Developers increasingly rely on AI coding assistants to speed up routine tasks, generate boilerplate, and catch bugs before they reach production. JetBrains IDEs—IntelliJ IDEA, PyCharm, WebStorm, and the rest—support a growing ecosystem of AI plugins that can be wired directly to large‑language‑model (LLM) APIs such as OpenAI, Anthropic, or self‑hosted endpoints.
This guide walks you through installing an AI coding assistant in any JetBrains product, configuring the LLM API, comparing the experience with VS Code extensions, and extending the assistant to automated code‑review steps in a CI/CD pipeline.
What You’ll Learn
- How to choose the right JetBrains AI plugin for your stack.
- Step‑by‑step installation and API configuration.
- Key differences between JetBrains and VS Code AI extensions.
- How to expose the LLM endpoint to CI/CD for automated reviews.
- Best‑practice tips to avoid common pitfalls.
Choosing the Right AI Plugin
JetBrains Marketplace lists several AI‑enabled plugins. The most widely used as of November 2026 are:
| Plugin | Core Model | Supported IDEs | Learning Curve | Pricing Model | Key Limits |
|---|---|---|---|---|---|
| Tabnine | OpenAI GPT‑4o, proprietary Tabnine model | All JetBrains IDEs | Low – UI‑driven settings | Freemium; paid tier adds unlimited completions (check Tabnine pricing) | Free tier limited to 2 k tokens per day |
| Code‑Whisper | OpenAI Whisper + GPT‑4o (via API) | IntelliJ IDEA, PyCharm, WebStorm | Medium – requires API key setup | Pay‑as‑you‑go via OpenAI usage (see OpenAI pricing) | Rate‑limited by OpenAI account limits |
| JetBrains AI Assistant (beta) | Anthropic Claude 3.5 Sonnet (default) | IntelliJ IDEA, PyCharm, GoLand, Rider | Low – built‑in UI | Free during beta; future pricing TBD (track JetBrains AI page) | Beta limits: 5 k tokens per request |
Select a plugin that matches your budget, token limits, and preferred LLM provider. For organizations that already have an OpenAI contract, Code‑Whisper offers the most transparent usage‑based billing.
Installing the Plugin in JetBrains IDE
Step 1 – Open the Marketplace
File → Settings (or Preferences on macOS) → Plugins → Marketplace
Step 2 – Search and Install
Enter the plugin name (e.g., “Tabnine”) and click **Install**. The IDE will prompt for a restart; accept it.
Step 3 – Verify Installation
After restart, open Help → About to confirm the plugin version appears in the list.
Configuring the LLM API Key
All three plugins require an API key for the underlying LLM. Store the key securely using JetBrains’ built‑in Password Safe rather than plain‑text files.
Tabnine Example
- Navigate to
File → Settings → Tools → Tabnine. - Choose **Custom LLM** and paste your OpenAI API key.
- Optionally set a custom endpoint if you run a self‑hosted OpenAI‑compatible server.
Code‑Whisper Example
<!-- settings.xml snippet for Maven projects -->
<properties>
<whisper.api.key>YOUR_OPENAI_API_KEY</whisper.api.key>
</properties>
Then in the IDE: File → Settings → Tools → Code‑Whisper → API Settings and select “Use Maven properties”.
JetBrains AI Assistant (beta)
The beta reads the key from the environment variable JETBRAINS_AI_API_KEY. Set it in your shell profile:
export JETBRAINS_AI_API_KEY=<YOUR_ANTHROPIC_API_KEY>
JetBrains vs. VS Code AI Extensions
Both ecosystems offer similar capabilities (code completion, doc generation, inline refactoring), but there are practical differences:
- Installation model: JetBrains uses a central Marketplace with signed plugins; VS Code relies on the open Marketplace, where extensions may be unsigned.
- Performance: JetBrains plugins run in the same JVM as the IDE, often resulting in lower latency for large projects.
- Configuration UI: JetBrains provides a dedicated Settings pane; VS Code extensions typically require editing
settings.json. - CI/CD hooks: JetBrains plugins expose a CLI (
jb-ai) that can be invoked in pipelines, whereas VS Code extensions usually need a separate script.
