solana-agent 17.0.1__py3-none-any.whl → 17.0.3__py3-none-any.whl

solana_agent/adapters/llm_adapter.py

@@ -19,7 +19,7 @@ class OpenAIAdapter(LLMProvider):
     def __init__(self, api_key: str):
         self.client = OpenAI(api_key=api_key)
         self.parse_model = "gpt-4o-mini"
-        self.search_model = "gpt-4o-mini-search-preview"
+        self.text_model = "gpt-4o-mini"
         self.transcription_model = "gpt-4o-mini-transcribe"
         self.tts_model = "tts-1"
 
@@ -106,7 +106,7 @@ class OpenAIAdapter(LLMProvider):
         prompt: str,
         system_prompt: str = "",
     ) -> AsyncGenerator[str, None]:  # pragma: no cover
-        """Generate text from OpenAI models with web search capability."""
+        """Generate text from OpenAI models."""
         messages = []
 
         if system_prompt:
@@ -118,7 +118,7 @@ class OpenAIAdapter(LLMProvider):
         request_params = {
             "messages": messages,
             "stream": True,
-            "model": self.search_model,
+            "model": self.text_model,
         }
         try:
             response = self.client.chat.completions.create(**request_params)

solana_agent/services/agent.py

@@ -39,6 +39,7 @@ class AgentService(AgentServiceInterface):
         self.agent_repository = agent_repository
         self.organization_mission = organization_mission
         self.config = config or {}
+        self.last_text_response = ""
 
         # Initialize tool registry with concrete implementation
         if tool_registry:
@@ -247,7 +248,10 @@ class AgentService(AgentServiceInterface):
             if memory_context:
                 system_prompt += f"\n\nMemory Context: {memory_context}"
 
-            # Buffer for collecting text when generating audio
+            # Keep track of the complete text response
+            complete_text_response = ""
+            json_buffer = ""
+            is_json = False
             text_buffer = ""
 
             # Generate and stream response
@@ -255,40 +259,107 @@ class AgentService(AgentServiceInterface):
                 prompt=query_text,
                 system_prompt=system_prompt,
             ):
-                if chunk.strip().startswith("{"):
-                    # Handle tool calls
-                    result = await self._handle_tool_call(
-                        agent_name, chunk,
-                    )
-                    if output_format == "audio":
-                        async for audio_chunk in self.llm_provider.tts(result, instructions=audio_instructions, response_format=audio_output_format, voice=audio_voice):
-                            yield audio_chunk
-                    else:
-                        yield result
+                # Check for JSON start
+                if chunk.strip().startswith("{") and not is_json:
+                    is_json = True
+                    json_buffer = chunk
+                    continue
+
+                # Collect JSON or handle normal text
+                if is_json:
+                    json_buffer += chunk
+                    try:
+                        # Try to parse complete JSON
+                        data = json.loads(json_buffer)
+
+                        # Valid JSON found, handle it
+                        if "tool_call" in data:
+                            # Process tool call with existing method
+                            response_text = await self._handle_tool_call(
+                                agent_name=agent_name,
+                                json_chunk=json_buffer
+                            )
+
+                            # Add to complete text response
+                            complete_text_response += response_text
+
+                            # Output response based on format
+                            if output_format == "audio":
+                                async for audio_chunk in self.llm_provider.tts(
+                                    text=response_text,
+                                    voice=audio_voice,
+                                    response_format=audio_output_format
+                                ):
+                                    yield audio_chunk
+                            else:
+                                yield response_text
+                        else:
+                            # For non-tool JSON, still capture the text
+                            complete_text_response += json_buffer
+
+                            if output_format == "audio":
+                                async for audio_chunk in self.llm_provider.tts(
+                                    text=json_buffer,
+                                    voice=audio_voice,
+                                    response_format=audio_output_format
+                                ):
+                                    yield audio_chunk
+                            else:
+                                yield json_buffer
+
+                        # Reset JSON handling
+                        is_json = False
+                        json_buffer = ""
+
+                    except json.JSONDecodeError:
+                        pass
                 else:
+                    # For regular text, always add to the complete response
+                    complete_text_response += chunk
+
+                    # Handle audio buffering or direct text output
                     if output_format == "audio":
-                        # Buffer text until we have a complete sentence
                         text_buffer += chunk
                         if any(punct in chunk for punct in ".!?"):
                             async for audio_chunk in self.llm_provider.tts(
-                                text_buffer, instructions=audio_instructions, response_format=audio_output_format, voice=audio_voice
+                                text=text_buffer,
+                                voice=audio_voice,
+                                response_format=audio_output_format
                             ):
                                 yield audio_chunk
                             text_buffer = ""
                     else:
                         yield chunk
 
