v0.1.4

What's New

Real LLM Provider Support in Cookbooks

• All cookbooks now use real OpenAI/Anthropic API calls instead of mock providers
• Auto-detection pattern: uses Anthropic if ANTHROPIC_API_KEY is set, falls back to OpenAI

Bug Fixes

• Fix tool call argument serialization in ReAct agentic loop (OpenAI API requires JSON strings, not dicts)
• Fix CI test for tool argument format
• Fix OpenAI timeout test to work in CI without openai installed
• Add httpx to dev dependencies for FastAPI TestClient

Security & Hardening (since v0.1.3)

• Configurable CORS, API key auth, and sanitized errors for FlowServer
• SSRF and path traversal vulnerability fixes in built-in tasks
• Max size limits and LRU eviction for in-memory storage/caches
• Sandbox builtins restriction and import blocking
• Thread safety with asyncio.Lock across storage, cache, and event systems
• Configurable timeouts for LLM providers and storage backends
• Constructor validation for Task, Scheduler, FallbackChain

New Features (since v0.1.3)

• Flow debugger with step-through, breakpoints, and time-travel
• Agent benchmarking suite with tool use and instruction benchmarks
• Multi-turn conversation management with state tracking
• Structured output with JSON mode and Pydantic validation
• Comprehensive test suite for engine, flow, storage, cache, and built-in tasks