crewAI v1.14.7a1 adds native Snowflake Cortex LLM provider
New Snowflake Cortex and Databricks integrations streamline agent orchestration with lazy-loaded imports.
crewAI, the open-source multi-agent orchestration framework, published version 1.14.7a1. This pre-release introduces a native Snowflake Cortex LLM provider, letting agents directly invoke Snowflake’s managed language models without additional adapters. A complementary Snowflake integration guide and a new Databricks integration guide are included, making it easier to set up agent pipelines on those platforms. Additionally, the release adds support for crew trained agents files—a way to save and reuse agent configurations across runs.
On the engineering side, the team improved import speed by lazy-loading docling dependencies (reducing cold-start overhead) and refactored flow.py into separate DSL, definition, and runtime modules for cleaner code. Bug fixes address CLI restoration for UV tool install, unreliable file input, incomplete tool result histories in Snowflake Claude, and stringified tool calls. Contributors include jessemiller, Luzk, lorenzejay, and others. The release is tagged as 1.14.7a1 and is available on GitHub.
- Native Snowflake Cortex LLM provider allows agents to use Snowflake's models directly
- New Databricks and Snowflake integration guides reduce setup friction for enterprise users
- Import speed improved by lazy-loading docling dependencies; flow.py refactored for maintainability
Why It Matters
Enterprise teams can now integrate Snowflake and Databricks LLMs into AI agents with native support, accelerating production-ready multi-agent workflows.