JetBrains AI Assistant and Junie: What Each Does

JetBrains AI Assistant is the AI feature set inside IntelliJ IDEA, PyCharm, WebStorm and the other JetBrains IDEs. It gives you a chat that knows your project, code completion and next edit suggestions, inline generation, commit messages and test drafts, and a place to run coding agents. Junie is one of those agents: JetBrains’ own, built to plan and carry out multi-step changes across files. So the two are not rivals. AI Assistant is the host, and Junie is the agent you hand a larger job to, alongside Claude Agent, Codex and GitHub Copilot, which AI Assistant can also run. Both use models from Anthropic, OpenAI, Google and xAI through a JetBrains AI subscription, and both can connect to MCP servers. Everything below is as of October 1, 2026; JetBrains is renaming and adding products quickly, so check the dates on the pages linked.

AI Assistant and Junie at a glance

• AI Assistant: built into the IDE. Chat, completion, next edit suggestions, inline prompts, explanations, refactoring help, documentation, unit tests, commit messages and pull request summaries. You stay in charge of every edit.
• Junie: a coding agent. You give it a task, it plans, edits several files, runs tests and reports back. It runs in AI Assistant’s chat, as its own IDE plugin, as a CLI, and from GitHub or GitLab.
• Shared: one JetBrains AI subscription covers both, and both can call tools on MCP servers you configure.
• Different: AI Assistant works at the size of a question or an edit; Junie works at the size of a task.

What JetBrains AI Assistant does

The AI Assistant documentation↗ (opens in a new tab) groups its features into five areas, and they map well onto a working day:

• Context-aware chat: ask about your code and your project structure, and get answers grounded in the files you have open and the project around them.
• Coding agents: hand off “complex, multi-step tasks” to an agent that can work across many files.
• In-editor help: generate or change code from a natural-language prompt, and accept inline code completion and next edit suggestions as you type.
• Code insight: explain a function, suggest improvements, help with a refactoring, or point out likely problems.
• Routine work: documentation, unit tests, commit messages and pull request summaries.

It works in ten JetBrains IDEs plus Android Studio and ReSharper. Project rules, a prompt library, MCP and the Agent Client Protocol (ACP) are available in all of them; a few language-specific features, such as Python type annotations, are limited to PyCharm.

What Junie does

Junie is the part of JetBrains AI that does the work rather than suggests it. You describe a change, and it explores the project, proposes a plan, edits the files, runs the tests and tells you what it did, asking before terminal commands and other risky steps. Its advantage inside a JetBrains IDE is that it uses what the IDE already knows: indexes, inspections, refactorings and the test runner. Junie now has its own site and documentation at junie.jetbrains.com, and it runs in several places: inside AI Assistant’s chat, as a separate IDE plugin, as a terminal CLI, and from GitHub issues and pull requests or GitLab CI/CD.

Its modes, approvals, AGENTS.md guidelines and how it compares with Anthropic’s agent are covered in JetBrains Junie vs Claude Code; this page does not repeat them.

The other agents inside AI Assistant

AI Assistant is no longer only a JetBrains product with JetBrains models. Its activation options↗ (opens in a new tab) list four ways in: a JetBrains AI subscription; integrated agents (Junie, Claude Agent, Codex and GitHub Copilot), each signed in through JetBrains AI, the provider’s own account, or an API key; a third-party model provider with your own key; and any external agent that speaks ACP. Which one you pick decides who bills you for the usage.

If you already use Claude Code, there is also the separate route of running its CLI with Anthropic’s JetBrains plugin; Claude Code in JetBrains IDEs covers that setup.

Two newer names sit next to AI Assistant. JetBrains Air is a workspace for directing several agents at once inside the IDE, offered as a plugin that its Marketplace page calls a public preview. And in a September 22, 2026 post, Introducing JetBrains Air↗ (opens in a new tab), JetBrains renamed JetBrains Central, its governance and cost-control layer for organizations, to Air Governance. The main JetBrains AI page still said “JetBrains Central” on October 1, so expect to see both names for a while.

