26 Oct 20:53

arthurbrenno

b5a1d7c

v0.9.21

Changelog

v0.9.21

feat(apis): new boolean type, recursive object schema serialization, request config and more

Assets 2

25 Oct 18:26

arthurbrenno

v0.9.20

5eaeb47

v0.9.20

Changelog

v0.9.20

Assets 2

25 Oct 17:53

arthurbrenno

v0.9.19

ad3d504

v0.9.19

Changelog

v0.9.19

fix(whatsapp): Optimize typing indicator and response handling

Fix response handling logic in EvolutionAPI provider
Modify typing indicator to send only once before sending multiple messages
Improve error handling for typing indicator failures
Prevent typing indicator errors from interrupting message sending process
Optimize logging for typing indicator and message sending

v0.9.18

refactor(whatsapp): Update expected_status handling in API requests

Modify expected_status parameter in request methods to accept both int and Sequence[int]
Update all relevant API request calls to use a list for expected_status values
Enhance resilience in handling HTTP responses by allowing multiple acceptable status codes

v0.9.17

feat(whatsapp): Add recording indicator for TTS audio synthesis

Implement send_recording_indicator method in WhatsApp base provider
Add recording indicator support for Evolution API provider
Update WhatsApp bot to show recording indicator during TTS audio synthesis
Enhance TTS audio response flow with async synthesis and recording indicator
Improve logging and error handling for recording indicator functionality
Adds a visual cue for users during text-to-speech audio generation, providing a more interactive and responsive messaging experience.

v0.9.16

feat(whatsapp): Add text-to-speech (TTS) integration for WhatsApp bot

Implement speech configuration in WhatsAppBotConfig
Add speech_play_chance and speech_config fields to control TTS behavior
Create send_audio_message method in WhatsAppProvider base class
Implement send_audio_message for EvolutionAPI provider
Add documentation and example for WhatsApp TTS integration
Enhance WhatsApp bot to support optional audio message generation
This change enables flexible text-to-speech capabilities for WhatsApp bots, allowing dynamic audio response generation with configurable probability and provider settings.

v0.9.15

rollback changes

v0.9.10

feat(docx-parser, pdf-parser): Implement max_output_tokens for response generation

Add max_output_tokens attribute to DocxFileParser and PDFFileParser for controlling token generation in responses
Update relevant methods to utilize max_output_tokens in generation configuration

v0.9.9

fix(openrouter-adapter): Improve tool call argument parsing with robust error handling

Add comprehensive JSON parsing error handling for tool call arguments
Implement fallback mechanism to extract first valid JSON object when parsing fails
Log detailed warnings and errors for malformed JSON input
Prevent adapter from breaking when encountering invalid JSON in tool call arguments
Ensure graceful degradation by using empty dict if parsing completely fails

v0.9.8

fix(whatsapp): Improve phone number ID extraction for WhatsApp webhook payloads

Update phone number ID extraction logic for linked IDs
Set remoteJid to remoteJidAlt when "@lid" is present in key
Ensure consistent phone number ID handling for different WhatsApp message types

v0.9.7

fix(whatsapp): not returning early when "lid" is present.

v0.9.6

fix(stdio-mcp-server): Improve JSON message parsing and connection state handling

Enhance message parsing to skip non-JSON lines in stdout
Add debug logging for skipped lines and invalid JSON
Modify connection state checks to allow requests/notifications during initialization
Prevent potential errors from processing non-JSON output from MCP servers
Improve error handling and logging for message decoding

v0.9.5

fix(whatsapp): Improve WhatsApp webhook payload phone number ID extraction

Update phone number ID extraction logic for WhatsApp webhook payloads
Add fallback extraction method using remoteJid when remoteJidAlt is not available
Ensure consistent phone number ID extraction across different webhook scenarios
Prevent potential null or incomplete phone number ID assignments

v0.9.4

feat(whatsapp): Improve WhatsApp webhook payload and key handling

Add remoteJidAlt field to Key model for enhanced JID parsing
Update phone number ID extraction logic in webhook payload processing
Modify message handling to use remoteJidAlt for linked WhatsApp accounts
Enhance robustness of JID parsing for different WhatsApp account types

v0.9.3

🚀 New Responses API (Experimental)

feat(responses): Brand new experimental Responses API with comprehensive OpenRouter integration
- Complete OpenRouter API compatibility with streaming support and structured outputs
- Advanced function tool management with FunctionCallStore for tracking and executing function calls
- Enhanced FunctionTool class with robust type safety and parameter conversion
- ToolPair system for managing tool definitions and call history
- Support for both synchronous and asynchronous tool execution
- Comprehensive response streaming with AsyncStream implementation
- Integration with orjson for improved JSON handling performance
- Support for reasoning, background processing, and conversation management
- Factory methods for easy instantiation (openrouter(), openai())

🌐 Web Extraction & Automation

feat(web): Complete web extraction and automation system
- Extractor class: Intelligent web content extraction with structured output support
- Browser automation: Full Playwright integration with configurable options
- Action system: Comprehensive set of web actions including:
  - Click, scroll, write text, take screenshots
  - Generate PDFs, execute JavaScript, press keys
  - Wait for conditions, viewport management
- Extraction preferences: Advanced configuration options:
  - Geolocation support, mobile/desktop modes
  - Ad blocking, TLS verification control
  - Custom headers, timeouts, and proxy settings
  - Content filtering (include/exclude tags)
- Structured prompts: AI-powered content analysis with markdown conversion
- Error handling: Graceful handling of invalid URLs and extraction failures

🔧 Enhanced Assistant & Tracing

feat(assistant): Enhanced Assistant model with new fields and Responder integration
feat(tracing): Improved tracing system with BaseModel inheritance
- Refactored LangfuseOtelClient and NoOpOtelClient to inherit from BaseModel
- Enhanced type safety and error handling across tracing components
- Better parameter conversion and callable management

🛠️ Function Call Management

feat(responses): Advanced function call management system
- Methods for retrieving, removing, and clearing function calls
- Enhanced tool presence checking and call count retrieval
- Support for multiple function calls with improved access patterns
- Example usage documentation for better developer experience

🐛 Bug Fixes & Improvements

fix(responses): Multiple fixes for structured outputs and JSON schema handling
- Proper JSON schema "additionalProperties" setting and retrieval
- Improved handling of incomplete structured outputs with clearer error messaging
- Fixed tool calls and response object text format handling
- Streamlined response object creation and event handling
refactor(file): Commented out MIME type validation for broader file type acceptance
fix(responses): Enhanced OpenRouter example with updated max output tokens

📈 Performance & Developer Experience

feat(responses): AsyncStream implementation for efficient streaming responses
refactor(responses, web): Simplified JSON response handling and improved output parsing
feat(utils): Added needs utility for dependency management
Enhanced error messaging and debugging capabilities across all new features

v0.9.2

feat(agent): cache instructions in Agent class and Developer message for the providers that supports it (cache ephemeral)

v0.9.1

🎯 Pricing Integration

feat(generations): Added Pricing model to Generation class with input_pricing, output_pricing, and total_pricing fields
feat(providers): All major providers now automatically populate pricing information in Generation objects
- OpenRouter: Dynamic pricing from API with automatic per-token to per-million conversion
- OpenAI: Static pricing table with comprehensive model coverage
- Google (Gemini): Static pricing for all Gemini models
- Cerebras: Static pricing for Llama models
- Amazon Bedrock: Static pricing for Claude and other models
feat(adapters): Added async adapt_async() methods to all provider adapters for pricing calculation
fix(openrouter): Corrected pricing calculation - OpenRouter returns per-token prices, now properly converted to per-million-tokens
Pricing is calculated using the abstract methods price_per_million_tokens_input() and price_per_million_tokens_output()
Costs are rounded to 8 decimal places for precision
Graceful fallback to empty Pricing object if calculation fails

🐛 Bug Fixes

fix(openrouter): when fallback models are used, use "models" in the requests, not "model"
fix(openrouter): accept single string in with_fallback_models() method (was treating string as iterable)

🚨 Error Handling

feat(openrouter): comprehensive custom exception hierarchy with detailed error messages
- Created 16+ custom exception classes for all documented OpenRouter errors
- Each exception includes clear descriptions, possible causes, and actionable solutions
- Automatic error parsing from response status codes and error messages
- Examples: ModelNotFoundError, InsufficientCreditsError, ContextLengthExceededError, etc.
- See examples/openrouter_error_handling.py for usage examples

v0.9.0 🚀

Release Date: October 18, 2024

We're thrilled to announce Agentle v0.9.0, a major milestone packed with powerful new features, critical fixes, and performance improvements. This release r...

