diff --git a/collection.go b/collection.go
index 4262a92..114026d 100644
--- a/collection.go
+++ b/collection.go
@@ -3,6 +3,7 @@ package chromem
 import (
 	"context"
 	"errors"
+	"fmt"
 	"sync"
 )
 
@@ -41,6 +42,20 @@ func newCollection(name string, metadata map[string]string, embed EmbeddingFunc)
 //
 // A row-based API will be added when Chroma adds it (they already plan to).
 func (c *Collection) Add(ctx context.Context, ids []string, embeddings [][]float32, metadatas []map[string]string, documents []string) error {
+	return c.add(ctx, ids, documents, embeddings, metadatas, 1)
+}
+
+// AddConcurrently is like Add, but adds embeddings concurrently.
+// This is mostly useful when you don't pass any embeddings so they have to be created.
+// Upon error, concurrently running operations are canceled and the error is returned.
+func (c *Collection) AddConcurrently(ctx context.Context, ids []string, embeddings [][]float32, metadatas []map[string]string, documents []string, concurrency int) error {
+	if concurrency < 1 {
+		return errors.New("concurrency must be at least 1")
+	}
+	return c.add(ctx, ids, documents, embeddings, metadatas, concurrency)
+}
+
+func (c *Collection) add(ctx context.Context, ids []string, documents []string, embeddings [][]float32, metadatas []map[string]string, concurrency int) error {
 	if len(ids) == 0 || len(documents) == 0 {
 		return errors.New("ids and documents must not be empty")
 	}
@@ -54,23 +69,68 @@ func (c *Collection) Add(ctx context.Context, ids []string, embeddings [][]float
 		return errors.New("ids, metadatas and documents must have the same length")
 	}
 
-	var embedding []float32
-	var metadata map[string]string
-	var err error
+	ctx, cancel := context.WithCancelCause(ctx)
+
+	var wg sync.WaitGroup
+	var globalErr error
+	var globalErrLock sync.RWMutex
+	semaphore := make(chan struct{}, concurrency)
 	for i, document := range documents {
+		var embedding []float32
+		var metadata map[string]string
 		if len(embeddings) != 0 {
 			embedding = embeddings[i]
 		}
 		if len(metadatas) != 0 {
 			metadata = metadatas[i]
 		}
-		c.documentsLock.Lock()
-		// We don't defer the unlock because we want to unlock much earlier
-		c.documents[ids[i]], err = newDocument(ctx, ids[i], embedding, metadata, document, c.embed)
-		c.documentsLock.Unlock()
-		if err != nil {
-			return err
-		}
+
+		wg.Add(1)
+		go func(id string, embedding []float32, metadata map[string]string, document string) {
+			defer wg.Done()
+
+			// Don't even start if we already have an error
+			globalErrLock.RLock()
+			// We don't defer the unlock because we want to unlock much earlier.
+			if globalErr != nil {
+				globalErrLock.RUnlock()
+				return
+			}
+			globalErrLock.RUnlock()
+
+			// Wait here while $concurrency other goroutines are creating documents.
+			semaphore <- struct{}{}
+			defer func() { <-semaphore }()
+
+			err := c.addRow(ctx, id, document, embedding, metadata)
+			if err != nil {
+				globalErrLock.Lock()
+				defer globalErrLock.Unlock()
+				// Another goroutine might have already set the error.
+				if globalErr == nil {
+					globalErr = err
+					// Cancel the operation for all other goroutines.
+					cancel(globalErr)
+				}
+				return
+			}
+		}(ids[i], embedding, metadata, document)
 	}
+
+	wg.Wait()
+
+	return globalErr
+}
+
+func (c *Collection) addRow(ctx context.Context, id string, document string, embedding []float32, metadata map[string]string) error {
+	doc, err := newDocument(ctx, id, embedding, metadata, document, c.embed)
+	if err != nil {
+		return fmt.Errorf("couldn't create document '%s': %w", id, err)
+	}
+
+	c.documentsLock.Lock()
+	defer c.documentsLock.Unlock()
+	c.documents[id] = doc
+
 	return nil
 }