- Code assistance in the IDE with completion, generation, chat, and diff transformations.
- Advanced Context: Remote Repositories with @ and Customization in Enterprise.
- Editions: Individual (free), Standard and Enterprise with cloud security and extensions.
- What's new: Agent Mode in VS Code/IntelliJ and improvements to Gemini CLI and IDE.
Gemini Code Assist has become one of the most talked-about programming assistants on Google Cloud, with versions for individual users and businesses. Although it's still evolving, it already allows you to complete, generate, transform, and review code directly from your IDE , with contextual chat and smart actions that reduce the effort required for repetitive tasks.
Beyond classic autocomplete features, the service includes code transformation with difference view, remote repository context, and an agent mode for step-by-step problem-solving. And for those who want to get started at no cost, there's an individual edition with very generous usage limits in Visual Studio Code and JetBrains IDEs.
What is Gemini Code Assist and who is it for?
Gemini Code Assist is an AI-powered collaborator integrated into the editor, designed to help developers of all levels work faster without losing control over their code . It builds upon the legacy "Duet AI for Developers" and expands its capabilities to perform code generation, explanation, refactoring, and documentation tasks in real time.
There are three main plans: the Individual edition (free), the Standard edition, and the Enterprise edition. The Individual edition lets you start without a credit card and offers high limits: up to 6.000 autocompletes per day, 240 daily chat interactions, and code reviews . In addition, the chat supports a large context window of up to 128.000 input tokens, enough to handle large files and accommodate a significant portion of your local database.
For organizations, Standard and Enterprise focus on security, data governance, and extensions in Google Cloud . Enterprise adds code customization to tailor suggestions to private repositories, as well as more enterprise cloud integrations and options like VPC-SC or Private Access to Google.
In large-scale scenarios, some capabilities of the Gemini ecosystem have demonstrated context windows of up to one million tokens in certain flows , useful for understanding large projects or extensive documentation. In any case, it's always advisable to verify the validity of each response: the tool itself warns that it can generate plausible but incorrect outputs.
Main functions in the IDE
As you type, Gemini Code Assist suggests inline completions that you can instantly accept or ignore . This behavior, known as code completion, speeds up repetitive tasks and helps maintain style consistency. If you prefer, you can disable autocomplete in the plugin settings, and if the default shortcut for generating code isn't working for you, you can remap the key combination in your editor.
Another key element is code transformation. Using commands or natural language in menus like "Quick Pick," you can request specific modifications to files and review a diff before applying changes . This approach minimizes errors, as the flow shows you the proposed adjustments and allows you to approve or reject each block.
The integrated chat works as a copilot with context from the editor. You can select a snippet and request assistance: from "Function to create a Cloud Storage bucket" to "Help me debug my code." or "Write unit tests for my code.”. The tool processes your request, cites sources where appropriate and submits commented proposals, with the possibility of inserting the result into the active file.
When an error is detected, the wizard offers a quick fix with code transformation . You'll see a diff with the suggested solution and have full control to apply, edit, or discard it. This pattern reduces back-and-forth between the IDE and the documentation, and cuts down on debugging time.
Important note: If you use other plugins that capture the same shortcuts or IDE APIs, build behavior, autocomplete, and transformations may be non-deterministic . If you notice inconsistencies, check for shortcut conflicts and the extension settings in your environment.
Advanced Context: Remote Repositories and Customization
If your work revolves around specific microservices, libraries, or modules, you can ask Gemini to focus on specific remote repositories. To do this, you must first index and configure the repositories in code customization (required for Enterprise) and then invoke them in the chat using "@".
The flow is simple: you open the IDE chat, start the instruction with “@”, you select one or more repositories from the list and finish writing the request. With that, the system prioritizes the context of the chosen repos and generates more relevant responses. It's especially useful for aligning with your team's own patterns, conventions, and practices.
Some useful queries with remote repositories are: understand the repo structure ( "@REPOSITORY_NAME What is the overall structure of this repository?”), ask for a refactor towards the project conventions (“@REPOSITORY_NAME Refactor the following code to follow the conventions in the selected repository.") either generate tests adapted to existing examples ( "@UNIT_TEST_FILE_NAME Generate unit tests for module-x based on the examples in the selected file.
You can also orchestrate specific actions on built-in tools with the “@” symbol followed by the name, for example: “@GitHub get issues assigned to me”. And, if you need to fine-tune things even further, the model supports style requests “Use the library-x in @REPOSITORY_A_NAME-A and implement the function-x" for rely on specific components of the ecosystem that your organization already uses.