Contributors

lid

Assets 2

25 Oct 17:27

arthurbrenno

v0.9.18

6f7ff7f

v0.9.18

Changelog

v0.9.18

refactor(whatsapp): Update expected_status handling in API requests

Modify expected_status parameter in request methods to accept both int and Sequence[int]
Update all relevant API request calls to use a list for expected_status values
Enhance resilience in handling HTTP responses by allowing multiple acceptable status codes

v0.9.17

feat(whatsapp): Add recording indicator for TTS audio synthesis

Implement send_recording_indicator method in WhatsApp base provider
Add recording indicator support for Evolution API provider
Update WhatsApp bot to show recording indicator during TTS audio synthesis
Enhance TTS audio response flow with async synthesis and recording indicator
Improve logging and error handling for recording indicator functionality
Adds a visual cue for users during text-to-speech audio generation, providing a more interactive and responsive messaging experience.

v0.9.16

feat(whatsapp): Add text-to-speech (TTS) integration for WhatsApp bot

Implement speech configuration in WhatsAppBotConfig
Add speech_play_chance and speech_config fields to control TTS behavior
Create send_audio_message method in WhatsAppProvider base class
Implement send_audio_message for EvolutionAPI provider
Add documentation and example for WhatsApp TTS integration
Enhance WhatsApp bot to support optional audio message generation
This change enables flexible text-to-speech capabilities for WhatsApp bots, allowing dynamic audio response generation with configurable probability and provider settings.

v0.9.15

rollback changes

v0.9.10

feat(docx-parser, pdf-parser): Implement max_output_tokens for response generation

Add max_output_tokens attribute to DocxFileParser and PDFFileParser for controlling token generation in responses
Update relevant methods to utilize max_output_tokens in generation configuration

v0.9.9

fix(openrouter-adapter): Improve tool call argument parsing with robust error handling

Add comprehensive JSON parsing error handling for tool call arguments
Implement fallback mechanism to extract first valid JSON object when parsing fails
Log detailed warnings and errors for malformed JSON input
Prevent adapter from breaking when encountering invalid JSON in tool call arguments
Ensure graceful degradation by using empty dict if parsing completely fails

v0.9.8

fix(whatsapp): Improve phone number ID extraction for WhatsApp webhook payloads

Update phone number ID extraction logic for linked IDs
Set remoteJid to remoteJidAlt when "@lid" is present in key
Ensure consistent phone number ID handling for different WhatsApp message types

v0.9.7

fix(whatsapp): not returning early when "lid" is present.

v0.9.6

fix(stdio-mcp-server): Improve JSON message parsing and connection state handling

Enhance message parsing to skip non-JSON lines in stdout
Add debug logging for skipped lines and invalid JSON
Modify connection state checks to allow requests/notifications during initialization
Prevent potential errors from processing non-JSON output from MCP servers
Improve error handling and logging for message decoding

v0.9.5

fix(whatsapp): Improve WhatsApp webhook payload phone number ID extraction

Update phone number ID extraction logic for WhatsApp webhook payloads
Add fallback extraction method using remoteJid when remoteJidAlt is not available
Ensure consistent phone number ID extraction across different webhook scenarios
Prevent potential null or incomplete phone number ID assignments

v0.9.4

feat(whatsapp): Improve WhatsApp webhook payload and key handling

Add remoteJidAlt field to Key model for enhanced JID parsing
Update phone number ID extraction logic in webhook payload processing
Modify message handling to use remoteJidAlt for linked WhatsApp accounts
Enhance robustness of JID parsing for different WhatsApp account types

v0.9.3

🚀 New Responses API (Experimental)

feat(responses): Brand new experimental Responses API with comprehensive OpenRouter integration
- Complete OpenRouter API compatibility with streaming support and structured outputs
- Advanced function tool management with FunctionCallStore for tracking and executing function calls
- Enhanced FunctionTool class with robust type safety and parameter conversion
- ToolPair system for managing tool definitions and call history
- Support for both synchronous and asynchronous tool execution
- Comprehensive response streaming with AsyncStream implementation
- Integration with orjson for improved JSON handling performance
- Support for reasoning, background processing, and conversation management
- Factory methods for easy instantiation (openrouter(), openai())

🌐 Web Extraction & Automation

feat(web): Complete web extraction and automation system
- Extractor class: Intelligent web content extraction with structured output support
- Browser automation: Full Playwright integration with configurable options
- Action system: Comprehensive set of web actions including:
  - Click, scroll, write text, take screenshots
  - Generate PDFs, execute JavaScript, press keys
  - Wait for conditions, viewport management
- Extraction preferences: Advanced configuration options:
  - Geolocation support, mobile/desktop modes
  - Ad blocking, TLS verification control
  - Custom headers, timeouts, and proxy settings
  - Content filtering (include/exclude tags)
- Structured prompts: AI-powered content analysis with markdown conversion
- Error handling: Graceful handling of invalid URLs and extraction failures

🔧 Enhanced Assistant & Tracing

feat(assistant): Enhanced Assistant model with new fields and Responder integration
feat(tracing): Improved tracing system with BaseModel inheritance
- Refactored LangfuseOtelClient and NoOpOtelClient to inherit from BaseModel
- Enhanced type safety and error handling across tracing components
- Better parameter conversion and callable management

🛠️ Function Call Management

feat(responses): Advanced function call management system
- Methods for retrieving, removing, and clearing function calls
- Enhanced tool presence checking and call count retrieval
- Support for multiple function calls with improved access patterns
- Example usage documentation for better developer experience

🐛 Bug Fixes & Improvements

fix(responses): Multiple fixes for structured outputs and JSON schema handling
- Proper JSON schema "additionalProperties" setting and retrieval
- Improved handling of incomplete structured outputs with clearer error messaging
- Fixed tool calls and response object text format handling
- Streamlined response object creation and event handling
refactor(file): Commented out MIME type validation for broader file type acceptance
fix(responses): Enhanced OpenRouter example with updated max output tokens

📈 Performance & Developer Experience

feat(responses): AsyncStream implementation for efficient streaming responses
refactor(responses, web): Simplified JSON response handling and improved output parsing
feat(utils): Added needs utility for dependency management
Enhanced error messaging and debugging capabilities across all new features

v0.9.2

feat(agent): cache instructions in Agent class and Developer message for the providers that supports it (cache ephemeral)

v0.9.1

🎯 Pricing Integration

feat(generations): Added Pricing model to Generation class with input_pricing, output_pricing, and total_pricing fields
feat(providers): All major providers now automatically populate pricing information in Generation objects
- OpenRouter: Dynamic pricing from API with automatic per-token to per-million conversion
- OpenAI: Static pricing table with comprehensive model coverage
- Google (Gemini): Static pricing for all Gemini models
- Cerebras: Static pricing for Llama models
- Amazon Bedrock: Static pricing for Claude and other models
feat(adapters): Added async adapt_async() methods to all provider adapters for pricing calculation
fix(openrouter): Corrected pricing calculation - OpenRouter returns per-token prices, now properly converted to per-million-tokens
Pricing is calculated using the abstract methods price_per_million_tokens_input() and price_per_million_tokens_output()
Costs are rounded to 8 decimal places for precision
Graceful fallback to empty Pricing object if calculation fails

🐛 Bug Fixes

fix(openrouter): when fallback models are used, use "models" in the requests, not "model"
fix(openrouter): accept single string in with_fallback_models() method (was treating string as iterable)

🚨 Error Handling

feat(openrouter): comprehensive custom exception hierarchy with detailed error messages
- Created 16+ custom exception classes for all documented OpenRouter errors
- Each exception includes clear descriptions, possible causes, and actionable solutions
- Automatic error parsing from response status codes and error messages
- Examples: ModelNotFoundError, InsufficientCreditsError, ContextLengthExceededError, etc.
- See examples/openrouter_error_handling.py for usage examples

v0.9.0 🚀

Release Date: October 18, 2024

We're thrilled to announce Agentle v0.9.0, a major milestone packed with powerful new features, critical fixes, and performance improvements. This release represents months of refinement across our parsing, provider, and agent ecosystems.

✨ Highlights

🛡️ Guardrails System

NEW: Comprehensive guardrails integration for safer AI interactions
Tool Leakage Validator: Prevent sensitive tool information from leaking in responses

🔌 OpenRouter Provider

Full OpenRouter integration with streaming support and ...

