JetBrains AI Assistant

What is JetBrains AI Assistant?

JetBrains AI Assistant is an AI-powered programming plugin provided by JetBrains for development environments such as IntelliJ IDEA, PyCharm, WebStorm, GoLand, PhpStorm, and Rider. It utilizes the code structure, symbols, files, version control data, and execution information available in the IDE to offer developers capabilities for code completion, question answering, explanation, generation, refactoring, testing, and review.

Today, JetBrains AI is no longer just a chat window. AI Chat allows switching between a regular chat interface and various programming agents; it can utilize JetBrains’ own Junie, as well as Codex, Claude Agent, Gemini CLI, and other agents that are compatible with ACP.

AI Chat within IDE

AI Chat can answer questions related to projects, explain errors, generate code and terminal commands, and obtain context from manually added files, directories, symbols, images, or submissions. Regular Chat mainly provides suggestions that developers review before applying them; it does not automatically apply all changes to the project.

Long conversations continuously expand the context and increase data usage. Different tasks should require separate sessions, and sensitive files should not be included in the context arbitrarily.

Code repository context

The AI Assistant makes use of the indexing and program structure provided by the JetBrains IDE to understand classes, functions, references, dependencies, and any current changes. RAG retrieval searches through large projects for relevant files, enabling responses that are more aligned with the codebase than those derived from just the current tab.

The retrieved context may still omit runtime configurations, external services, and business constraints. For complex modifications, it is necessary to clearly identify the modules, interfaces, and behaviors that cannot be changed.

In-line code completion

AI Completion offers single-line or multi-line suggestions as you type. JetBrains AI Free, Pro, and Ultimate all provide unlimited code completion capabilities, both locally and in the cloud; however, the actual level of usability is still influenced by the IDE, version, and account settings.

The completed content must undergo static analysis and testing just like ordinary code; in particular, attention should be paid to the API version, exception handling, and security boundaries.

Next Edit Suggestions

The next step in the editing process suggests the areas that may need to be modified after making the current changes, such as synchronization parameters, imports, data types, and related calls. This feature is currently available for users with AI Pro, Ultimate, and Enterprise subscriptions.

Predictive editing is suitable for mechanical coordination and should not automatically replace architectural decisions; it is still necessary to understand the scope of impacts before making modifications across modules.

Code generation and multi-file editing

Users can ask the AI to implement functions, create classes, or modify multiple files. The editing mode uses project search to locate relevant code and presents the changes in a diff format, making it suitable for small features, refactoring, and repairs.

The objectives of a single change should include acceptance criteria, testing procedures, and the areas that are prohibited from being modified. Before applying any changes, it is necessary to check whether the generated code removes any comments, overwrites manual modifications, or alters the public interfaces.

Code explanation and document generation

The AI Assistant can explain selected code, complex expressions, and project structure; it can also generate comments, Docstrings, and API documentation. It is useful for taking over unfamiliar codebases, adding explanations to existing modules, and learning how to use various frameworks.

The explanation indicates the results inferred by the model from the code, but it does not prove the behavior at runtime. The documentation should be cross-checked against tests and the actual implementation.

Unit test generation

Plugins can generate test drafts around classes, functions, or modifications, and create common inputs, boundary conditions, and mocks. The existing test runner and code analysis tools in the JetBrains IDE can be used to verify the results.

AI tends to follow the path that leads to happiness. The key components should also take into account failure, concurrency, permissions, time, randomness, and external dependencies.

AI reconstruction and error correction

Developers can describe the goals of refactoring in natural language, or request fixes for compilation, validation, and runtime errors. The combination of AI and IDE semantic analysis enables more accurate identification of symbol and type-related issues.

Automatic repair may only eliminate the error without addressing its root cause; after applying it, tests, static analysis, and performance benchmarks should be run again.

Version control and commit information

AI can generate commit messages, summarize one or more commits, explain diffs, and conduct self-reviews of both uncommitted and committed changes. Teams can also set review rules to ensure that the AI pays attention to security, performance, and project standards.

AI review cannot replace the approval of responsible persons and CI. It may fail to detect impacts across different services, or it might mistake style preferences for defects.