-            # Handle any remaining text in buffer
-            if output_format == "audio" and text_buffer:
-                async for audio_chunk in self.llm_provider.tts(
-                    text_buffer, instructions=audio_instructions, response_format=audio_output_format, voice=audio_voice
-                ):
-                    yield audio_chunk
+            # Handle any remaining text or incomplete JSON
+            remaining_text = ""
+            if text_buffer:
+                remaining_text += text_buffer
+            if is_json and json_buffer:
+                remaining_text += json_buffer
+
+            if remaining_text:
+                # Add remaining text to complete response
+                complete_text_response += remaining_text
+
+                if output_format == "audio":
+                    async for audio_chunk in self.llm_provider.tts(
+                        text=remaining_text,
+                        voice=audio_voice,
+                        response_format=audio_output_format
+                    ):
+                        yield audio_chunk
+                else:
+                    yield remaining_text
+
+            # Store the complete text response for the caller to access
+            # This needs to be done in the query service using the self.last_text_response
+            self.last_text_response = complete_text_response
 
         except Exception as e:
             error_msg = f"I apologize, but I encountered an error: {str(e)}"
             if output_format == "audio":
-                async for chunk in self.llm_provider.tts(error_msg, instructions=audio_instructions, response_format=audio_output_format, voice=audio_voice):
+                async for chunk in self.llm_provider.tts(error_msg, voice=audio_voice, response_format=audio_output_format):
                     yield chunk
             else:
                 yield error_msg

solana_agent/services/query.py

@@ -52,11 +52,8 @@ class QueryService(QueryServiceInterface):
         Args:
             user_id: User ID
             query: Text query or audio bytes
-            output_format: Response format ("text" or "audio")
-            audio_voice: Voice to use for audio output
-            audio_instructions: Optional instructions for audio synthesis
-            audio_output_format: Audio output format
-            audio_input_format: Audio input format
+            output_format: Response format ("text" or "audio") 
+            voice: Voice to use for audio output
 
         Yields:
             Response chunks (text strings or audio bytes)
@@ -74,7 +71,11 @@ class QueryService(QueryServiceInterface):
             if user_text.strip().lower() in ["test", "hello", "hi", "hey", "ping"]:
                 response = "Hello! How can I help you today?"
                 if output_format == "audio":
-                    async for chunk in self.agent_service.llm_provider.tts(response, instructions=audio_instructions, response_format=audio_output_format, voice=audio_voice):
+                    async for chunk in self.agent_service.llm_provider.tts(
+                        text=response,
+                        voice=audio_voice,
+                        response_format=audio_output_format
+                    ):
                         yield chunk
                 else:
                     yield response
@@ -91,42 +92,47 @@ class QueryService(QueryServiceInterface):
 
             # Route query to appropriate agent
             agent_name = await self.routing_service.route_query(user_text)
+            print(f"DEBUG: Routed to agent: {agent_name}")
 
-            # Generate response using agent service
-            full_response = ""
-            async for chunk in self.agent_service.generate_response(
-                agent_name=agent_name,
-                user_id=user_id,
-                query=user_text,
-                memory_context=memory_context,
-                output_format=output_format,
-                audio_voice=audio_voice,
-                audio_output_format=audio_output_format,
-            ):
-                yield chunk
-                if output_format == "text":
+            # For text output, we can directly collect and yield chunks
+            if output_format == "text":
+                full_response = ""
+                async for chunk in self.agent_service.generate_response(
+                    agent_name=agent_name,
+                    user_id=user_id,
+                    query=user_text,
+                    memory_context=memory_context,
+                    output_format="text",
+                ):
+                    yield chunk
                     full_response += chunk
 
-            # For audio responses, get transcription for storage
-            if output_format == "audio":
-                # Re-generate response in text format for storage
+            # For audio output, we'll yield audio chunks while collecting text separately
+            else:
                 async for chunk in self.agent_service.generate_response(
                     agent_name=agent_name,
                     user_id=user_id,
                     query=user_text,
                     memory_context=memory_context,
-                    output_format="text"
+                    output_format="audio",
+                    audio_input_format=audio_input_format,
+                    audio_output_format=audio_output_format,
+                    audio_voice=audio_voice,
                 ):
-                    full_response += chunk
+                    yield chunk
 
-            # Store conversation and extract insights
+            # Store conversation with the full text response
             if self.memory_provider:
-                await self._store_conversation(user_id, user_text, full_response)
+                await self._store_conversation(user_id, user_text, self.agent_service.last_text_response)
 
         except Exception as e:
             error_msg = f"I apologize for the technical difficulty. {str(e)}"
             if output_format == "audio":
-                async for chunk in self.agent_service.llm_provider.tts(error_msg, instructions=audio_instructions, response_format=audio_output_format, voice=audio_voice):
+                async for chunk in self.agent_service.llm_provider.tts(
+                    text=error_msg,
+                    voice=audio_voice,
+                    response_format=audio_output_format
+                ):
                     yield chunk
             else:
                 yield error_msg