Contributors

lid

Assets 2

25 Oct 17:22

arthurbrenno

v0.9.17

44e5f90

v0.9.17

Changelog

v0.9.17

feat(whatsapp): Add recording indicator for TTS audio synthesis

Implement send_recording_indicator method in WhatsApp base provider
Add recording indicator support for Evolution API provider
Update WhatsApp bot to show recording indicator during TTS audio synthesis
Enhance TTS audio response flow with async synthesis and recording indicator
Improve logging and error handling for recording indicator functionality
Adds a visual cue for users during text-to-speech audio generation, providing a more interactive and responsive messaging experience.

v0.9.16

feat(whatsapp): Add text-to-speech (TTS) integration for WhatsApp bot

Implement speech configuration in WhatsAppBotConfig
Add speech_play_chance and speech_config fields to control TTS behavior
Create send_audio_message method in WhatsAppProvider base class
Implement send_audio_message for EvolutionAPI provider
Add documentation and example for WhatsApp TTS integration
Enhance WhatsApp bot to support optional audio message generation
This change enables flexible text-to-speech capabilities for WhatsApp bots, allowing dynamic audio response generation with configurable probability and provider settings.

v0.9.15

rollback changes

v0.9.10

feat(docx-parser, pdf-parser): Implement max_output_tokens for response generation

Add max_output_tokens attribute to DocxFileParser and PDFFileParser for controlling token generation in responses
Update relevant methods to utilize max_output_tokens in generation configuration

v0.9.9

fix(openrouter-adapter): Improve tool call argument parsing with robust error handling

Add comprehensive JSON parsing error handling for tool call arguments
Implement fallback mechanism to extract first valid JSON object when parsing fails
Log detailed warnings and errors for malformed JSON input
Prevent adapter from breaking when encountering invalid JSON in tool call arguments
Ensure graceful degradation by using empty dict if parsing completely fails

v0.9.8

fix(whatsapp): Improve phone number ID extraction for WhatsApp webhook payloads

Update phone number ID extraction logic for linked IDs
Set remoteJid to remoteJidAlt when "@lid" is present in key
Ensure consistent phone number ID handling for different WhatsApp message types

v0.9.7

fix(whatsapp): not returning early when "lid" is present.

v0.9.6

fix(stdio-mcp-server): Improve JSON message parsing and connection state handling

Enhance message parsing to skip non-JSON lines in stdout
Add debug logging for skipped lines and invalid JSON
Modify connection state checks to allow requests/notifications during initialization
Prevent potential errors from processing non-JSON output from MCP servers
Improve error handling and logging for message decoding

v0.9.5

fix(whatsapp): Improve WhatsApp webhook payload phone number ID extraction

Update phone number ID extraction logic for WhatsApp webhook payloads
Add fallback extraction method using remoteJid when remoteJidAlt is not available
Ensure consistent phone number ID extraction across different webhook scenarios
Prevent potential null or incomplete phone number ID assignments

v0.9.4

feat(whatsapp): Improve WhatsApp webhook payload and key handling

Add remoteJidAlt field to Key model for enhanced JID parsing
Update phone number ID extraction logic in webhook payload processing
Modify message handling to use remoteJidAlt for linked WhatsApp accounts
Enhance robustness of JID parsing for different WhatsApp account types

v0.9.3

🚀 New Responses API (Experimental)

feat(responses): Brand new experimental Responses API with comprehensive OpenRouter integration
- Complete OpenRouter API compatibility with streaming support and structured outputs
- Advanced function tool management with FunctionCallStore for tracking and executing function calls
- Enhanced FunctionTool class with robust type safety and parameter conversion
- ToolPair system for managing tool definitions and call history
- Support for both synchronous and asynchronous tool execution
- Comprehensive response streaming with AsyncStream implementation
- Integration with orjson for improved JSON handling performance
- Support for reasoning, background processing, and conversation management
- Factory methods for easy instantiation (openrouter(), openai())

🌐 Web Extraction & Automation

feat(web): Complete web extraction and automation system
- Extractor class: Intelligent web content extraction with structured output support
- Browser automation: Full Playwright integration with configurable options
- Action system: Comprehensive set of web actions including:
  - Click, scroll, write text, take screenshots
  - Generate PDFs, execute JavaScript, press keys
  - Wait for conditions, viewport management
- Extraction preferences: Advanced configuration options:
  - Geolocation support, mobile/desktop modes
  - Ad blocking, TLS verification control
  - Custom headers, timeouts, and proxy settings
  - Content filtering (include/exclude tags)
- Structured prompts: AI-powered content analysis with markdown conversion
- Error handling: Graceful handling of invalid URLs and extraction failures

🔧 Enhanced Assistant & Tracing

feat(assistant): Enhanced Assistant model with new fields and Responder integration
feat(tracing): Improved tracing system with BaseModel inheritance
- Refactored LangfuseOtelClient and NoOpOtelClient to inherit from BaseModel
- Enhanced type safety and error handling across tracing components
- Better parameter conversion and callable management

🛠️ Function Call Management

feat(responses): Advanced function call management system
- Methods for retrieving, removing, and clearing function calls
- Enhanced tool presence checking and call count retrieval
- Support for multiple function calls with improved access patterns
- Example usage documentation for better developer experience

🐛 Bug Fixes & Improvements

fix(responses): Multiple fixes for structured outputs and JSON schema handling
- Proper JSON schema "additionalProperties" setting and retrieval
- Improved handling of incomplete structured outputs with clearer error messaging
- Fixed tool calls and response object text format handling
- Streamlined response object creation and event handling
refactor(file): Commented out MIME type validation for broader file type acceptance
fix(responses): Enhanced OpenRouter example with updated max output tokens

📈 Performance & Developer Experience

feat(responses): AsyncStream implementation for efficient streaming responses
refactor(responses, web): Simplified JSON response handling and improved output parsing
feat(utils): Added needs utility for dependency management
Enhanced error messaging and debugging capabilities across all new features

v0.9.2

feat(agent): cache instructions in Agent class and Developer message for the providers that supports it (cache ephemeral)

v0.9.1

🎯 Pricing Integration

feat(generations): Added Pricing model to Generation class with input_pricing, output_pricing, and total_pricing fields
feat(providers): All major providers now automatically populate pricing information in Generation objects
- OpenRouter: Dynamic pricing from API with automatic per-token to per-million conversion
- OpenAI: Static pricing table with comprehensive model coverage
- Google (Gemini): Static pricing for all Gemini models
- Cerebras: Static pricing for Llama models
- Amazon Bedrock: Static pricing for Claude and other models
feat(adapters): Added async adapt_async() methods to all provider adapters for pricing calculation
fix(openrouter): Corrected pricing calculation - OpenRouter returns per-token prices, now properly converted to per-million-tokens
Pricing is calculated using the abstract methods price_per_million_tokens_input() and price_per_million_tokens_output()
Costs are rounded to 8 decimal places for precision
Graceful fallback to empty Pricing object if calculation fails

🐛 Bug Fixes

fix(openrouter): when fallback models are used, use "models" in the requests, not "model"
fix(openrouter): accept single string in with_fallback_models() method (was treating string as iterable)

🚨 Error Handling

feat(openrouter): comprehensive custom exception hierarchy with detailed error messages
- Created 16+ custom exception classes for all documented OpenRouter errors
- Each exception includes clear descriptions, possible causes, and actionable solutions
- Automatic error parsing from response status codes and error messages
- Examples: ModelNotFoundError, InsufficientCreditsError, ContextLengthExceededError, etc.
- See examples/openrouter_error_handling.py for usage examples

v0.9.0 🚀

Release Date: October 18, 2024

✨ Highlights

🛡️ Guardrails System

NEW: Comprehensive guardrails integration for safer AI interactions
Tool Leakage Validator: Prevent sensitive tool information from leaking in responses

🔌 OpenRouter Provider

Full OpenRouter integration with streaming support and structured outputs
Dynamic price fetching for cost-aware model selection
Model fallbacks for improved reliability
Factory methods for easier instantiation
Enhanced tool adapter with complex type expansion in JSON Schema conversion
Configurable httpx client timeout for generation tasks

📄 **Document Parsing Revolution*...