Junie Programming Agent

Junie is a multi-step programming agent developed by JetBrains; it can analyze tasks, create plans, modify code, execute commands and tests, and provide updates on progress. Currently, Junie has been integrated into the unified AI Chat, though the standalone plugin still exists as an option for migration.

The Junie support plan mode involves first generating the requirements, technical design, delivery phases, and testing strategies, after which the user can edit and approve them. Agents are capable of executing commands; however, Action Allowlists and manual confirmation are required to control sensitive operations.

Junie debugging mode

In a supported environment, Junie can use the debugger MCP tool to start or attach to programs, set breakpoints, check their operation status, and evaluate expressions; it also provides assistance in identifying problems using natural language.

Debugging involves accessing runtime data. Production credentials, customer data, and write operations should be isolated to prevent the agent from running in an incorrect environment.

Junie CLI

Junie also offers a standalone CLI that can be used in terminals, IDEs, GitHub, GitLab, and CI/CD environments; it enables issue resolution, feature implementation, Pull Request review, and automatic handling of repository events. The official website provides a free tier without the need for a subscription, and it allows the use of one’s own model keys.

The ways in which the free tiers of CLI and JetBrains AI subscriptions are displayed are different, so they cannot be simply added together. For CI operations, it is necessary to set budgets, branch permissions, and release controls.

Codex, Claude, and ACP agents

AI Chat supports Junie, OpenAI Codex, Claude Agent, Gemini CLI, as well as other agents that comply with the Agent Client Protocol. Developers can switch between these agents depending on the task at hand, without having to leave their familiar JetBrains IDE.

The accounts, fees, data processing, and tool permissions related to third-party agents are determined by the respective providers. The JetBrains AI subscription does not automatically cover the costs of all external services.

Agent Skills

Skills Manager and the Skills Repository enable the discovery, installation, and management of Agent Skills from within the IDE, allowing framework specifications, testing procedures, deployment instructions, and team knowledge to be reused by agents. Currently, different agents may provide varying levels of support for skill storage and execution.

Before installing a skill, read the source material and instructions to prevent the skill from causing the agent to read keys or execute commands with high privileges.

Integration of MCP tools

JetBrains AI and Junie can act as MCP clients to connect to databases, documents, browsers, debuggers, and internal tools. MCP enables agents to not only generate code but also to query external systems and verify results.

Each MCP server should employ the principle of least privilege, use separate tokens, and rely on allowlists. Actions such as payment, deletion, deployment, and production-level writes should not be approved automatically by default.

BYOK comes with its own keys

Users can connect to keys provided by third-party services such as Anthropic, OpenAI, Google, and OpenAI-compatible endpoints, in order to use models directly within AI Chat and the supported agents. BYOK does not require a JetBrains AI subscription; however, the costs for making calls are charged by the model providers.

The key should be stored in the IDE’s credential settings, rather than being included in the project. Different third-party models may not support all of the AI Assistant functions, tool calls, and completion capabilities.

Ollama and LM Studio local models

The AI Assistant can be connected to Ollama, LM Studio, and other local OpenAI-compatible services; it is suitable for users who want to keep their code on the local device, test open-source models, or control their cloud costs. It also allows specifying compatible models for certain completion functions.

Standard chat models may not support Fill-in-the-Middle or predictions for the next edit; the performance of local models is also affected by video memory, context, and quantization.

Multi-model selection

The JetBrains AI service offers a variety of models from providers such as OpenAI, Anthropic, and Google, with these models being updated regularly. Users can choose based on factors such as inference performance, speed, context handling, and cost; high-cost models are marked accordingly in the interface.

The list of models changes rapidly; the production process should keep track of the actual models and their versions, and models from older tutorials should not be considered permanently available.

Supported IDEs

The AI Assistant is compatible with the major JetBrains IDEs, and can also be used in environments such as Android Studio and Visual Studio with ReSharper installed. The specific features vary depending on the IDE, programming language, and version; not all functions are available in every product.

Before installation, it is necessary to update to the officially supported version and check the compatibility of the plugin. Older IDEs may not be able to recognize new license tiers such as Free or Ultimate.