Which models JetBrains AI runs

Through a JetBrains AI subscription, AI Assistant and Junie run models from Anthropic, OpenAI, Google and xAI. Code completion and next edit suggestions rely on JetBrains’ own models, such as Mellum, which JetBrains describes as optimized for coding. Beyond the subscription you have two other routes: bring your own key for a supported provider, or run local models through Ollama or LM Studio, so AI features work without a cloud service.

One thing to watch: JetBrains’ own pages disagree on the current lineup. The supported models page↗ (opens in a new tab) in the AI Assistant 2026.2 documentation lists, for example, Claude Opus 5 and GPT 5.6 Sol, while Junie’s home page lists Claude Opus 5.5, Claude Fable 5.1 and GPT-6.1 SOL. Treat the model picker in your own IDE as the answer, and update the IDE and plugins if a model you expect is missing.

MCP in AI Assistant and in Junie

Both act as MCP clients, but they keep their server lists in different places. In AI Assistant, open Settings, then Tools, then AI Assistant, then Model Context Protocol (MCP). The AI Assistant MCP page↗ (opens in a new tab) supports three transports: STDIO, where the IDE starts the server as a subprocess; Streamable HTTP; and SSE, kept for older servers. Each server can be global or limited to the current project, and its tools are either called automatically when the chat needs them or by you, with a / command.

Junie keeps its own list in mcp.json: .junie/mcp/mcp.json in the project root for one project, or ~/.junie/mcp/mcp.json for every project. In the CLI, the /mcp command opens an assistant that adds servers from a registry or from scratch, and according to Junie’s MCP configuration page↗ (opens in a new tab), remote servers that need OAuth show an “Authorization required” status until you sign in.

The traffic also runs the other way. Since version 2025.2, JetBrains IDEs include an MCP server of their own, so Claude Code, Codex, VS Code and other clients can call the IDE’s tools. Turn it on under Settings, Tools, MCP Server, and use Auto-Configure to write the entry into a detected client’s configuration.

{
  "mcpServers": {
    "fenbs": {
      "command": "npx",
      "args": ["-y", "fenbs-mcp"],
      "env": { "FENBS_TOKEN": "your-token" }
    }
  }
}

JetBrains AI plans by name

JetBrains sells AI as a separate license with four tiers: AI Free, AI Pro, AI Ultimate and AI Enterprise. AI Free, AI Pro and AI Ultimate are available to individuals and organizations; AI Enterprise is for organizations only. There is also an AI Trial. The tiers differ mainly in how many AI Credits you get each 30 days, the unit JetBrains uses to measure cloud usage, and AI Pro and AI Ultimate can be topped up when you run out. JetBrains says the All Products Pack includes AI Pro, and Junie’s site labels AI Ultimate “Recommended for Junie”. Prices change and vary by region, so check JetBrains’ own plan page rather than a summary.

Which one to open first

• You want a quick answer, an explanation or a small edit you will review line by line: AI Assistant’s chat or an inline prompt.
• You are typing and want the next lines filled in: completion and next edit suggestions, which run on JetBrains’ coding models.
• You have a task with several files and a test to pass: Junie, from the chat or its own plugin, with the plan reviewed before it edits.
• You already pay for Claude, ChatGPT or Copilot: sign in to that agent inside AI Assistant instead of buying a second subscription.
• You want an agent in CI or on a server: the Junie CLI, or Junie on GitHub or GitLab.
• You run several agents in parallel: try JetBrains Air, and expect preview rough edges.

Keeping track of what the agents did

An agent’s chat history is not a record anyone else can read. fenbs is a task board the agents connect to over MCP at https://fenbs.ai/api/mcp, with OAuth sign-in or a token you issue by hand under Settings, with a name, scopes and an optional expiry. Clients that only speak STDIO, which includes a plain mcpServers entry like the one above, use the bridge npx -y fenbs-mcp. Junie, Claude Agent or Codex can then read the next task, move it from To Do through Next Up and In Progress to Completed, write how it was tested on the task, and History records each change under the assistant’s name. Rules you set on the Decisions and rules page are read first by every connected assistant. fenbs is deliberately small: no sprints, due dates or settable assignee.