meshagent-openai 0.0.29__py3-none-any.whl → 0.0.30__py3-none-any.whl

meshagent/openai/proxy/__init__.py

@@ -0,0 +1 @@
+from .proxy import get_client

meshagent/openai/proxy/proxy.py

@@ -0,0 +1,31 @@
+import os
+from meshagent.api import RoomClient
+from openai import AsyncOpenAI
+
+def get_client(*, room: RoomClient) -> AsyncOpenAI:
+
+    if os.getenv("MESHAGENT_API_URL") == None:
+        token : str = room.protocol.token
+        url : str = room.room_url
+        
+        room_proxy_url = f"{url}/v1"
+        if room_proxy_url.startswith("ws:") or room_proxy_url.startswith("wss:"):
+            room_proxy_url = room_proxy_url.replace("ws","http",1)
+
+        openai=AsyncOpenAI(
+            api_key=token,
+            base_url=room_proxy_url,
+            default_headers={
+                "Meshagent-Session" : room.session_id
+            }
+        )
+
+    else:
+    
+        openai=AsyncOpenAI(
+            default_headers={
+                "Meshagent-Session" : room.session_id
+            }
+        )
+
+    return openai

meshagent/openai/tools/stt.py

@@ -0,0 +1,90 @@
+from meshagent.tools import ToolContext, Tool, Toolkit, JsonResponse, TextResponse
+from openai import AsyncOpenAI
+from meshagent.openai.proxy import get_client
+from typing import Optional
+import io
+import pathlib
+
+from meshagent.openai.proxy import get_client
+
+async def _transcribe(*, client: AsyncOpenAI, data: bytes, model: str, filename: str, response_format: str, timestamp_granularities: list[str] = None, prompt: Optional[str] = None, language: Optional[str] = None):
+
+    buf = io.BytesIO(data)
+    buf.name = filename
+    transcript = await client.audio.transcriptions.create(
+        model=model,
+        response_format=response_format,
+        file=buf,
+        prompt=prompt,
+        language=language,
+        timestamp_granularities=timestamp_granularities,
+        stream=False
+    )
+
+    if isinstance(transcript, str):
+        return TextResponse(text=transcript)
+    
+    return JsonResponse(json=transcript.to_dict(mode="json"))
+
+class OpenAIAudioFileSTT(Tool):
+    def __init__(self, *, client: Optional[AsyncOpenAI] = None):
+        super().__init__(
+            name="openai-file-stt",
+            input_schema={
+                "type" : "object",
+                "additionalProperties" : False,
+                "required" : [ "model", "path", "response_format", "timestamp_granularities", "prompt"],
+                "properties" : {
+                    "path" : {
+                        "type": "string",
+                        "description" : "the path to a file in the room storage",
+                    },
+                    "prompt" : {
+                        "type": "string",
+                        "description" : "a prompt. can improve the accuracy of the transcript",
+                    },
+                    "model" : {
+                        "type" : "string",
+                        "enum" : [
+                            "whisper-1", "gpt-4o-mini-transcribe", "gpt-4o-transcribe"
+                        ]
+                    },
+                    "response_format" : {
+                        "type" : "string",
+                        "description" : "text and json are supported for all models, srt, verbose_json, and vtt are only supported for whisper-1",
+                        "enum" : [ "text", "json", "srt", "verbose_json", "vtt" ]
+                    },
+                    "timestamp_granularities" : {
+                        "description" : "timestamp_granularities are only valid with whisper-1",
+                        "type" : "array",
+                        "items" : {
+                            "type" : "string",
+                            "enum" : [ "word", "segment" ]
+                        }
+                    }
+                }
+            },
+            title="OpenAI audio file STT",
+            description="transcribes an audio file to text"
+        )
+        self.client = client
+
+    async def execute(self, context: ToolContext, *, model: str, prompt: str, path: str, response_format: str, timestamp_granularities: list):
+
+        file_data = await context.room.storage.download(path=path)
+        client = self.client
+        if client == None:
+            client = get_client(room=context.room)
+
+        return await _transcribe(client=client, data=file_data.data, model=model, prompt=prompt, filename=pathlib.Path(path).name, response_format=response_format)
+
+
+class OpenAISTTToolkit(Toolkit):
+    def __init__(self):
+        super().__init__(
+            name="openai-stt",
+            description="tools for speech to text using openai",
+            tools=[
+                OpenAIAudioFileSTT()
+            ]
+        )

meshagent/openai/version.py

@@ -1 +1 @@
-__version__ = "0.0.29"
+__version__ = "0.0.30"