AI Trial trial version

New users can take advantage of a 30-day AI trial under eligible paid or free IDE licenses; the individual trial includes 10 AI Credits, while the trial quota for organizations may vary.

The trial version offers features similar to those of AI Pro, but with time and usage limits.

The trial requires a JetBrains Account; eligibility and availability depend on the details on the account page.

Price and version comparison

AI Free free version

For individuals and organizations, AI Free represents the free tier: it includes 3 AI Credits every 30 days, as well as access to the AI Assistant, Junie, Grazie Pro, browser extensions, and unlimited code completion features, both locally and in the cloud.

AI Free requires a valid JetBrains IDE license; it is not available for IntelliJ IDEA Community, PyCharm Community, nor the free version of PyCharm Unified, and it is also not available for Android Studio. Both Free and Ultimate versions are currently subject to geographical restrictions, with an exception for the Chinese mainland.

Price of Personal AI Pro

The Personal AI Pro plan costs $10 per month; it includes 10 AI Credits every 30 days. The annual subscription comes out to approximately $8.33 per month.

It is suitable for everyday conversations, code generation, and occasional use of Junie; it is also included in the paid All Products Pack and dotUltimate subscriptions.

Once the package quota is exhausted, you can purchase a Top-up, or switch to BYOK and local models.

Price of Personal AI Ultimate

Personal AI Ultimate costs $30 per month; it includes 35 AI Credits every 30 days. The annual subscription amounts to approximately $25 per month.

It is designed for more frequent Junie and proxy tasks, offering a bonus of 5 points in addition to the value corresponding to the price.

The agent carries out multiple rounds of retrieval, editing, and testing; the actual time required is usually much higher than that in simple conversations, and Ultimate is not available for unlimited use.

Prices for AI Pro and Ultimate editions

Organizations on AI Pro currently pay 20 dollars per user per month and receive 20 AI Credits, while those on AI Ultimate pay 60 dollars per user per month and get 70 AI Credits. Organizations that migrate to JetBrains Central have their license quotas combined into a shared pool, with the administrator setting default limits as well as individual limits.

Organizational pricing is different from individual pricing; it’s not possible to estimate the cost of group purchases by using the individual prices of $10 or $30.

AI Enterprise

AI Enterprise is offered through JetBrains IDE Services; the current starting price is 60 dollars per user per month. The resource quota provided is equivalent to or higher than that of AI Ultimate, and it includes features such as access management, corporate data sources, BYOK, auditing, and enterprise deployment capabilities.

Companies can evaluate cloud, on-premises, or hybrid deployment options; the final amount and price shall be determined through the sales contract.

AI Credits and Consumption

Each AI Credit corresponds to the value of $1 worth of model resources, but the amount consumed by a request is not fixed; the number of tokens, the length of the context, the price of the model, and the number of proxy steps all influence the charging amount.

The official examples indicate that 1 Credit is sufficient for approximately 10 generations of chat codes, 40 generations of editor code, 25 code interpretations, 280 document creations, or 140 submissions – but these are only estimates.

The monthly quota is reset every 30 days starting from the date of initial activation; any unused subscription quota is not carried over. The widget allows you to view the remaining balance and the date of the next reset.

Top-up AI Credits

Once the subscription quota is exhausted, a top-up can be purchased. The system uses the current quota first and then the amount topped up. A top-up is usually valid for 12 months from the date of purchase;

The top-up amounts purchased by the organization can be shared and restricted by the administrator.

Top-up is currently not available in the Chinese mainland. The region, taxes and fees, as well as the currency, shall be based on the settlement page.

Data and Privacy

When cloud models are used, prompts, necessary code snippets, file types, and framework information are sent to the model provider. JetBrains also offers optional detailed data sharing, which may include complete conversations and code snippets, for product improvement and model training.

This option is disabled by default.

Developers can view AI request logs. Before corporate code is incorporated into any cloud platform, it is necessary to verify the company’s policies, the supplier’s terms, and the location of the data.

.aiignore and disabling AI

Projects can use .aiignore files to exclude certain files, and they are also compatible with the ignore files used by some other AI tools; the .noai file in the root directory can disable the AI Assistant for the entire project, and plugins can be disabled or uninstalled at the IDE level.

