Integration of Custom AI Models in JetBrains IDEs with AI Enterprise – Support | JetBrains

Question

Can a custom AI model be integrated for inline editing and code completions in JetBrains IDEs without using the JetBrains AI subscription, specifically within the context of AI Enterprise?

Answer

In the current JetBrains AI architecture, this depends on whether the setup is managed individually in the IDE or centrally through AI Enterprise:

For Individual IDE Instances (Non-AI Enterprise):
- A custom or local AI model can be configured directly in supported JetBrains IDEs. This is done through the IDE settings under Settings | Tools | AI Assistant | Providers & API Keys. This allows users to connect a custom AI provider and use it for supported AI Assistant features, depending on the IDE version and available provider capabilities.
- This method does not require an AI Enterprise subscription and remains independent of JetBrains’ centralized AI services.
For AI Enterprise:
- AI Enterprise does not currently allow replacing the JetBrains-provided code completion model with a custom third-party model.
- For code completion, AI Enterprise uses JetBrains Mellum, which is optimized by JetBrains for software development workflows. Mellum may be used as a JetBrains-hosted service or, where applicable, as a self-hosted deployment in the customer’s infrastructure, but it cannot currently be replaced with another model for AI Enterprise-managed code completion.
- AI Enterprise focuses on centralizing the management and provisioning of AI tools across an organization and does not currently support plugging in custom models for code completion through its infrastructure.

Summary

When using individually configured IDE instances, a custom AI provider may be configured directly in the IDE.

When using AI Enterprise, code completion is handled by JetBrains Mellum and cannot currently be redirected to a custom model.

Therefore, if the main requirement is to use a customer-owned code-completion model, the available approach is to configure it directly in each IDE instance rather than through AI Enterprise.

References