{meshagent_openai-0.0.29.dist-info → meshagent_openai-0.0.30.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: meshagent-openai
-Version: 0.0.29
+Version: 0.0.30
 Summary: OpenAI Building Blocks for Meshagent
 License-Expression: Apache-2.0
 Project-URL: Documentation, https://docs.meshagent.com
@@ -13,9 +13,9 @@ Requires-Dist: pyjwt~=2.10.1
 Requires-Dist: pytest~=8.3.5
 Requires-Dist: pytest-asyncio~=0.26.0
 Requires-Dist: openai~=1.70.0
-Requires-Dist: meshagent-api~=0.0.29
-Requires-Dist: meshagent-agents~=0.0.29
-Requires-Dist: meshagent-tools~=0.0.29
+Requires-Dist: meshagent-api~=0.0.30
+Requires-Dist: meshagent-agents~=0.0.30
+Requires-Dist: meshagent-tools~=0.0.30
 Dynamic: license-file
 
 ### Meshagent OpenAI

meshagent_openai-0.0.30.dist-info/RECORD

@@ -0,0 +1,13 @@
+meshagent/openai/__init__.py,sha256=4JRby-ltGfJzrNYhJkMNIpVc2ml2zL_JkkFC0T1_8Vk,174
+meshagent/openai/version.py,sha256=Ji8g_yOyFfpoV8jQQPNpvlINEqW1-F23R31bqwShK9o,22
+meshagent/openai/proxy/__init__.py,sha256=SqoueAmMXHbDKd8O4EeqGkI0gEiC3xLTLlpESGxySPU,30
+meshagent/openai/proxy/proxy.py,sha256=FBlFYVwBVorxif1hbYAG3fz-1J3_XPogpBIHS_OGRW0,822
+meshagent/openai/tools/__init__.py,sha256=RBU_J4qRDuBaaUdi6jpgpuMlIbvT30QmTBrZrYLwsUU,185
+meshagent/openai/tools/completions_adapter.py,sha256=4xEfs57O5GrIRnz1_yVRd3nJpC521lMKFO7RANczRnw,15191
+meshagent/openai/tools/responses_adapter.py,sha256=QdTnnSvsGK0nMgceEX-kwo1lAc59ot7G1CYre-n6aFM,23390
+meshagent/openai/tools/stt.py,sha256=qMuzkOss7Z5zRVsxmZRASugfgsvuWMG3_6toezq68s8,3612
+meshagent_openai-0.0.30.dist-info/licenses/LICENSE,sha256=eTt0SPW-sVNdkZe9PS_S8WfCIyLjRXRl7sUBWdlteFg,10254
+meshagent_openai-0.0.30.dist-info/METADATA,sha256=XQb4qd9uTDHuHGJnuvt2Cp3oJ8lgx0xJZWWvrdhFlUg,660
+meshagent_openai-0.0.30.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+meshagent_openai-0.0.30.dist-info/top_level.txt,sha256=GlcXnHtRP6m7zlG3Df04M35OsHtNXy_DY09oFwWrH74,10
+meshagent_openai-0.0.30.dist-info/RECORD,,

meshagent_openai-0.0.29.dist-info/RECORD

@@ -1,10 +0,0 @@
-meshagent/openai/__init__.py,sha256=4JRby-ltGfJzrNYhJkMNIpVc2ml2zL_JkkFC0T1_8Vk,174
-meshagent/openai/version.py,sha256=kKjqpKrSxIZWidrpy6lRpfZS596w03NuINSgPGAgFN4,22
-meshagent/openai/tools/__init__.py,sha256=RBU_J4qRDuBaaUdi6jpgpuMlIbvT30QmTBrZrYLwsUU,185
-meshagent/openai/tools/completions_adapter.py,sha256=4xEfs57O5GrIRnz1_yVRd3nJpC521lMKFO7RANczRnw,15191
-meshagent/openai/tools/responses_adapter.py,sha256=QdTnnSvsGK0nMgceEX-kwo1lAc59ot7G1CYre-n6aFM,23390
-meshagent_openai-0.0.29.dist-info/licenses/LICENSE,sha256=eTt0SPW-sVNdkZe9PS_S8WfCIyLjRXRl7sUBWdlteFg,10254
-meshagent_openai-0.0.29.dist-info/METADATA,sha256=3WaZFf6lEZzbj3n6N-A8-um_A42JLGhcPrMioy0TWu4,660
-meshagent_openai-0.0.29.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-meshagent_openai-0.0.29.dist-info/top_level.txt,sha256=GlcXnHtRP6m7zlG3Df04M35OsHtNXy_DY09oFwWrH74,10
-meshagent_openai-0.0.29.dist-info/RECORD,,