1.14.3a1
The new release enables agents to save progress and resume tasks, preventing costly restarts.
CrewAI Inc. has pushed a significant pre-release update, version 1.14.3a1, to its open-source framework for orchestrating role-playing AI agents. The headline feature is the addition of checkpoint and fork support for standalone agents. This allows developers to save the exact state of an agent's execution—its memory, context, and progress on a task—and resume it later from that point. This is a game-changer for long-running or complex workflows, preventing the need to restart from scratch if a process is interrupted or needs to be parallelized.
Alongside this core functionality, the release includes several important bug fixes. It ensures the `thought_signature` is preserved during Gemini streaming tool calls, maintaining the structured reasoning output crucial for debugging agent logic. The update also refines the checkpoint Text User Interface (TUI) and correctly emits a `task_started` event when resuming from a forked state. These improvements enhance the developer experience and stability, making the framework more robust for building sophisticated AI crews that can handle real-world, multi-step problems.
- Adds checkpoint/fork support, enabling agents to save state and resume tasks later.
- Fixes a bug to preserve `thought_signature` in Gemini model streaming tool calls.
- Improves the checkpoint TUI and event handling for a smoother developer workflow.
Why It Matters
Enables reliable, long-running AI agent systems for production use, reducing cost and complexity from task failures.