Automating Code Reviews in CI/CD
Once the assistant can generate suggestions locally, you can reuse the same LLM endpoint in a CI pipeline to run automated reviews on pull requests.
Prerequisites
- A CI system that supports Docker containers (GitHub Actions, GitLab CI, Azure Pipelines, etc.).
- An LLM API key stored as a secret variable.
- The
jb-aiCLI (bundled with JetBrains AI Assistant) or a lightweight wrapper script for other plugins.
Sample GitHub Actions Workflow
name: AI Code Review
on:
pull_request:
branches: [ main ]
jobs:
review:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Set up Java (required for jb-ai)
uses: actions/setup-java@v4
with:
distribution: temurin
java-version: '21'
- name: Install jb-ai CLI
run: |
curl -L https://download.jetbrains.com/ai/cli/latest/jb-ai-linux.tar.gz | tar xz
sudo mv jb-ai /usr/local/bin/
- name: Run AI review
env:
JETBRAINS_AI_API_KEY: ${{ secrets.JETBRAINS_AI_API_KEY }}
run: |
jb-ai review \
--repo ${{ github.repository }} \
--pr ${{ github.event.pull_request.number }} \
--output review.md
- name: Post review as comment
uses: peter-evans/create-or-update-comment@v3
with:
token: ${{ secrets.GITHUB_TOKEN }}
issue-number: ${{ github.event.pull_request.number }}
body-path: review.md
The jb-ai review command sends the diff to the LLM, receives a markdown report, and posts it back to the PR. Adjust the command line for Tabnine or Code‑Whisper by swapping the CLI name and flags (see each plugin’s docs).
Common Mistakes & How to Avoid Them
- Missing API key in CI: Always reference the secret variable; hard‑coding the key will cause a build failure and expose credentials.
- Rate‑limit errors: Batch diffs or throttle the workflow using
jobs.and a retry strategy..timeout-minutes - Large repos causing timeouts: Limit the review to changed files with
--files $(git diff --name-only ${{ github.base_ref }} ${{ github.head_ref }}).
Pro Tips
Store the LLM API key in JetBrains Password Safe and also as a CI secret. This dual storage prevents accidental exposure while keeping the local IDE experience seamless.
When using a self‑hosted LLM (e.g., an open‑source model behind an
vLLMserver), set the endpoint in the plugin’s “Custom URL” field and enable TLS verification to avoid man‑in‑the‑middle risks.
Frequently Asked Questions
Can I use the same AI assistant across multiple JetBrains IDEs?
Yes. Plugins installed from the Marketplace are shared across the JetBrains family; a single API key works for all IDEs on the same machine.
Do AI assistants increase memory usage noticeably?
Because the heavy LLM inference runs on remote servers, the IDE only holds lightweight client libraries. Typical memory overhead is 100–200 MB, which is negligible for modern workstations.
Is it possible to restrict the assistant to specific languages?
Most plugins allow language filters in their settings panel. For example, Tabnine’s “Enabled Languages” list lets you deselect unwanted languages.
How do I audit the suggestions for compliance?
Enable the “Log prompts and responses” option in the plugin settings. The logs are stored in ~/.config/JetBrains/ and can be reviewed or shipped to a SIEM.
What happens if the LLM API is unavailable?
The plugin falls back to offline completions (if the vendor provides a local model) or simply disables suggestions until connectivity is restored.
Conclusion
By installing a JetBrains AI plugin, configuring the LLM API key, and wiring the same endpoint into your CI/CD pipelines, you gain a consistent AI coding assistant experience across development and delivery stages. The next steps are:
- Pick the plugin that aligns with your budget and model preference.
- Complete the installation and test a few completions in a sandbox project.
- Add the
jb-ai(or equivalent) step to your CI workflow and monitor token usage. - Iterate on prompt engineering in the CI step to surface the most useful review comments.
With these actions in place, the AI assistant becomes a continuous partner that helps you write, review, and ship code faster.
Official documentation
- JetBrains Marketplace
- Tabnine documentation
- Code‑Whisper docs
- JetBrains Password Safe
- OpenAI API reference
- GitHub Actions documentation