Contributors

lid

Assets 2

25 Oct 16:43

arthurbrenno

v0.9.16

1c49dd4

v0.9.16

Changelog

v0.9.16

feat(whatsapp): Add text-to-speech (TTS) integration for WhatsApp bot

Implement speech configuration in WhatsAppBotConfig
Add speech_play_chance and speech_config fields to control TTS behavior
Create send_audio_message method in WhatsAppProvider base class
Implement send_audio_message for EvolutionAPI provider
Add documentation and example for WhatsApp TTS integration
Enhance WhatsApp bot to support optional audio message generation
This change enables flexible text-to-speech capabilities for WhatsApp bots, allowing dynamic audio response generation with configurable probability and provider settings.

v0.9.15

rollback changes

v0.9.10

feat(docx-parser, pdf-parser): Implement max_output_tokens for response generation

Add max_output_tokens attribute to DocxFileParser and PDFFileParser for controlling token generation in responses
Update relevant methods to utilize max_output_tokens in generation configuration

v0.9.9

fix(openrouter-adapter): Improve tool call argument parsing with robust error handling

Add comprehensive JSON parsing error handling for tool call arguments
Implement fallback mechanism to extract first valid JSON object when parsing fails
Log detailed warnings and errors for malformed JSON input
Prevent adapter from breaking when encountering invalid JSON in tool call arguments
Ensure graceful degradation by using empty dict if parsing completely fails

v0.9.8

fix(whatsapp): Improve phone number ID extraction for WhatsApp webhook payloads

Update phone number ID extraction logic for linked IDs
Set remoteJid to remoteJidAlt when "@lid" is present in key
Ensure consistent phone number ID handling for different WhatsApp message types

v0.9.7

fix(whatsapp): not returning early when "lid" is present.

v0.9.6

fix(stdio-mcp-server): Improve JSON message parsing and connection state handling

Enhance message parsing to skip non-JSON lines in stdout
Add debug logging for skipped lines and invalid JSON
Modify connection state checks to allow requests/notifications during initialization
Prevent potential errors from processing non-JSON output from MCP servers
Improve error handling and logging for message decoding

v0.9.5

fix(whatsapp): Improve WhatsApp webhook payload phone number ID extraction

Update phone number ID extraction logic for WhatsApp webhook payloads
Add fallback extraction method using remoteJid when remoteJidAlt is not available
Ensure consistent phone number ID extraction across different webhook scenarios
Prevent potential null or incomplete phone number ID assignments

v0.9.4

feat(whatsapp): Improve WhatsApp webhook payload and key handling

Add remoteJidAlt field to Key model for enhanced JID parsing
Update phone number ID extraction logic in webhook payload processing
Modify message handling to use remoteJidAlt for linked WhatsApp accounts
Enhance robustness of JID parsing for different WhatsApp account types

v0.9.3

🚀 New Responses API (Experimental)

feat(responses): Brand new experimental Responses API with comprehensive OpenRouter integration
- Complete OpenRouter API compatibility with streaming support and structured outputs
- Advanced function tool management with FunctionCallStore for tracking and executing function calls
- Enhanced FunctionTool class with robust type safety and parameter conversion
- ToolPair system for managing tool definitions and call history
- Support for both synchronous and asynchronous tool execution
- Comprehensive response streaming with AsyncStream implementation
- Integration with orjson for improved JSON handling performance
- Support for reasoning, background processing, and conversation management
- Factory methods for easy instantiation (openrouter(), openai())

🌐 Web Extraction & Automation

feat(web): Complete web extraction and automation system
- Extractor class: Intelligent web content extraction with structured output support
- Browser automation: Full Playwright integration with configurable options
- Action system: Comprehensive set of web actions including:
  - Click, scroll, write text, take screenshots
  - Generate PDFs, execute JavaScript, press keys
  - Wait for conditions, viewport management
- Extraction preferences: Advanced configuration options:
  - Geolocation support, mobile/desktop modes
  - Ad blocking, TLS verification control
  - Custom headers, timeouts, and proxy settings
  - Content filtering (include/exclude tags)
- Structured prompts: AI-powered content analysis with markdown conversion
- Error handling: Graceful handling of invalid URLs and extraction failures

🔧 Enhanced Assistant & Tracing

feat(assistant): Enhanced Assistant model with new fields and Responder integration
feat(tracing): Improved tracing system with BaseModel inheritance
- Refactored LangfuseOtelClient and NoOpOtelClient to inherit from BaseModel
- Enhanced type safety and error handling across tracing components
- Better parameter conversion and callable management

🛠️ Function Call Management

feat(responses): Advanced function call management system
- Methods for retrieving, removing, and clearing function calls
- Enhanced tool presence checking and call count retrieval
- Support for multiple function calls with improved access patterns
- Example usage documentation for better developer experience

🐛 Bug Fixes & Improvements

fix(responses): Multiple fixes for structured outputs and JSON schema handling
- Proper JSON schema "additionalProperties" setting and retrieval
- Improved handling of incomplete structured outputs with clearer error messaging
- Fixed tool calls and response object text format handling
- Streamlined response object creation and event handling
refactor(file): Commented out MIME type validation for broader file type acceptance
fix(responses): Enhanced OpenRouter example with updated max output tokens

📈 Performance & Developer Experience

feat(responses): AsyncStream implementation for efficient streaming responses
refactor(responses, web): Simplified JSON response handling and improved output parsing
feat(utils): Added needs utility for dependency management
Enhanced error messaging and debugging capabilities across all new features

v0.9.2

feat(agent): cache instructions in Agent class and Developer message for the providers that supports it (cache ephemeral)

v0.9.1

🎯 Pricing Integration

feat(generations): Added Pricing model to Generation class with input_pricing, output_pricing, and total_pricing fields
feat(providers): All major providers now automatically populate pricing information in Generation objects
- OpenRouter: Dynamic pricing from API with automatic per-token to per-million conversion
- OpenAI: Static pricing table with comprehensive model coverage
- Google (Gemini): Static pricing for all Gemini models
- Cerebras: Static pricing for Llama models
- Amazon Bedrock: Static pricing for Claude and other models
feat(adapters): Added async adapt_async() methods to all provider adapters for pricing calculation
fix(openrouter): Corrected pricing calculation - OpenRouter returns per-token prices, now properly converted to per-million-tokens
Pricing is calculated using the abstract methods price_per_million_tokens_input() and price_per_million_tokens_output()
Costs are rounded to 8 decimal places for precision
Graceful fallback to empty Pricing object if calculation fails

🐛 Bug Fixes

fix(openrouter): when fallback models are used, use "models" in the requests, not "model"
fix(openrouter): accept single string in with_fallback_models() method (was treating string as iterable)

🚨 Error Handling

feat(openrouter): comprehensive custom exception hierarchy with detailed error messages
- Created 16+ custom exception classes for all documented OpenRouter errors
- Each exception includes clear descriptions, possible causes, and actionable solutions
- Automatic error parsing from response status codes and error messages
- Examples: ModelNotFoundError, InsufficientCreditsError, ContextLengthExceededError, etc.
- See examples/openrouter_error_handling.py for usage examples

v0.9.0 🚀

Release Date: October 18, 2024

✨ Highlights

🛡️ Guardrails System

NEW: Comprehensive guardrails integration for safer AI interactions
Tool Leakage Validator: Prevent sensitive tool information from leaking in responses

🔌 OpenRouter Provider

Full OpenRouter integration with streaming support and structured outputs
Dynamic price fetching for cost-aware model selection
Model fallbacks for improved reliability
Factory methods for easier instantiation
Enhanced tool adapter with complex type expansion in JSON Schema conversion
Configurable httpx client timeout for generation tasks

📄 Document Parsing Revolution

Native PDF processing with PyMuPDF for superior performance
RTF Document Parser - brand new format support
Native DOCX processing option for faster document handling
Enhanced PDF parsing with encrypted file support and structured exception handling
PPTX parsing improvements with legacy conversion support
Legacy .doc file conversion with improved error handling
Optimized archive processing (ZIP, RAR) with better error messaging
Page screenshot optimization to reduce per-image analysis costs
Configurable render_scale for PDF screenshot...

