From eab6281923ddc525c17c47b844cd07e2099e3b87 Mon Sep 17 00:00:00 2001
From: Avi Fenesh <55081586+avifenesh@users.noreply.github.com>
Date: Mon, 17 Aug 2026 13:15:51 +0300
Subject: [PATCH] =?UTF-8?q?fix:=20messages=20and=20responses=20are=20suppo?=
 =?UTF-8?q?rted=20=E2=80=94=20measured,=20not=20assumed?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

I first declared both false on the grounds that I had not tested LiteLLM's bridge
against this endpoint. That was the wrong way to resolve it; I ran the calls instead.

Against litellm 1.97.0 with a live key:

  litellm.responses(...).output_text            -> 'ok'
  await litellm.anthropic_messages(...)         -> Anthropic-shaped content blocks,
                                                   with reasoning mapped into a
                                                   'thinking' block
  litellm.completion(..., stream=True)          -> content plus 110 reasoning_content
                                                   chunks
  litellm.text_completion(...)                  -> 'Paris.'
  tool_calls, response_format json_schema       -> both correct

Everything left false was probed too, rather than assumed: embeddings and moderations
fall through to another provider's credentials, audio and rerank raise 'tiyuvta is not
a valid LlmProviders', batches needs a files API this endpoint does not have, and
litellm 1.97.0 exposes no a2a entry point.
---
 provider_endpoints_support.json | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/provider_endpoints_support.json b/provider_endpoints_support.json
index 38bcb15588f..4530e3716de 100644
--- a/provider_endpoints_support.json
+++ b/provider_endpoints_support.json
@@ -2371,8 +2371,8 @@
       "url": "https://inference.tiyuvta.ai/docs",
       "endpoints": {
         "chat_completions": true,
-        "messages": false,
-        "responses": false,
+        "messages": true,
+        "responses": true,
         "embeddings": false,
         "image_generations": false,
         "audio_transcriptions": false,