These rules apply only to JetBrains AI Assistant and do not necessarily restrict third-party plugins or external agents. Controls should also be implemented at the organizational and network levels for sensitive repositories.

GitHub and the open-source status

JetBrains has made available on GitHub a large number of projects, including IntelliJ Community, Kotlin, the ACP protocol, and the Junie repository. The Junie repository allows users to download the CLI tool, as well as view releases and submit issues; however, its license states that all rights are reserved and that it is governed by JetBrains’ AI terms of service, so it is not a standard open-source license.

JetBrains AI Assistant, cloud services, and Junie core cannot be classified as open source based on this. The open-source IntelliJ platform and ACP protocol do not imply that the source code for AI models or commercial plugins is also available openly.

Tutorial for Using JetBrains AI Assistant

Complete a basic task.

1. Define the code tasks to be carried out in JetBrains AI Assistant, the scope of the repository, and the acceptance criteria.
2. Connect to or import the test project, and first back up the current branch;
3. First, use the AI Chat within the IDE to create a plan, and then review the files that need to be modified.
4. Make minor changes using the code repository context;
5. Run tests, static checks, and builds; unverified code is not accepted directly.
6. Manually check permissions, keys, dependencies, and handle exceptions before merging;

Create reusable professional workflows

1. Select a low-risk, real-world project as a template;
2. Fix the order of use for AI Chat, repository context, and in-line code completion within the IDE;
3. Record the environment, model, prompts, and failure conditions;
4. Set up manual approval for write, deploy, and delete actions;
5. Compare speed, cost, test pass rate, and the amount of rework;
6. Expand to a team or production environment only after stability has been verified;

Which users is it suitable for?

• Developers using IntelliJ IDEA, PyCharm, WebStorm, GoLand, or Rider;
• Individuals who need code completion, explanation, testing, and refactoring;
• Teams that wish to use proxies such as Junie, Codex, and Claude consistently within the IDE;
• Users who are concerned about privacy and costs and therefore need BYOK, Ollama, or LM Studio;
• Enterprises that require access control, auditing, sharing quotas, and on-premises deployment.

Product advantages

• Deep integration with JetBrains’ code indexing, refactoring, debugging, and VCS tools;
• Normal chatting, editing mode, and multi-step proxy are unified;
• It supports the Junie, Codex, Claude, and ACP ecosystems;
• You can choose from the JetBrains cloud model, BYOK, or a local model;
• Offers project ignoring, disabling, permissions, and enterprise governance;
• The All Products Pack and dotUltimate include AI Pro.

Restrictions and Precautions

• Free eligibility is subject to IDE version, license, product, and region restrictions;
• Cloud services are billed on a credits basis; long-running contexts and proxy tasks can quickly deplete these credits, while the use of third-party keys incurs additional costs.
• AI can perform multiple file modifications, as well as command execution and debugging tasks; it is necessary to review any differences, use version control, run tests, and restrict access to production environments.

Frequently Asked Questions

Is JetBrains AI Assistant free?

Users who meet the criteria can use AI Free, which includes 3 credits every 30 days as well as unlimited opportunities to complete tasks; it is not available for the Community version, Android Studio, and in some regions.

How much does JetBrains AI cost?

The individual AI Pro plan costs $10 per month, while the Ultimate plan costs $30 per month; for organizational versions, the costs are $20 and $60 per user per month respectively.

Corporate plans and taxes are subject to the settlement terms or the contract.

Is Junie included in JetBrains AI?

Included. AI Free, Pro, and Ultimate can use Junie and share Credits;

Junie CLI also offers BYOK as well as a separate free entry point.

Can I use my own API key?

Yes, it supports OpenAI, Anthropic, Google, and various compatible endpoints; it is also possible to connect to local models such as Ollama and LM Studio. The costs are borne by the respective providers.

Is Codex and Claude Agent supported?

Supported. Unified AI Chat allows switching between Junie, Codex, Claude Agent, Gemini CLI, and other ACP agents.

Is JetBrains AI Assistant open source?

It is not open-source. JetBrains has made the IDE platform, the ACP protocol, and the Junie distribution repository available publicly, but AI Assistant and the core components of Junie are subject to commercial service terms.