Contributors

lid

Assets 2

25 Oct 14:35

arthurbrenno

v0.9.15

ed66211

v0.9.15

Changelog

v0.9.15

rollback changes

v0.9.10

feat(docx-parser, pdf-parser): Implement max_output_tokens for response generation

Add max_output_tokens attribute to DocxFileParser and PDFFileParser for controlling token generation in responses
Update relevant methods to utilize max_output_tokens in generation configuration

v0.9.9

fix(openrouter-adapter): Improve tool call argument parsing with robust error handling

Add comprehensive JSON parsing error handling for tool call arguments
Implement fallback mechanism to extract first valid JSON object when parsing fails
Log detailed warnings and errors for malformed JSON input
Prevent adapter from breaking when encountering invalid JSON in tool call arguments
Ensure graceful degradation by using empty dict if parsing completely fails

v0.9.8

fix(whatsapp): Improve phone number ID extraction for WhatsApp webhook payloads

Update phone number ID extraction logic for linked IDs
Set remoteJid to remoteJidAlt when "@lid" is present in key
Ensure consistent phone number ID handling for different WhatsApp message types

v0.9.7

fix(whatsapp): not returning early when "lid" is present.

v0.9.6

fix(stdio-mcp-server): Improve JSON message parsing and connection state handling

Enhance message parsing to skip non-JSON lines in stdout
Add debug logging for skipped lines and invalid JSON
Modify connection state checks to allow requests/notifications during initialization
Prevent potential errors from processing non-JSON output from MCP servers
Improve error handling and logging for message decoding

v0.9.5

fix(whatsapp): Improve WhatsApp webhook payload phone number ID extraction

Update phone number ID extraction logic for WhatsApp webhook payloads
Add fallback extraction method using remoteJid when remoteJidAlt is not available
Ensure consistent phone number ID extraction across different webhook scenarios
Prevent potential null or incomplete phone number ID assignments

v0.9.4

feat(whatsapp): Improve WhatsApp webhook payload and key handling

Add remoteJidAlt field to Key model for enhanced JID parsing
Update phone number ID extraction logic in webhook payload processing
Modify message handling to use remoteJidAlt for linked WhatsApp accounts
Enhance robustness of JID parsing for different WhatsApp account types

v0.9.3

🚀 New Responses API (Experimental)

feat(responses): Brand new experimental Responses API with comprehensive OpenRouter integration
- Complete OpenRouter API compatibility with streaming support and structured outputs
- Advanced function tool management with FunctionCallStore for tracking and executing function calls
- Enhanced FunctionTool class with robust type safety and parameter conversion
- ToolPair system for managing tool definitions and call history
- Support for both synchronous and asynchronous tool execution
- Comprehensive response streaming with AsyncStream implementation
- Integration with orjson for improved JSON handling performance
- Support for reasoning, background processing, and conversation management
- Factory methods for easy instantiation (openrouter(), openai())

🌐 Web Extraction & Automation

feat(web): Complete web extraction and automation system
- Extractor class: Intelligent web content extraction with structured output support
- Browser automation: Full Playwright integration with configurable options
- Action system: Comprehensive set of web actions including:
  - Click, scroll, write text, take screenshots
  - Generate PDFs, execute JavaScript, press keys
  - Wait for conditions, viewport management
- Extraction preferences: Advanced configuration options:
  - Geolocation support, mobile/desktop modes
  - Ad blocking, TLS verification control
  - Custom headers, timeouts, and proxy settings
  - Content filtering (include/exclude tags)
- Structured prompts: AI-powered content analysis with markdown conversion
- Error handling: Graceful handling of invalid URLs and extraction failures

🔧 Enhanced Assistant & Tracing

feat(assistant): Enhanced Assistant model with new fields and Responder integration
feat(tracing): Improved tracing system with BaseModel inheritance
- Refactored LangfuseOtelClient and NoOpOtelClient to inherit from BaseModel
- Enhanced type safety and error handling across tracing components
- Better parameter conversion and callable management

🛠️ Function Call Management

feat(responses): Advanced function call management system
- Methods for retrieving, removing, and clearing function calls
- Enhanced tool presence checking and call count retrieval
- Support for multiple function calls with improved access patterns
- Example usage documentation for better developer experience

🐛 Bug Fixes & Improvements

fix(responses): Multiple fixes for structured outputs and JSON schema handling
- Proper JSON schema "additionalProperties" setting and retrieval
- Improved handling of incomplete structured outputs with clearer error messaging
- Fixed tool calls and response object text format handling
- Streamlined response object creation and event handling
refactor(file): Commented out MIME type validation for broader file type acceptance
fix(responses): Enhanced OpenRouter example with updated max output tokens

📈 Performance & Developer Experience

feat(responses): AsyncStream implementation for efficient streaming responses
refactor(responses, web): Simplified JSON response handling and improved output parsing
feat(utils): Added needs utility for dependency management
Enhanced error messaging and debugging capabilities across all new features

v0.9.2

feat(agent): cache instructions in Agent class and Developer message for the providers that supports it (cache ephemeral)

v0.9.1

🎯 Pricing Integration

feat(generations): Added Pricing model to Generation class with input_pricing, output_pricing, and total_pricing fields
feat(providers): All major providers now automatically populate pricing information in Generation objects
- OpenRouter: Dynamic pricing from API with automatic per-token to per-million conversion
- OpenAI: Static pricing table with comprehensive model coverage
- Google (Gemini): Static pricing for all Gemini models
- Cerebras: Static pricing for Llama models
- Amazon Bedrock: Static pricing for Claude and other models
feat(adapters): Added async adapt_async() methods to all provider adapters for pricing calculation
fix(openrouter): Corrected pricing calculation - OpenRouter returns per-token prices, now properly converted to per-million-tokens
Pricing is calculated using the abstract methods price_per_million_tokens_input() and price_per_million_tokens_output()
Costs are rounded to 8 decimal places for precision
Graceful fallback to empty Pricing object if calculation fails

🐛 Bug Fixes

fix(openrouter): when fallback models are used, use "models" in the requests, not "model"
fix(openrouter): accept single string in with_fallback_models() method (was treating string as iterable)

🚨 Error Handling

feat(openrouter): comprehensive custom exception hierarchy with detailed error messages
- Created 16+ custom exception classes for all documented OpenRouter errors
- Each exception includes clear descriptions, possible causes, and actionable solutions
- Automatic error parsing from response status codes and error messages
- Examples: ModelNotFoundError, InsufficientCreditsError, ContextLengthExceededError, etc.
- See examples/openrouter_error_handling.py for usage examples

v0.9.0 🚀

Release Date: October 18, 2024

✨ Highlights

🛡️ Guardrails System

NEW: Comprehensive guardrails integration for safer AI interactions
Tool Leakage Validator: Prevent sensitive tool information from leaking in responses

🔌 OpenRouter Provider

Full OpenRouter integration with streaming support and structured outputs
Dynamic price fetching for cost-aware model selection
Model fallbacks for improved reliability
Factory methods for easier instantiation
Enhanced tool adapter with complex type expansion in JSON Schema conversion
Configurable httpx client timeout for generation tasks

📄 Document Parsing Revolution

Native PDF processing with PyMuPDF for superior performance
RTF Document Parser - brand new format support
Native DOCX processing option for faster document handling
Enhanced PDF parsing with encrypted file support and structured exception handling
PPTX parsing improvements with legacy conversion support
Legacy .doc file conversion with improved error handling
Optimized archive processing (ZIP, RAR) with better error messaging
Page screenshot optimization to reduce per-image analysis costs
Configurable render_scale for PDF screenshot quality control
max_concurrent_pages field for PDF parsing concurrency control

🧠 Embeddings & Vector Stores

DeepInfra Embedding Provider - OpenAI-compatible API integration
Batch embedding generation methods for async processing
Enhanced async embedding methods with None value handling
Load balancer implementation (DuckDB and in-memory) with provider quotas and ranking

💬 WhatsApp Enhancements

Enhanced markdown formatting (headers, tables, blockquotes, horizontal rules)
Improved message splitting with list item grouping
send_message method for independent message sending
Multiple ca...

