livellm 1.1.0__tar.gz → 1.1.1__tar.gz

{livellm-1.1.0 → livellm-1.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: livellm
-Version: 1.1.0
+Version: 1.1.1
 Summary: Python client for the LiveLLM Server
 Project-URL: Homepage, https://github.com/qalby-tech/livellm-client-py
 Project-URL: Repository, https://github.com/qalby-tech/livellm-client-py
@@ -39,18 +39,19 @@ Python client library for the LiveLLM Server - a unified proxy for AI agent, aud
 - 🛠️ **Agent tools** - Web search and MCP server integration
 - 🎙️ **Audio services** - Text-to-speech and transcription
 - ⚡ **Fallback strategies** - Sequential and parallel fallback handling
-- 📦 **Context manager support** - Automatic cleanup with async context managers
+- 📦 **Smart resource management** - Automatic cleanup via GC, context managers, or manual control
+- 🧹 **Memory safe** - No resource leaks with multiple cleanup strategies
 
 ## Installation
 
 ```bash
-pip install livellm-client
+pip install livellm
 ```
 
 Or with development dependencies:
 
 ```bash
-pip install livellm-client[testing]
+pip install livellm[testing]
 ```
 
 ## Quick Start
@@ -181,6 +182,18 @@ print(f"Output: {response.output}")
 print(f"Tokens used - Input: {response.usage.input_tokens}, Output: {response.usage.output_tokens}")
 ```
 
+**Note:** You can use either `MessageRole` enum or string values for the `role` parameter:
+
+```python
+# Using enum (recommended for type safety)
+TextMessage(role=MessageRole.USER, content="Hello")
+
+# Using string (more convenient)
+TextMessage(role="user", content="Hello")
+
+# Both work identically and serialize correctly
+```
+
 #### Streaming Agent Response
 
 ```python
@@ -417,9 +430,32 @@ fallback_request = AudioFallbackRequest(
 audio = await client.speak(fallback_request)
 ```
 
-## Context Manager Support
+## Resource Management
+
+The client provides multiple ways to manage resources and cleanup:
+
+### 1. Automatic Cleanup (Garbage Collection)
+
+The client automatically cleans up when garbage collected:
+
+```python
+async def main():
+    client = LivellmClient(base_url="http://localhost:8000")
+    
+    # Use client...
+    response = await client.ping()
+    
+    # No explicit cleanup needed - handled automatically when object is destroyed
+    # Note: Provider configs are deleted synchronously from the server
+
+asyncio.run(main())
+```
+
+**Note**: While automatic cleanup works, it shows a `ResourceWarning` if configs exist to encourage explicit cleanup for immediate resource release.
 
-The client supports async context managers for automatic cleanup:
+### 2. Context Manager (Recommended)
+
+Use async context managers for guaranteed cleanup:
 
 ```python
 async with LivellmClient(base_url="http://localhost:8000") as client:
@@ -433,17 +469,32 @@ async with LivellmClient(base_url="http://localhost:8000") as client:
 # Automatically cleans up configs and closes HTTP client
 ```
 
-Or manually:
+### 3. Manual Cleanup
+
+Explicitly call cleanup in a try/finally block:
 
