CrewAI v1.14.7rc1 improves runtime state management and conversation logic

CrewAI, the open-source framework for orchestrating multi-agent AI workflows, has shipped version 1.14.7rc1 as a pre-release. The update introduces several key features aimed at improving runtime control and conversation management. Most notably, developers can now call `reset_runtime_state` to release accumulated bus state, helping prevent memory bloat in long-running agent systems. The release also adds support for handling both custom prompts natively, and decouples conversation logic from the runtime by introducing a `conversational_definition` — allowing more flexible and modular dialogue systems.

Bug fixes address critical stability issues: runtime state is now scoped per run to bound memory growth and isolate concurrent executions, fixing a long-standing challenge in multi-agent environments. Telemetry setup errors on `crewai-login` are resolved, and the `suppress_flow_events` flag now correctly applies to method-execution events. Additionally, flow condition evaluation has been simplified to be stateless per event, and conversation routing now uses one less cycle. Documentation was updated to reflect new OpenTelemetry collector images. Contributors include greysonlalonde, lorenzejay, lucasgomide, and vinibrsl.