CrewAI 1.15.18
CrewAI 1.15.18 stabilizes conversational flows, expands flow configurations, improves project/deployment tracking, and resolves several Anthropic, messaging, and task-recording bugs.
Why this signal matters
CrewAI release 1.15.18 introduces stable conversational flows along with enhanced documentation and APIs, such as declarative router response formats, custom state shapes, and crew-style LLM configuration support. Tracking improvements include recording deployments with UUIDs, capturing project creation IDs, backfilling project IDs across user commands, and logging whether runs received inputs without storing the inputs directly. The release addresses multiple bug fixes: preserving tool results when final answers are empty, accurately mapping default Claude Sonnet 4.6 context window to 1M, elevating Anthropic default max tokens for large tool calls, rendering message parts as text instead of Python reprs, retaining message roles during Agent.kickoff, correctly flagging task failures instead of successes, and fixing crew memory recording as a string rather than a boolean.
Actionable summary
CrewAI v1.15.18 marks conversational flows as stable, adds support for custom state shapes and router response formats in declarative chat flows, and fixes key runtime bugs including task failure reporting and tool result preservation on empty final answers.
- Agent usefulness
- 62/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade to crewai version 1.15.18.
- Verify conversational flows transitioned from experimental or preview to the stable API surface.
- Review task failure monitoring and dashboards, as failed tasks will now be correctly reported as failures instead of successes.