CrewAI 1.15.29 makes AgentExecutor stable, upgrades eval

CrewAI published 1.15.29 on October 23, 2026, two weeks after 1.15.28. The main changes are a stable AgentExecutor, a more complete crewAI eval command and a batch of memory and tool-loop fixes. A follow-up 1.15.30 arrived a few hours later, on October 24 (Istanbul time), with a single AGENTS.md fix for the evaluation login flow.

Release facts

What changed

• AgentExecutor is stable. The release promotes AgentExecutor out of experimental status.
• crewAI eval runs end to end. crewAI eval can now run the crew itself and return a verdict from one command, keeps the evaluation's full data in the project and can export evaluations as files. An offline routing and guardrail evaluation runner is also added.
• LiteLLM-style arguments. CrewAI now accepts LiteLLM-compatible kwargs.
• New data sources. An Azure DocumentDB vector search integration, a read-only notebook text loader and a standard-library JSONL loader.
• CLI and platform. Organization selection during login, platform application search, and integration app actions run directly through a call: tool.
• Flow tracing. Flows now trace conversation routes and public replies.
• Compaction refactor. Context compaction becomes a BaseLLM capability.

Fixes worth knowing

• Memory privacy. Private memories without a source no longer leak into every recall.
• Repeated tool calls. Agents stop repeating the same tool call.
• Tool budgets. Tool actions honor failure policies and get a fresh usage budget on each kickoff.
• LanceDB. Memory records come back in the right order, and scope and id values are escaped in LanceDB filters.
• Replanning. The task description is restored when replanning fails.

Questions

Is AgentExecutor still experimental in CrewAI? No. The 1.15.29 release notes say it was promoted out of experimental status.

Can crewAI eval run my crew for me? Yes. Per the release notes, crewAI eval can now run the crew itself and produce a verdict from one command, and it can export evaluations as files.

Should I install 1.15.29 or 1.15.30? 1.15.30 is the newer build. Its only listed fix is to AGENTS.md for the evaluation login flow, so the feature set is the one described for 1.15.29.

For more, see the CrewAI tool hub, the earlier CrewAI 1.15.27 launch, the Frameworks category and LangChain vs CrewAI.