From 82686e731af1d1778d6f1bb5c5e271b2d5e46499 Mon Sep 17 00:00:00 2001
From: Anton Devson <99696803+antondevson@users.noreply.github.com>
Date: Tue, 5 Aug 2025 17:39:14 -0400
Subject: [PATCH 1/2] Use correct URL for Ollama embeddings

Since the prefixing of the base url with `/v1` (OpenAI API standard) happens elsewhere and is even saved in the database, it's easier to fix this here for now.
---
 letta/embeddings.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/letta/embeddings.py b/letta/embeddings.py
index dffcd571..0da697f4 100644
--- a/letta/embeddings.py
+++ b/letta/embeddings.py
@@ -159,7 +159,7 @@ class OllamaEmbeddings:
 
         with httpx.Client() as client:
             response = client.post(
-                f"{self.base_url}/api/embeddings",
+                f"{self.base_url}/embeddings",
                 headers=headers,
                 json=json_data,
             )

From 2203ff9092840ca9f9cee02ee911cfda0b7ee0bc Mon Sep 17 00:00:00 2001
From: natn0 <natn0@hotmail.com>
Date: Wed, 6 Aug 2025 01:58:50 -0400
Subject: [PATCH 2/2] Updated call to OpenAI API compatible endpoint in Ollama
 to correct request and return data

---
 letta/embeddings.py | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/letta/embeddings.py b/letta/embeddings.py
index 0da697f4..7dace6ee 100644
--- a/letta/embeddings.py
+++ b/letta/embeddings.py
@@ -139,10 +139,11 @@ class AzureOpenAIEmbedding:
 
 class OllamaEmbeddings:
 
+    # Uses OpenAI API standard
     # Format:
-    # curl http://localhost:11434/api/embeddings -d '{
+    # curl http://localhost:11434/v1/embeddings -d '{
     #   "model": "mxbai-embed-large",
-    #   "prompt": "Llamas are members of the camelid family"
+    #   "input": "Llamas are members of the camelid family"
     # }'
 
     def __init__(self, model: str, base_url: str, ollama_additional_kwargs: dict):
@@ -154,7 +155,7 @@ class OllamaEmbeddings:
         import httpx
 
         headers = {"Content-Type": "application/json"}
-        json_data = {"model": self.model, "prompt": text}
+        json_data = {"model": self.model, "input": text}
         json_data.update(self.ollama_additional_kwargs)
 
         with httpx.Client() as client:
@@ -165,7 +166,7 @@ class OllamaEmbeddings:
             )
 
         response_json = response.json()
-        return response_json["embedding"]
+        return response_json["data"][0]["embedding"]
 
 
 class GoogleEmbeddings: