v2.0.0 - Modernized Architecture & Model Improvements

Major Improvements

Code Modernization

• Native OpenAI Format: Refactored history storage to use native OpenAI/OpenRouter message format (removed custom paraphrasing layer)
• Simplified LLM Calling: Added call_llm_simple() for tool-free tasks, removing unnecessary complexity for summarization
• Cleaner Codebase: Removed legacy workarounds and jank previously needed for older models

Model Updates

• Smart Fallback Chain: --alt now tries multiple models automatically (polaris-alpha → grok-code-fast-1 → kimi-k2-thinking → gpt-5-mini)
• Default Model: gpt-5-mini for reliability and performance
• Thinking Budget Support: Added thinking_budget parameter support via extra_body_params

New Features

• Tavily Web Search: Integrated web search and URL extraction tools (optional with TAVILY_API_KEY)
• Better Tool Calling: Improved visual presentation and tool execution feedback
• History Management: Enhanced --compact and --summary with better model support (now works with phi-4)

Developer Experience

• Test Harnesses: Added bash test scripts for validation
• Better Error Handling: Improved retry logic and fallback mechanisms
• API Key Status: Shows configuration status in --help output

Breaking Changes

None - fully backward compatible

Files

• ask - Main executable (Python 3.11+)
• README.md - Documentation
• requirements.txt - Dependencies (openai, tavily-python)

Commits Since v1.0

• Refactored to native OpenAI message format
• Added call_llm_simple for non-tool tasks
• Thinking budget optimization
• Tavily web search integration
• Model fallback chains
• Removed Groq dependency
• Enhanced history management

See full changelog: f78837f...56afa16

initial

1.0.1

Update README.md

initial

releases