{solana_agent-17.0.1.dist-info → solana_agent-17.0.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: solana-agent
-Version: 17.0.1
+Version: 17.0.3
 Summary: The Future of Work
 License: MIT
 Keywords: ai,openai,ai agents,agi
@@ -37,7 +37,6 @@ Description-Content-Type: text/markdown
 * Multi-modal input-output streaming with text or audio by AI Agents
 * Conversational memory per user shared by all AI Agents
 * Routing based on AI Agent specializations
-* Built-in Internet Search for all AI Agents
 * Organizational mission, values, goals, and guidance for all AI Agents
 * Robust AI Agent tool plugins based on standard python packages
 
@@ -110,7 +109,7 @@ async for response in solana_agent.process("user123", "What are the latest AI de
 ```
 
 ## Models Used
-* The model used for AI Agents is `gpt-4o-mini-search-preview`
+* The model used for AI Agents is `gpt-4o-mini`
 * The model used for internal structured outputs is `gpt-4o-mini`
 * The model used for audio_transcription is `gpt-4o-mini-transcribe`
 * The model used for tts is `tts-1`

{solana_agent-17.0.1.dist-info → solana_agent-17.0.3.dist-info}/RECORD

@@ -1,6 +1,6 @@
 solana_agent/__init__.py,sha256=ceYeUpjIitpln8YK1r0JVJU8mzG6cRPYu-HLny3d-Tw,887
 solana_agent/adapters/__init__.py,sha256=tiEEuuy0NF3ngc_tGEcRTt71zVI58v3dYY9RvMrF2Cg,204
-solana_agent/adapters/llm_adapter.py,sha256=s9eZkS1b-m7hK99hPLK421AFkud2jySFg4u4qNB9ayE,6080
+solana_agent/adapters/llm_adapter.py,sha256=hi2JYj6CvhNLJxznIv_7Ef7Y0mk7aztHd_OMYe2scMQ,6034
 solana_agent/adapters/mongodb_adapter.py,sha256=qqEFbY_v1XGyFXBmwd5HSXSSHnA9wWo-Hm1vGEyIG0k,2718
 solana_agent/client/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 solana_agent/client/solana_agent.py,sha256=G0m6GD6N8J3tpjfLoKBMzIR3cZU9qs9cDBfY1mngvzI,4207
@@ -28,10 +28,10 @@ solana_agent/repositories/__init__.py,sha256=fP83w83CGzXLnSdq-C5wbw9EhWTYtqE2lQT
 solana_agent/repositories/agent.py,sha256=e1rnsQiigkKwJNLKro86a3b6TBiky3GMfmCRc5b_jPw,3187
 solana_agent/repositories/memory.py,sha256=GABGwaz00thjviHewLvb18NeKE8dkBROxy_stsiiWrE,4722
 solana_agent/services/__init__.py,sha256=ab_NXJmwYUCmCrCzuTlZ47bJZINW0Y0F5jfQ9OovidU,163
-solana_agent/services/agent.py,sha256=5BCzy8h37HAzvxiuOanu-RM_ARxwIdOUNie8xPLGDj0,14219
-solana_agent/services/query.py,sha256=N0RMcQm7o4B0MGrbFFrZ_Ar5z9r-UUDgn1xzgoxQtqg,10758
+solana_agent/services/agent.py,sha256=j0aI_BGaY5Nhkb9ga0r4BqI3qMKu5TOc4QPQsU3NHyQ,17000
+solana_agent/services/query.py,sha256=1BAb4UjpydwtJjut8oq724BJohe9ceMigMX-rkI5j7s,10820
 solana_agent/services/routing.py,sha256=TPJ2Pas4acE93QzMEV6ZP670OtTNrVEPa76fz6urEV4,4996
-solana_agent-17.0.1.dist-info/LICENSE,sha256=BnSRc-NSFuyF2s496l_4EyrwAP6YimvxWcjPiJ0J7g4,1057
-solana_agent-17.0.1.dist-info/METADATA,sha256=-KgZSOm4jpeQsuy8tCQs6lGFqfwJjTsqf3tZRLs_OwI,4948
-solana_agent-17.0.1.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
-solana_agent-17.0.1.dist-info/RECORD,,
+solana_agent-17.0.3.dist-info/LICENSE,sha256=BnSRc-NSFuyF2s496l_4EyrwAP6YimvxWcjPiJ0J7g4,1057
+solana_agent-17.0.3.dist-info/METADATA,sha256=kioM7hvaI97S8skANaI4oAcS2gnKgGMvI2jKTTHf5vg,4888
+solana_agent-17.0.3.dist-info/WHEEL,sha256=XbeZDeTWKc1w7CSIyre5aMDU_-PohRwTQceYnisIYYY,88
+solana_agent-17.0.3.dist-info/RECORD,,