Contributors

lid

Assets 2

25 Oct 14:21

arthurbrenno

v0.9.14

1cea58c

v0.9.14

Changelog

v0.9.14

fix(openrouter-adapter): Suppress verbose HTTP library logging

Add log level configuration for httpcore and httpx loggers
Set log levels to WARNING to prevent debug logs from interfering with response streams
Improve logging clarity and reduce unnecessary log noise in OpenRouter adapter

v0.9.13

fix(openrouter-adapter): Add debug logging for OpenRouter response

Add debug logging to track raw OpenRouter API response
Helps with troubleshooting and understanding API interactions
Provides visibility into response data before parsing

v0.9.12

fix(openrouter-adapter): Improve JSON response parsing with robust error handling

Add more detailed error logging for JSON parsing failures
Read response text before attempting JSON parsing to prevent body consumption
Include response length in error log for better diagnostics
Use json.loads() instead of .json() method for more explicit parsing
Enhance error message with additional context about response parsing

v0.9.11

fix(openrouter-adapter): Improve JSON response parsing with robust error handling

Add comprehensive error handling for JSON parsing in OpenRouter response
Capture and log detailed error information when JSON parsing fails
Include response status code, content type, and preview in error logging
Raise a descriptive ValueError with context when JSON parsing fails
Enhance error traceability for debugging OpenRouter API interactions

v0.9.10

feat(docx-parser, pdf-parser): Implement max_output_tokens for response generation

Add max_output_tokens attribute to DocxFileParser and PDFFileParser for controlling token generation in responses
Update relevant methods to utilize max_output_tokens in generation configuration

v0.9.9

fix(openrouter-adapter): Improve tool call argument parsing with robust error handling

Add comprehensive JSON parsing error handling for tool call arguments
Implement fallback mechanism to extract first valid JSON object when parsing fails
Log detailed warnings and errors for malformed JSON input
Prevent adapter from breaking when encountering invalid JSON in tool call arguments
Ensure graceful degradation by using empty dict if parsing completely fails

v0.9.8

fix(whatsapp): Improve phone number ID extraction for WhatsApp webhook payloads

Update phone number ID extraction logic for linked IDs
Set remoteJid to remoteJidAlt when "@lid" is present in key
Ensure consistent phone number ID handling for different WhatsApp message types

v0.9.7

fix(whatsapp): not returning early when "lid" is present.

v0.9.6

fix(stdio-mcp-server): Improve JSON message parsing and connection state handling

Enhance message parsing to skip non-JSON lines in stdout
Add debug logging for skipped lines and invalid JSON
Modify connection state checks to allow requests/notifications during initialization
Prevent potential errors from processing non-JSON output from MCP servers
Improve error handling and logging for message decoding

v0.9.5

fix(whatsapp): Improve WhatsApp webhook payload phone number ID extraction

Update phone number ID extraction logic for WhatsApp webhook payloads
Add fallback extraction method using remoteJid when remoteJidAlt is not available
Ensure consistent phone number ID extraction across different webhook scenarios
Prevent potential null or incomplete phone number ID assignments

v0.9.4

feat(whatsapp): Improve WhatsApp webhook payload and key handling

Add remoteJidAlt field to Key model for enhanced JID parsing
Update phone number ID extraction logic in webhook payload processing
Modify message handling to use remoteJidAlt for linked WhatsApp accounts
Enhance robustness of JID parsing for different WhatsApp account types

v0.9.3

🚀 New Responses API (Experimental)

feat(responses): Brand new experimental Responses API with comprehensive OpenRouter integration
- Complete OpenRouter API compatibility with streaming support and structured outputs
- Advanced function tool management with FunctionCallStore for tracking and executing function calls
- Enhanced FunctionTool class with robust type safety and parameter conversion
- ToolPair system for managing tool definitions and call history
- Support for both synchronous and asynchronous tool execution
- Comprehensive response streaming with AsyncStream implementation
- Integration with orjson for improved JSON handling performance
- Support for reasoning, background processing, and conversation management
- Factory methods for easy instantiation (openrouter(), openai())

🌐 Web Extraction & Automation

feat(web): Complete web extraction and automation system
- Extractor class: Intelligent web content extraction with structured output support
- Browser automation: Full Playwright integration with configurable options
- Action system: Comprehensive set of web actions including:
  - Click, scroll, write text, take screenshots
  - Generate PDFs, execute JavaScript, press keys
  - Wait for conditions, viewport management
- Extraction preferences: Advanced configuration options:
  - Geolocation support, mobile/desktop modes
  - Ad blocking, TLS verification control
  - Custom headers, timeouts, and proxy settings
  - Content filtering (include/exclude tags)
- Structured prompts: AI-powered content analysis with markdown conversion
- Error handling: Graceful handling of invalid URLs and extraction failures

🔧 Enhanced Assistant & Tracing

feat(assistant): Enhanced Assistant model with new fields and Responder integration
feat(tracing): Improved tracing system with BaseModel inheritance
- Refactored LangfuseOtelClient and NoOpOtelClient to inherit from BaseModel
- Enhanced type safety and error handling across tracing components
- Better parameter conversion and callable management

🛠️ Function Call Management

feat(responses): Advanced function call management system
- Methods for retrieving, removing, and clearing function calls
- Enhanced tool presence checking and call count retrieval
- Support for multiple function calls with improved access patterns
- Example usage documentation for better developer experience

🐛 Bug Fixes & Improvements

fix(responses): Multiple fixes for structured outputs and JSON schema handling
- Proper JSON schema "additionalProperties" setting and retrieval
- Improved handling of incomplete structured outputs with clearer error messaging
- Fixed tool calls and response object text format handling
- Streamlined response object creation and event handling
refactor(file): Commented out MIME type validation for broader file type acceptance
fix(responses): Enhanced OpenRouter example with updated max output tokens

📈 Performance & Developer Experience

feat(responses): AsyncStream implementation for efficient streaming responses
refactor(responses, web): Simplified JSON response handling and improved output parsing
feat(utils): Added needs utility for dependency management
Enhanced error messaging and debugging capabilities across all new features

v0.9.2

feat(agent): cache instructions in Agent class and Developer message for the providers that supports it (cache ephemeral)

v0.9.1

🎯 Pricing Integration

feat(generations): Added Pricing model to Generation class with input_pricing, output_pricing, and total_pricing fields
feat(providers): All major providers now automatically populate pricing information in Generation objects
- OpenRouter: Dynamic pricing from API with automatic per-token to per-million conversion
- OpenAI: Static pricing table with comprehensive model coverage
- Google (Gemini): Static pricing for all Gemini models
- Cerebras: Static pricing for Llama models
- Amazon Bedrock: Static pricing for Claude and other models
feat(adapters): Added async adapt_async() methods to all provider adapters for pricing calculation
fix(openrouter): Corrected pricing calculation - OpenRouter returns per-token prices, now properly converted to per-million-tokens
Pricing is calculated using the abstract methods price_per_million_tokens_input() and price_per_million_tokens_output()
Costs are rounded to 8 decimal places for precision
Graceful fallback to empty Pricing object if calculation fails

🐛 Bug Fixes

fix(openrouter): when fallback models are used, use "models" in the requests, not "model"
fix(openrouter): accept single string in with_fallback_models() method (was treating string as iterable)

🚨 Error Handling

feat(openrouter): comprehensive custom exception hierarchy with detailed error messages
- Created 16+ custom exception classes for all documented OpenRouter errors
- Each exception includes clear descriptions, possible causes, and actionable solutions
- Automatic error parsing from response status codes and error messages
- Examples: ModelNotFoundError, InsufficientCreditsError, ContextLengthExceededError, etc.
- See examples/openrouter_error_handling.py for usage examples

v0.9.0 🚀

Release Date: October 18, 2024

✨ Highlights

🛡️ Guardrails System

NEW: Comprehensive guardrails integration for safer AI interactions
Tool Leakage Validator: Prevent sensitive tool information from leaking in responses

🔌 OpenRouter Provider

Full OpenRouter integration with streaming support and structured outputs
Dynamic price fetching for cost-aware model selection
Model fallbacks for improved reliabilit...