 ```python
 client = LivellmClient(base_url="http://localhost:8000")
 try:
     # Use client...
-    pass
+    response = await client.ping()
 finally:
     await client.cleanup()
 ```
 
+### Cleanup Behavior
+
+The `cleanup()` method:
+- Deletes all provider configs created by the client
+- Closes the HTTP client connection
+- Is idempotent (safe to call multiple times)
+
+The `__del__()` destructor (automatic cleanup):
+- Triggers when the object is garbage collected
+- Synchronously deletes provider configs from the server
+- Closes the HTTP client connection
+- Shows a `ResourceWarning` if configs exist (to encourage explicit cleanup)
+
 ## API Reference
 
 ### Client Methods
@@ -468,8 +519,9 @@ finally:
 - `transcribe_json(request: TranscribeRequest | TranscribeFallbackRequest) -> TranscribeResponse` - JSON transcription
 
 #### Cleanup
-- `cleanup() -> None` - Clean up resources and close client
+- `cleanup() -> None` - Clean up resources and close client (async)
 - `__aenter__() / __aexit__()` - Async context manager support
+- `__del__()` - Automatic cleanup when garbage collected (sync)
 
 ### Models
 

{livellm-1.1.0 → livellm-1.1.1}/README.md

@@ -14,18 +14,19 @@ Python client library for the LiveLLM Server - a unified proxy for AI agent, aud
 - 🛠️ **Agent tools** - Web search and MCP server integration
 - 🎙️ **Audio services** - Text-to-speech and transcription
 - ⚡ **Fallback strategies** - Sequential and parallel fallback handling
-- 📦 **Context manager support** - Automatic cleanup with async context managers
+- 📦 **Smart resource management** - Automatic cleanup via GC, context managers, or manual control
+- 🧹 **Memory safe** - No resource leaks with multiple cleanup strategies
 
 ## Installation
 
 ```bash
-pip install livellm-client
+pip install livellm
 ```
 
 Or with development dependencies:
 
 ```bash
-pip install livellm-client[testing]
+pip install livellm[testing]
 ```
 
 ## Quick Start
@@ -156,6 +157,18 @@ print(f"Output: {response.output}")
 print(f"Tokens used - Input: {response.usage.input_tokens}, Output: {response.usage.output_tokens}")
 ```
 
+**Note:** You can use either `MessageRole` enum or string values for the `role` parameter:
+
+```python
+# Using enum (recommended for type safety)
+TextMessage(role=MessageRole.USER, content="Hello")
+
+# Using string (more convenient)
+TextMessage(role="user", content="Hello")
+
+# Both work identically and serialize correctly
+```
+
 #### Streaming Agent Response
 
 ```python
@@ -392,9 +405,32 @@ fallback_request = AudioFallbackRequest(
 audio = await client.speak(fallback_request)
 ```
 
-## Context Manager Support
+## Resource Management
+
+The client provides multiple ways to manage resources and cleanup:
+
+### 1. Automatic Cleanup (Garbage Collection)
+
+The client automatically cleans up when garbage collected:
+
+```python
+async def main():
+    client = LivellmClient(base_url="http://localhost:8000")
+    
+    # Use client...
+    response = await client.ping()
+    
+    # No explicit cleanup needed - handled automatically when object is destroyed
+    # Note: Provider configs are deleted synchronously from the server
+
+asyncio.run(main())
+```
+
+**Note**: While automatic cleanup works, it shows a `ResourceWarning` if configs exist to encourage explicit cleanup for immediate resource release.
 
-The client supports async context managers for automatic cleanup:
+### 2. Context Manager (Recommended)
+
+Use async context managers for guaranteed cleanup:
 
 ```python
 async with LivellmClient(base_url="http://localhost:8000") as client:
@@ -408,17 +444,32 @@ async with LivellmClient(base_url="http://localhost:8000") as client:
 # Automatically cleans up configs and closes HTTP client
 ```
 
-Or manually:
+### 3. Manual Cleanup
+
+Explicitly call cleanup in a try/finally block:
 
 ```python
 client = LivellmClient(base_url="http://localhost:8000")
 try:
     # Use client...
-    pass
+    response = await client.ping()
 finally:
     await client.cleanup()
 ```
 
+### Cleanup Behavior
+
+The `cleanup()` method:
+- Deletes all provider configs created by the client
+- Closes the HTTP client connection
+- Is idempotent (safe to call multiple times)
+
+The `__del__()` destructor (automatic cleanup):
+- Triggers when the object is garbage collected
+- Synchronously deletes provider configs from the server
+- Closes the HTTP client connection
+- Shows a `ResourceWarning` if configs exist (to encourage explicit cleanup)
+
 ## API Reference
 
 ### Client Methods
@@ -443,8 +494,9 @@ finally:
 - `transcribe_json(request: TranscribeRequest | TranscribeFallbackRequest) -> TranscribeResponse` - JSON transcription
 
