crewAI 1.14.7a2 adds conversational flows and chat API
New pre-release lets agents hold natural conversations with turn-based handling.
crewAI, the popular open-source framework for orchestrating autonomous AI agents, has released v1.14.7a2 as a pre-release. This update introduces conversational flow traces and a dedicated chat API, enabling agents to engage in turn-based dialogues. Developers can now use `handle_turn` for structured conversation handling, and the conversational flow documentation has been updated accordingly. The release also makes DSL triggers route-aware decorators, improving modularity in flow definitions.
Under the hood, the LLM event system now surfaces real `finish_reason`, sampling parameters, and `response.id`, giving developers deeper insight into model behavior. LiteLLM cache and reasoning usage sub-counts are flattened in `_usage_to_dict` for cleaner metrics. Additionally, a new documentation section covers NVIDIA Nemotron LLM integration. The locking backend is now overridable in the lock store, and the flow DSL monolith has been split into focused decorator modules. Contributors include @alex-clawd, @gvieira, @lorenzejay, @lucasgomide, @mattatcha, and @vinibrsl.
- Conversational flow traces and chat API with turn-based `handle_turn` support.
- DSL triggers now route-aware decorators; split monolithic flow DSL into modules.
- Exposed `finish_reason`, sampling parameters, and `response.id` in LLM events; added NVIDIA Nemotron LLM documentation.
Why It Matters
Enables natural multi-turn agent conversations and better observability, key for production multi-agent systems.