Contributors

lid

Assets 2

25 Oct 14:06

arthurbrenno

v0.9.13

66cbd40

v0.9.13

Changelog

v0.9.13

fix(openrouter-adapter): Add debug logging for OpenRouter response

Add debug logging to track raw OpenRouter API response
Helps with troubleshooting and understanding API interactions
Provides visibility into response data before parsing

v0.9.12

fix(openrouter-adapter): Improve JSON response parsing with robust error handling

Add more detailed error logging for JSON parsing failures
Read response text before attempting JSON parsing to prevent body consumption
Include response length in error log for better diagnostics
Use json.loads() instead of .json() method for more explicit parsing
Enhance error message with additional context about response parsing

v0.9.11

fix(openrouter-adapter): Improve JSON response parsing with robust error handling

Add comprehensive error handling for JSON parsing in OpenRouter response
Capture and log detailed error information when JSON parsing fails
Include response status code, content type, and preview in error logging
Raise a descriptive ValueError with context when JSON parsing fails
Enhance error traceability for debugging OpenRouter API interactions

v0.9.10

feat(docx-parser, pdf-parser): Implement max_output_tokens for response generation

Add max_output_tokens attribute to DocxFileParser and PDFFileParser for controlling token generation in responses
Update relevant methods to utilize max_output_tokens in generation configuration

v0.9.9

fix(openrouter-adapter): Improve tool call argument parsing with robust error handling

Add comprehensive JSON parsing error handling for tool call arguments
Implement fallback mechanism to extract first valid JSON object when parsing fails
Log detailed warnings and errors for malformed JSON input
Prevent adapter from breaking when encountering invalid JSON in tool call arguments
Ensure graceful degradation by using empty dict if parsing completely fails

v0.9.8

fix(whatsapp): Improve phone number ID extraction for WhatsApp webhook payloads

Update phone number ID extraction logic for linked IDs
Set remoteJid to remoteJidAlt when "@lid" is present in key
Ensure consistent phone number ID handling for different WhatsApp message types

v0.9.7

fix(whatsapp): not returning early when "lid" is present.

v0.9.6

fix(stdio-mcp-server): Improve JSON message parsing and connection state handling

Enhance message parsing to skip non-JSON lines in stdout
Add debug logging for skipped lines and invalid JSON
Modify connection state checks to allow requests/notifications during initialization
Prevent potential errors from processing non-JSON output from MCP servers
Improve error handling and logging for message decoding

v0.9.5

fix(whatsapp): Improve WhatsApp webhook payload phone number ID extraction

Update phone number ID extraction logic for WhatsApp webhook payloads
Add fallback extraction method using remoteJid when remoteJidAlt is not available
Ensure consistent phone number ID extraction across different webhook scenarios
Prevent potential null or incomplete phone number ID assignments

v0.9.4

feat(whatsapp): Improve WhatsApp webhook payload and key handling

Add remoteJidAlt field to Key model for enhanced JID parsing
Update phone number ID extraction logic in webhook payload processing
Modify message handling to use remoteJidAlt for linked WhatsApp accounts
Enhance robustness of JID parsing for different WhatsApp account types

v0.9.3

🚀 New Responses API (Experimental)

feat(responses): Brand new experimental Responses API with comprehensive OpenRouter integration
- Complete OpenRouter API compatibility with streaming support and structured outputs
- Advanced function tool management with FunctionCallStore for tracking and executing function calls
- Enhanced FunctionTool class with robust type safety and parameter conversion
- ToolPair system for managing tool definitions and call history
- Support for both synchronous and asynchronous tool execution
- Comprehensive response streaming with AsyncStream implementation
- Integration with orjson for improved JSON handling performance
- Support for reasoning, background processing, and conversation management
- Factory methods for easy instantiation (openrouter(), openai())

🌐 Web Extraction & Automation

feat(web): Complete web extraction and automation system
- Extractor class: Intelligent web content extraction with structured output support
- Browser automation: Full Playwright integration with configurable options
- Action system: Comprehensive set of web actions including:
  - Click, scroll, write text, take screenshots
  - Generate PDFs, execute JavaScript, press keys
  - Wait for conditions, viewport management
- Extraction preferences: Advanced configuration options:
  - Geolocation support, mobile/desktop modes
  - Ad blocking, TLS verification control
  - Custom headers, timeouts, and proxy settings
  - Content filtering (include/exclude tags)
- Structured prompts: AI-powered content analysis with markdown conversion
- Error handling: Graceful handling of invalid URLs and extraction failures

🔧 Enhanced Assistant & Tracing

feat(assistant): Enhanced Assistant model with new fields and Responder integration
feat(tracing): Improved tracing system with BaseModel inheritance
- Refactored LangfuseOtelClient and NoOpOtelClient to inherit from BaseModel
- Enhanced type safety and error handling across tracing components
- Better parameter conversion and callable management

🛠️ Function Call Management

feat(responses): Advanced function call management system
- Methods for retrieving, removing, and clearing function calls
- Enhanced tool presence checking and call count retrieval
- Support for multiple function calls with improved access patterns
- Example usage documentation for better developer experience

🐛 Bug Fixes & Improvements

fix(responses): Multiple fixes for structured outputs and JSON schema handling
- Proper JSON schema "additionalProperties" setting and retrieval
- Improved handling of incomplete structured outputs with clearer error messaging
- Fixed tool calls and response object text format handling
- Streamlined response object creation and event handling
refactor(file): Commented out MIME type validation for broader file type acceptance
fix(responses): Enhanced OpenRouter example with updated max output tokens

📈 Performance & Developer Experience

feat(responses): AsyncStream implementation for efficient streaming responses
refactor(responses, web): Simplified JSON response handling and improved output parsing
feat(utils): Added needs utility for dependency management
Enhanced error messaging and debugging capabilities across all new features

v0.9.2

feat(agent): cache instructions in Agent class and Developer message for the providers that supports it (cache ephemeral)

v0.9.1

🎯 Pricing Integration

feat(generations): Added Pricing model to Generation class with input_pricing, output_pricing, and total_pricing fields
feat(providers): All major providers now automatically populate pricing information in Generation objects
- OpenRouter: Dynamic pricing from API with automatic per-token to per-million conversion
- OpenAI: Static pricing table with comprehensive model coverage
- Google (Gemini): Static pricing for all Gemini models
- Cerebras: Static pricing for Llama models
- Amazon Bedrock: Static pricing for Claude and other models
feat(adapters): Added async adapt_async() methods to all provider adapters for pricing calculation
fix(openrouter): Corrected pricing calculation - OpenRouter returns per-token prices, now properly converted to per-million-tokens
Pricing is calculated using the abstract methods price_per_million_tokens_input() and price_per_million_tokens_output()
Costs are rounded to 8 decimal places for precision
Graceful fallback to empty Pricing object if calculation fails

🐛 Bug Fixes

fix(openrouter): when fallback models are used, use "models" in the requests, not "model"
fix(openrouter): accept single string in with_fallback_models() method (was treating string as iterable)

🚨 Error Handling

feat(openrouter): comprehensive custom exception hierarchy with detailed error messages
- Created 16+ custom exception classes for all documented OpenRouter errors
- Each exception includes clear descriptions, possible causes, and actionable solutions
- Automatic error parsing from response status codes and error messages
- Examples: ModelNotFoundError, InsufficientCreditsError, ContextLengthExceededError, etc.
- See examples/openrouter_error_handling.py for usage examples

v0.9.0 🚀

Release Date: October 18, 2024

✨ Highlights

🛡️ Guardrails System

NEW: Comprehensive guardrails integration for safer AI interactions
Tool Leakage Validator: Prevent sensitive tool information from leaking in responses

🔌 OpenRouter Provider

Full OpenRouter integration with streaming support and structured outputs
Dynamic price fetching for cost-aware model selection
Model fallbacks for improved reliability
Factory methods for easier instantiation
Enhanced tool adapter with complex type expansion in JSON Schema conversion
Configurable httpx client timeout for generation tasks

📄 Document Parsing Revolution

Native PDF processing with PyMuPDF for superior performance
**RTF Document ...

Contributors

lid

Assets 2

25 Oct 13:53

arthurbrenno

v0.9.12

e6167f8

v0.9.12

Changelog

v0.9.12

fix(openrouter-adapter): Improve JSON response parsing with robust error handling

Add more detailed error logging for JSON parsing failures
Read response text before attempting JSON parsing to prevent body consumption
Include response length in error log for better diagnostics
Use json.loads() instead of .json() method for more explicit parsing
Enhance error message with additional context about response parsing

v0.9.11

fix(openrouter-adapter): Improve JSON response parsing with robust error handling

Add comprehensive error handling for JSON parsing in OpenRouter response
Capture and log detailed error information when JSON parsing fails
Include response status code, content type, and preview in error logging
Raise a descriptive ValueError with context when JSON parsing fails
Enhance error traceability for debugging OpenRouter API interactions

v0.9.10

feat(docx-parser, pdf-parser): Implement max_output_tokens for response generation

Add max_output_tokens attribute to DocxFileParser and PDFFileParser for controlling token generation in responses
Update relevant methods to utilize max_output_tokens in generation configuration

v0.9.9

fix(openrouter-adapter): Improve tool call argument parsing with robust error handling

