Add nomic embedding daemon for fast repeated queries

Background daemon holds the nomic model in memory over a Unix socket,
so subsequent queries within 5 minutes skip the ~1s model load. Cache
and runtime files live under project-local .rekal/nomic/ instead of
global ~/.cache/. Model is pre-decompressed during rekal init.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: Frank Guo

cmd/rekal/cli/checkpoint.go

@@ -357,7 +357,7 @@ func updateIndexIncremental(gitRoot string, sessionIDs []string, checkpointID st
 		return err
 	}
 
-	if err := buildNomicEmbeddings(indexDB, sessionContent, w); err != nil {
+	if err := buildNomicEmbeddings(indexDB, sessionContent, w, gitRoot); err != nil {
 		fmt.Fprintf(w, "rekal: warning: nomic embeddings skipped: %v\n", err)
 	}
 

cmd/rekal/cli/index_cmd.go

@@ -116,7 +116,7 @@ func runIndex(cmd *cobra.Command, gitRoot string) error {
 		}
 
 		// Nomic pass (non-fatal).
-		if err := buildNomicEmbeddings(indexDB, sessionContent, w); err != nil {
+		if err := buildNomicEmbeddings(indexDB, sessionContent, w, gitRoot); err != nil {
 			fmt.Fprintf(w, "warning: nomic embeddings skipped: %v\n", err)
 		}
 	}
@@ -141,19 +141,19 @@ func runIndex(cmd *cobra.Command, gitRoot string) error {
 
 // buildNomicEmbeddings generates nomic-embed-text embeddings for all sessions
 // and stores them in the index DB. Non-fatal: returns error on any failure.
-func buildNomicEmbeddings(indexDB *sql.DB, sessionContent map[string]string, w io.Writer) error {
+func buildNomicEmbeddings(indexDB *sql.DB, sessionContent map[string]string, w io.Writer, gitRoot string) error {
 	if !nomic.Supported() {
 		return nil
 	}
 
 	fmt.Fprintln(w, "building nomic deep semantic embeddings...")
-	embedder, err := nomic.NewEmbedder()
+	client, err := nomic.NewClient(gitRoot)
 	if err != nil {
 		return err
 	}
-	defer embedder.Close()
+	defer client.Close()
 
-	vectors, err := embedder.EmbedSessions(sessionContent)
+	vectors, err := client.EmbedSessions(sessionContent)
 	if err != nil {
 		return err
 	}

cmd/rekal/cli/init.go

@@ -10,6 +10,7 @@ import (
 
 	"github.com/rekal-dev/rekal-cli/cmd/rekal/cli/codec"
 	"github.com/rekal-dev/rekal-cli/cmd/rekal/cli/db"
+	"github.com/rekal-dev/rekal-cli/cmd/rekal/cli/nomic"
 	"github.com/rekal-dev/rekal-cli/cmd/rekal/cli/skill"
 	"github.com/spf13/cobra"
 )
@@ -121,6 +122,12 @@ imported into the local data DB automatically.`,
 				fmt.Fprintf(cmd.ErrOrStderr(), "rekal: warning: initial checkpoint failed: %v\n", err)
 			}
 
+			// Pre-decompress nomic model so first query is fast.
+			nomicCacheDir := filepath.Join(rekalDir, "nomic")
+			if err := nomic.WarmCache(nomicCacheDir); err != nil {
+				fmt.Fprintf(cmd.ErrOrStderr(), "rekal: warning: nomic cache warm failed: %v\n", err)
+			}
+
 			// Detect other AI agents and print integration hints.
 			printAgentHints(cmd.ErrOrStderr(), gitRoot)
 

cmd/rekal/cli/nomic/client.go

@@ -0,0 +1,69 @@
+//go:build (darwin && arm64) || (linux && amd64)
+
+package nomic
+
+import "path/filepath"
+
+// Client provides nomic embeddings, transparently using the daemon when
+// available or falling back to in-process embedding.
+type Client struct {
+	daemon   *daemonClient
+	embedder *Embedder
+}
+
+// NewClient creates a Client that tries the daemon first for fast embedding.
+// If the daemon is unavailable, it falls back to loading the model in-process.
+// gitRoot is the git repository root (used to locate .rekal/nomic/).
+func NewClient(gitRoot string) (*Client, error) {
+	if !Supported() {
+		return nil, ErrNotSupported
+	}
+
+	// Try daemon first.
+	dc, err := ensureDaemon(gitRoot)
+	if err == nil {
+		return &Client{daemon: dc}, nil
+	}
+
+	// Fall back to in-process.
+	cacheDir := filepath.Join(gitRoot, ".rekal", "nomic")
+	embedder, err := NewEmbedder(cacheDir)
+	if err != nil {
+		return nil, err
+	}
+	return &Client{embedder: embedder}, nil
+}
+
+// EmbedQuery embeds text with the "search_query: " prefix.
+func (c *Client) EmbedQuery(text string) ([]float64, error) {
+	if c.daemon != nil {
+		return c.daemon.EmbedQuery(text)
+	}
+	return c.embedder.EmbedQuery(text)
+}
+
+// EmbedDocument embeds text with the "search_document: " prefix.
+func (c *Client) EmbedDocument(text string) ([]float64, error) {
+	if c.daemon != nil {
+		return c.daemon.EmbedDocument(text)
+	}
+	return c.embedder.EmbedDocument(text)
+}
+
+// EmbedSessions embeds multiple sessions in batch.
+func (c *Client) EmbedSessions(sessions map[string]string) (map[string][]float64, error) {
+	if c.daemon != nil {
+		return c.daemon.EmbedSessions(sessions)
+	}
+	return c.embedder.EmbedSessions(sessions)
+}
+
+// Close releases resources.
+func (c *Client) Close() {
+	if c.daemon != nil {
+		c.daemon.Close()
+	}
+	if c.embedder != nil {
+		c.embedder.Close()
+	}
+}