Skip to content

v0.9.5

Choose a tag to compare

View all tags
@arthurbrenno arthurbrenno released this 23 Oct 17:28
· 92 commits to master since this release
v0.9.5
91064bb

Changelog

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 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 callback support with semantic type aliases
  • Better table formatting (vertical lists for improved readability)

🔧 Developer Experience

  • Circuit breaker implementation with DuckDB backend
  • Static knowledge from_text class method for easier content creation
  • append_content method for dynamic ParsedFile content addition
  • Provider ID implementation across GenerationProvider subclasses
  • Fallback models parameter for provider resilience
  • Enhanced file validation with MIME type and content checks

🐛 Bug Fixes

WhatsApp

  • Fixed Brazilian phone number formatting after Meta's privacy update
  • Proper handling of Evolution API typed dicts (converted to BaseModels)
  • Fixed from_number extraction from sender
  • Corrected buffer handling for Evolution API fields
  • Improved thinking tag removal logic
  • Enhanced race condition protection and batch timing
  • Rate limit checks now happen before message processing

Agents & Tools

  • Fixed assistant message context handling before tool processing
  • Removed unnecessary cast in _stream_with_tools function
  • Corrected property "text" update logic
  • Fixed duplicate tools in vector stores
  • Resolved duplicated parameter issues

Parsing & Generation

  • Fixed PDF extraction error handling
  • Handled cases where no extraction is returned from PDF provider
  • Updated PyMuPDF import for better compatibility
  • Normalized model names (removed 'google/' prefix)
  • Fixed pickling errors in RAG search by avoiding callable serialization
  • Improved logging calls

⚡ Performance Improvements

  • Single page screenshot per page when images are present (avoiding duplicate OCR)
  • Optimized PDF page processing with PyMuPDF
  • Eliminated redundant temp file creation in archive processing
  • Better markdown parsing performance
  • Fail-fast on DOCX→PDF conversion errors

🔄 Refactoring & Code Quality

  • Converted Evolution API typed dicts to BaseModels for better type safety
  • Renamed 'agent' parameters to 'provider' for consistency
  • Renamed 'chunk_tokens' to 'output_tokens' with new 'input_tokens' property
  • Protocol classes follow single underscore convention
  • Cleaned up whitespace and improved exception handling
  • Streamlined parser class retrieval
  • Made Langfuse a parameter for better flexibility
  • Enhanced CallbackWithContext with token propagation

📚 Documentation & Testing

  • Added test for uppercase file extension handling
  • Testing failover Generation Provider
  • Updated examples with lightweight implementations
  • Comprehensive changelog updates

🙏 Thank You

This release wouldn't be possible without the dedication of our contributors and the valuable feedback from our community. We're committed to making Agentle the most powerful and reliable agentic AI framework.

Upgrade now and experience the future of AI agents!

pip install --upgrade agentle

v0.8.68

  • feat(openrouter) streaming with structured outputs
Assets 2
Loading