go and ask db for topics they memorised. Now they can access topics that aren't meant for them. (so memory should have an option: shareable; that indicates if that memory can be shared across chats);
 - delete chat option;
 - server mode: no tui but api calls with the func calling, rag, other middleware;
+- boolean flag to use/not use tools. I see it as a msg from a tool to an llm "Hey, it might be good idea to use me!";
 
 ### FIX:
 - bot responding (or hanging) blocks everything; +
diff --git a/config/config.go b/config/config.go
index 3c79564..eb5f3f1 100644
--- a/config/config.go
+++ b/config/config.go
@@ -12,6 +12,7 @@ type Config struct {
 	LogFile       string `toml:"LogFile"`
 	UserRole      string `toml:"UserRole"`
 	ToolRole      string `toml:"ToolRole"`
+	ToolUse       bool   `toml:"ToolUse"`
 	AssistantRole string `toml:"AssistantRole"`
 	AssistantIcon string `toml:"AssistantIcon"`
 	UserIcon      string `toml:"UserIcon"`
diff --git a/main.go b/main.go
index 8f05f5b..a8ae716 100644
--- a/main.go
+++ b/main.go
@@ -10,7 +10,7 @@ var (
 	botRespMode   = false
 	editMode      = false
 	selectedIndex = int(-1)
-	indexLine     = "F12 to show keys help; bot resp mode: %v; char: %s; chat: %s; RAGEnabled: %v; EmbedURL: %s"
+	indexLine     = "F12 to show keys help; bot resp mode: %v; char: %s; chat: %s; RAGEnabled: %v; toolUseAdviced: %v"
 	focusSwitcher = map[tview.Primitive]tview.Primitive{}
 )
 
diff --git a/models/db.go b/models/db.go
index db4ae75..ce1309a 100644
--- a/models/db.go
+++ b/models/db.go
@@ -39,7 +39,6 @@ type Memory struct {
 // vector models
 
 type VectorRow struct {
-	ID         uint32    `db:"id" json:"id"`
 	Embeddings []float32 `db:"embeddings" json:"embeddings"`
 	Slug       string    `db:"slug" json:"slug"`
 	RawText    string    `db:"raw_text" json:"raw_text"`
diff --git a/rag/main.go b/rag/main.go
index a7084bf..da919b4 100644
--- a/rag/main.go
+++ b/rag/main.go
@@ -88,7 +88,9 @@ func (r *RAG) writeVectors(vectorCh <-chan []models.VectorRow, doneCh <-chan boo
 		case batch := <-vectorCh:
 			for _, vector := range batch {
 				if err := r.store.WriteVector(&vector); err != nil {
-					return err
+					r.logger.Error("failed to write vector", "error", err, "slug", vector.Slug)
+					continue // a duplicate is not critical
+					// return err
 				}
 			}
 			r.logger.Info("wrote batch to db", "size", len(batch))
@@ -226,14 +228,14 @@ func (r *RAG) LineToVector(line string) ([]float32, error) {
 	return emb[0], nil
 }
 
-func (r *RAG) saveLine(topic, line string, emb *models.EmbeddingResp) error {
-	row := &models.VectorRow{
-		Embeddings: emb.Embedding,
-		Slug:       topic,
-		RawText:    line,
-	}
-	return r.store.WriteVector(row)
-}
+// func (r *RAG) saveLine(topic, line string, emb *models.EmbeddingResp) error {
+// 	row := &models.VectorRow{
+// 		Embeddings: emb.Embedding,
+// 		Slug:       topic,
+// 		RawText:    line,
+// 	}
+// 	return r.store.WriteVector(row)
+// }
 
 func (r *RAG) SearchEmb(emb *models.EmbeddingResp) ([]models.VectorRow, error) {
 	return r.store.SearchClosest(emb.Embedding)
diff --git a/storage/migrations/002_add_vector.up.sql b/storage/migrations/002_add_vector.up.sql
index f64aecb..6c8fb52 100644
--- a/storage/migrations/002_add_vector.up.sql
+++ b/storage/migrations/002_add_vector.up.sql
@@ -1,13 +1,11 @@
 CREATE VIRTUAL TABLE IF NOT EXISTS embeddings USING vec0(
-    id INTEGER PRIMARY KEY AUTOINCREMENT,
     embedding FLOAT[5120],
     slug TEXT NOT NULL,
-    raw_text TEXT NOT NULL
+    raw_text TEXT PRIMARY KEY,
 );
 
 CREATE VIRTUAL TABLE IF NOT EXISTS embeddings_384 USING vec0(
-    id INTEGER PRIMARY KEY AUTOINCREMENT,
     embedding FLOAT[384],
     slug TEXT NOT NULL,
-    raw_text TEXT NOT NULL
+    raw_text TEXT PRIMARY KEY
 );
diff --git a/storage/vector.go b/storage/vector.go
index 1579686..fe479d8 100644
--- a/storage/vector.go
+++ b/storage/vector.go
@@ -68,7 +68,6 @@ func (p ProviderSQL) WriteVector(row *models.VectorRow) error {
 	}
 	err = stmt.Exec()
 	if err != nil {
-		p.logger.Error("failed exec a stmt", "error", err)
 		return err
 	}
 	return nil
@@ -85,7 +84,6 @@ func (p ProviderSQL) SearchClosest(q []float32) ([]models.VectorRow, error) {
 	}
 	stmt, _, err := p.s3Conn.Prepare(
 		fmt.Sprintf(`SELECT
-			id,
 			distance,
 			embedding,
 			slug,
@@ -109,12 +107,11 @@ func (p ProviderSQL) SearchClosest(q []float32) ([]models.VectorRow, error) {
 	resp := []models.VectorRow{}
 	for stmt.Step() {
 		res := models.VectorRow{}
-		res.ID = uint32(stmt.ColumnInt64(0))
-		res.Distance = float32(stmt.ColumnFloat(1))
-		emb := stmt.ColumnRawText(2)
+		res.Distance = float32(stmt.ColumnFloat(0))
+		emb := stmt.ColumnRawText(1)
 		res.Embeddings = decodeUnsafe(emb)
-		res.Slug = stmt.ColumnText(3)
-		res.RawText = stmt.ColumnText(4)
+		res.Slug = stmt.ColumnText(2)
+		res.RawText = stmt.ColumnText(3)
 		resp = append(resp, res)
 	}
 	if err := stmt.Err(); err != nil {
diff --git a/tui.go b/tui.go
index 0f6a673..7f8f5d6 100644
--- a/tui.go
+++ b/tui.go
@@ -230,7 +230,7 @@ func colorText() {
 }
 
 func updateStatusLine() {
-	position.SetText(fmt.Sprintf(indexLine, botRespMode, cfg.AssistantRole, activeChatName, cfg.RAGEnabled, cfg.EmbedURL))
+	position.SetText(fmt.Sprintf(indexLine, botRespMode, cfg.AssistantRole, activeChatName, cfg.RAGEnabled, cfg.ToolUse))
 }
 
 func initSysCards() ([]string, error) {
diff --git a/vec0.so b/vec0.so
new file mode 100755
index 0000000..bd4c3ca
--- /dev/null
+++ b/vec0.so