 #### Cleanup
-- `cleanup() -> None` - Clean up resources and close client
+- `cleanup() -> None` - Clean up resources and close client (async)
 - `__aenter__() / __aexit__()` - Async context manager support
+- `__del__()` - Automatic cleanup when garbage collected (sync)
 
 ### Models
 

{livellm-1.1.0 → livellm-1.1.1}/livellm/livellm.py

@@ -1,6 +1,8 @@
 """LiveLLM Client - Python client for the LiveLLM Proxy and Realtime APIs."""
+import asyncio
 import httpx
 import json
+import warnings
 from typing import List, Optional, AsyncIterator, Union
 from .models.common import Settings, SuccessResponse
 from .models.agent.agent import AgentRequest, AgentResponse
@@ -164,6 +166,7 @@ class LivellmClient:
         Should be called when you're done using the client.
         """
         for config in self.settings:
+            config: Settings = config
             await self.delete_config(config.uid)
         await self.client.aclose()
     
@@ -175,6 +178,32 @@ class LivellmClient:
         """Async context manager exit."""
         await self.cleanup()
     
+    def __del__(self):
+        """
+        Destructor to clean up resources when the client is garbage collected.
+        This will close the HTTP client and attempt to delete configs if cleanup wasn't called.
+        Note: It's recommended to use the async context manager or call cleanup() explicitly.
+        """
+        # Warn user if cleanup wasn't called
+        if self.settings:
+            warnings.warn(
+                "LivellmClient is being garbage collected without explicit cleanup. "
+                "Provider configs may not be deleted from the server. "
+                "Consider using 'async with' or calling 'await client.cleanup()' explicitly.",
+                ResourceWarning,
+                stacklevel=2
+            )
+        
+        # Close the httpx client synchronously
+        # httpx.AsyncClient stores a sync Transport that needs cleanup
+        try:
+            with httpx.Client(base_url=self.base_url) as client:
+                for config in self.settings:
+                    config: Settings = config
+                    client.delete("providers/config/{config.uid}", headers=self.headers)
+        except Exception:
+            # Silently fail - we're in a destructor
+            pass
 
     async def agent_run(
         self,

{livellm-1.1.0 → livellm-1.1.1}/livellm/models/agent/chat.py

@@ -1,18 +1,17 @@
 # models for chat messages
-from pydantic import BaseModel, Field, model_validator
+from pydantic import BaseModel, Field, model_validator, field_serializer
 from enum import Enum
-from typing import Optional
+from typing import Optional, Union
 
-class MessageRole(Enum):
+class MessageRole(str, Enum):
     USER = "user"
     MODEL = "model"
     SYSTEM = "system"
 
 
 class Message(BaseModel):
-    role: MessageRole = Field(..., description="The role of the message")
-
-
+    role: Union[MessageRole, str] = Field(..., description="The role of the message")
+    
 class TextMessage(Message):
     content: str = Field(..., description="The content of the message")
 

{livellm-1.1.0 → livellm-1.1.1}/livellm/models/agent/tools.py

@@ -3,7 +3,7 @@ from pydantic import BaseModel, Field, field_validator
 from typing import Literal
 from enum import Enum
 
-class ToolKind(Enum):
+class ToolKind(str, Enum):
     WEB_SEARCH = "web_search"
     MCP_STREAMABLE_SERVER = "mcp_streamable_server"
 

{livellm-1.1.0 → livellm-1.1.1}/livellm/models/audio/speak.py

@@ -6,7 +6,7 @@ from ..common import BaseRequest
 SpeakStreamResponse: TypeAlias = Tuple[AsyncIterator[bytes], str, int]
 
 
-class SpeakMimeType(Enum):
+class SpeakMimeType(str, Enum):
     PCM = "audio/pcm"
     WAV = "audio/wav"
     MP3 = "audio/mpeg"

{livellm-1.1.0 → livellm-1.1.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "livellm"
-version = "1.1.0"
+version = "1.1.1"
 description = "Python client for the LiveLLM Server"
 readme = "README.md"
 requires-python = ">=3.10"