1.12.0a1
The framework decouples from LiteLLM, adds agent skills, and supports Arabic documentation.
CrewAI Inc. has launched a significant pre-release, version 1.12.0a1, for its popular open-source framework for orchestrating multi-agent AI systems. The headline feature is the addition of native, first-class support for OpenAI-compatible API providers, including OpenRouter, DeepSeek, Ollama, vLLM, Cerebras, and Alibaba's Dashscope. This move allows developers to seamlessly connect their AI crews to a much broader ecosystem of language models without relying on intermediary abstraction layers. Alongside this, the release introduces a new 'agent skills' system, enabling more granular and powerful specialization of individual agents within a crew.
Architecturally, this update represents a major refactor, decoupling CrewAI's core internal systems—like token counting and error handling—from the LiteLLM library. This shift aims to improve stability and give the framework more direct control over its operations. For global accessibility, the team has added a complete Modern Standard Arabic translation of all documentation. Other notable improvements include a new CLI command for analyzing code changes to auto-generate docs, fixes for agent memory persistence, and comprehensive resolution of type-checking errors across the codebase to enhance developer experience and code quality.
- Adds native support for 6+ OpenAI-compatible providers (OpenRouter, DeepSeek, Ollama) for flexible model choice
- Introduces a new 'agent skills' system for creating more specialized and capable AI agents
- Decouples core framework from LiteLLM and adds full Arabic documentation for broader stability and accessibility
Why It Matters
This update makes building complex, multi-agent AI applications more flexible, stable, and accessible to a global developer base.