Add comprehensive JSON parsing error handling for tool call arguments
Implement fallback mechanism to extract first valid JSON object when parsing fails
Log detailed warnings and errors for malformed JSON input
Prevent adapter from breaking when encountering invalid JSON in tool call arguments
Ensure graceful degradation by using empty dict if parsing completely fails

v0.9.8

fix(whatsapp): Improve phone number ID extraction for WhatsApp webhook payloads

Update phone number ID extraction logic for linked IDs
Set remoteJid to remoteJidAlt when "@lid" is present in key
Ensure consistent phone number ID handling for different WhatsApp message types

v0.9.7

fix(whatsapp): not returning early when "lid" is present.

v0.9.6

fix(stdio-mcp-server): Improve JSON message parsing and connection state handling

Enhance message parsing to skip non-JSON lines in stdout
Add debug logging for skipped lines and invalid JSON
Modify connection state checks to allow requests/notifications during initialization
Prevent potential errors from processing non-JSON output from MCP servers
Improve error handling and logging for message decoding

v0.9.5

fix(whatsapp): Improve WhatsApp webhook payload phone number ID extraction

Update phone number ID extraction logic for WhatsApp webhook payloads
Add fallback extraction method using remoteJid when remoteJidAlt is not available
Ensure consistent phone number ID extraction across different webhook scenarios
Prevent potential null or incomplete phone number ID assignments

v0.9.4

feat(whatsapp): Improve WhatsApp webhook payload and key handling

Add remoteJidAlt field to Key model for enhanced JID parsing
Update phone number ID extraction logic in webhook payload processing
Modify message handling to use remoteJidAlt for linked WhatsApp accounts
Enhance robustness of JID parsing for different WhatsApp account types

v0.9.3

🚀 New Responses API (Experimental)

feat(responses): Brand new experimental Responses API with comprehensive OpenRouter integration
- Complete OpenRouter API compatibility with streaming support and structured outputs
- Advanced function tool management with FunctionCallStore for tracking and executing function calls
- Enhanced FunctionTool class with robust type safety and parameter conversion
- ToolPair system for managing tool definitions and call history
- Support for both synchronous and asynchronous tool execution
- Comprehensive response streaming with AsyncStream implementation
- Integration with orjson for improved JSON handling performance
- Support for reasoning, background processing, and conversation management
- Factory methods for easy instantiation (openrouter(), openai())

🌐 Web Extraction & Automation

feat(web): Complete web extraction and automation system
- Extractor class: Intelligent web content extraction with structured output support
- Browser automation: Full Playwright integration with configurable options
- Action system: Comprehensive set of web actions including:
  - Click, scroll, write text, take screenshots
  - Generate PDFs, execute JavaScript, press keys
  - Wait for conditions, viewport management
- Extraction preferences: Advanced configuration options:
  - Geolocation support, mobile/desktop modes
  - Ad blocking, TLS verification control
  - Custom headers, timeouts, and proxy settings
  - Content filtering (include/exclude tags)
- Structured prompts: AI-powered content analysis with markdown conversion
- Error handling: Graceful handling of invalid URLs and extraction failures

🔧 Enhanced Assistant & Tracing

feat(assistant): Enhanced Assistant model with new fields and Responder integration
feat(tracing): Improved tracing system with BaseModel inheritance
- Refactored LangfuseOtelClient and NoOpOtelClient to inherit from BaseModel
- Enhanced type safety and error handling across tracing components
- Better parameter conversion and callable management

🛠️ Function Call Management

feat(responses): Advanced function call management system
- Methods for retrieving, removing, and clearing function calls
- Enhanced tool presence checking and call count retrieval
- Support for multiple function calls with improved access patterns
- Example usage documentation for better developer experience

🐛 Bug Fixes & Improvements

fix(responses): Multiple fixes for structured outputs and JSON schema handling
- Proper JSON schema "additionalProperties" setting and retrieval
- Improved handling of incomplete structured outputs with clearer error messaging
- Fixed tool calls and response object text format handling
- Streamlined response object creation and event handling
refactor(file): Commented out MIME type validation for broader file type acceptance
fix(responses): Enhanced OpenRouter example with updated max output tokens

📈 Performance & Developer Experience

feat(responses): AsyncStream implementation for efficient streaming responses
refactor(responses, web): Simplified JSON response handling and improved output parsing
feat(utils): Added needs utility for dependency management
Enhanced error messaging and debugging capabilities across all new features

v0.9.2

feat(agent): cache instructions in Agent class and Developer message for the providers that supports it (cache ephemeral)

v0.9.1

🎯 Pricing Integration

feat(generations): Added Pricing model to Generation class with input_pricing, output_pricing, and total_pricing fields
feat(providers): All major providers now automatically populate pricing information in Generation objects
- OpenRouter: Dynamic pricing from API with automatic per-token to per-million conversion
- OpenAI: Static pricing table with comprehensive model coverage
- Google (Gemini): Static pricing for all Gemini models
- Cerebras: Static pricing for Llama models
- Amazon Bedrock: Static pricing for Claude and other models
feat(adapters): Added async adapt_async() methods to all provider adapters for pricing calculation
fix(openrouter): Corrected pricing calculation - OpenRouter returns per-token prices, now properly converted to per-million-tokens
Pricing is calculated using the abstract methods price_per_million_tokens_input() and price_per_million_tokens_output()
Costs are rounded to 8 decimal places for precision
Graceful fallback to empty Pricing object if calculation fails

🐛 Bug Fixes

fix(openrouter): when fallback models are used, use "models" in the requests, not "model"
fix(openrouter): accept single string in with_fallback_models() method (was treating string as iterable)

🚨 Error Handling

feat(openrouter): comprehensive custom exception hierarchy with detailed error messages
- Created 16+ custom exception classes for all documented OpenRouter errors
- Each exception includes clear descriptions, possible causes, and actionable solutions
- Automatic error parsing from response status codes and error messages
- Examples: ModelNotFoundError, InsufficientCreditsError, ContextLengthExceededError, etc.
- See examples/openrouter_error_handling.py for usage examples

v0.9.0 🚀

Release Date: October 18, 2024

✨ Highlights

🛡️ Guardrails System

NEW: Comprehensive guardrails integration for safer AI interactions
Tool Leakage Validator: Prevent sensitive tool information from leaking in responses

🔌 OpenRouter Provider

Full OpenRouter integration with streaming support and structured outputs
Dynamic price fetching for cost-aware model selection
Model fallbacks for improved reliability
Factory methods for easier instantiation
Enhanced tool adapter with complex type expansion in JSON Schema conversion
Configurable httpx client timeout for generation tasks

📄 Document Parsing Revolution

Native PDF processing with PyMuPDF for superior performance
RTF Document Parser - brand new format support
Native DOCX processing option for faster document handling
Enhanced PDF parsing with encrypted file support and structured exception handling
PPTX parsing improvements with legacy conversion support
Legacy `.d...

Contributors

lid

Assets 2

Uh oh!

Releases: paragon-intelligence/agentle

v0.9.21

Changelog

v0.9.21

Uh oh!

v0.9.20

Changelog

v0.9.20

Uh oh!

v0.9.19

Changelog

v0.9.19

v0.9.18

v0.9.17

v0.9.16

v0.9.15

v0.9.10

v0.9.9

v0.9.8

v0.9.7

v0.9.6

v0.9.5

v0.9.4

v0.9.3

🚀 New Responses API (Experimental)

🌐 Web Extraction & Automation

🔧 Enhanced Assistant & Tracing

🛠️ Function Call Management

🐛 Bug Fixes & Improvements

📈 Performance & Developer Experience

v0.9.2

v0.9.1

🎯 Pricing Integration

🐛 Bug Fixes

🚨 Error Handling

v0.9.0 🚀

Contributors

Uh oh!

v0.9.18

Changelog

v0.9.18

v0.9.17

v0.9.16

v0.9.15

v0.9.10

v0.9.9

v0.9.8

v0.9.7

v0.9.6

v0.9.5

v0.9.4

v0.9.3

🚀 New Responses API (Experimental)

🌐 Web Extraction & Automation

🔧 Enhanced Assistant & Tracing

🛠️ Function Call Management

🐛 Bug Fixes & Improvements

📈 Performance & Developer Experience

v0.9.2

v0.9.1

🎯 Pricing Integration

🐛 Bug Fixes

🚨 Error Handling

v0.9.0 🚀

✨ Highlights

🛡️ Guardrails System

🔌 OpenRouter Provider

Contributors

Uh oh!

v0.9.17

Changelog

v0.9.17

v0.9.16

v0.9.15

v0.9.10

v0.9.9

v0.9.8

v0.9.7

v0.9.6

📄 **Document Parsing Revolution*...