To protect the privacy of the local context, Gemini Code Assist respects the “.aiexclude"And".gitignore”, preventing its content from being used in autocomplete, generation, transformations, or chat. Additionally, you can disable suggestions that match cited sources, so that the wizard doesn't suggest snippets that reproduce external code with added quotes.
Editions: Individual, Standard and Enterprise
The individual edition is designed for those who want to get started right away, with no costs or friction: it requires no credit card and allows you to explore key features in VS Code and JetBrains with generous limits. Google highlights that it offers up to 90 times more monthly completions than other popular free assistants, and chat with up to 128.000 input tokens for understanding long files and more local context.
In addition to completion and generation, the chat helps you document features, write comments, or create tests from requirements, with support for natural language in multiple languages . Everyday examples include “creating an HTML form with name, email, and message,” “writing a script that sends a daily email based on the weather,” or “explaining a Python snippet and detecting errors.”
In the Standard edition, the focus shifts to secure operation at scale, with robust data governance, secure infrastructure, and code hint compensation . It includes chat support, simplified IDE integration, local codebase awareness, and extensions for Google Cloud such as Firebase, Colab Enterprise, BigQuery (data insights), Cloud Run, and Database Studio.
Enterprise adds all of the above, plus customization with private codebases (GitHub, GitLab, Bitbucket) to accelerate development according to internal practices. It also expands integrations with Google Cloud (Apigee, Application Integration, Gemini Cloud Assist) and makes it easier for cloud teams to design, operate, and optimize applications and infrastructure more effectively.
In both paid editions, there are compliance capabilities such as source citations in the IDE and console , network controls (VPC-SC, Private Google Access) and a catalog of features by area (code and chat, CLI, smart actions, external tools) that can be mapped to the specific needs of each team.
Integrations and capabilities outside the IDE in Google Cloud
The scope of Gemini Code Assist extends beyond the editor. With Gemini Cloud Assist in the Google Cloud console, teams gain generative assistance with operations and configuration, with a feature set available to all users and expanded for those using Enterprise.
In API development, Gemini in Apigee provides in-ide and console assistance, accelerating design and debugging tasks. In Application Integration , AI guides the composition of integrations from the console, reducing deployment times and configuration errors.
In BigQuery Studio, Data Insights offers a library of queries generated from your table metadata, which are very useful for quickly exploring datasets and finding patterns. This contextual help streamlines the creation of dashboards and ad hoc analyses.
In databases, the tool can generate SQL statements in natural language , adapt queries to the actual schema, and optimize or explain existing queries. This democratizes data access and avoids friction between technical and business profiles.
What's New: Agent Mode, IDE and Gemini CLI Improvements
One of the major new features is Agent Mode , an agentic approach that solves complex, multi-step tasks with a plan you review before touching the code. For example, "refactoring the model, view, and controller" of a shopping cart to display applied discount coupons would involve multiple files; the agent proposes a detailed plan , waits for your approval, and executes changes in a controlled manner.
In VS Code, Agent Mode was initially released to the Insiders channel and is now in preview for everyone . It goes beyond command-response chat thanks to its integration with the Gemini CLI: it supports multiple file edits, full project context, built-in tools, and MCP support (with a human in the loop for review).
Recent improvements to Agent Mode in VS Code include: visual editing with integrated inline diffs , in-chat diffs to clarify proposals, persistence of Agent Mode in the history, clearer quota updates , real-time shell command output, batch approvals of tool calls, and a faster UI. Even the Agent Mode toggle state persists across IDE restarts.
In IntelliJ, Agent Mode is now available in the stable version . You can review everything before applying changes, request plan modifications, approve or reject steps, and activate an auto-approval mode to have the agent act on your behalf. At the end, you can review and revert any changes you deem necessary.
In addition, the team has refined the overall IDE experience: you can now pause ongoing chat responses to cut short long or rambling conversations, and "thinking tokens" have been added to IntelliJ to provide reasoning cues before the final response. The performance and quality of autocomplete suggestions in VS Code have also been improved , and several features that were in preview have been made generally available.
In terms of productivity, Gemini CLI has grown significantly. Gemini CLI GitHub Actions was launched as a free repository companion, capable of autonomously handling routine tasks or collaborating on demand. Key uses include intelligent issue triage , accelerated pull request reviews, and on-demand collaboration.
The CLI also integrates more deeply with VS Code , becoming aware of open files and selected text to provide more relevant suggestions. With native diffs in the editor , you can see proposed changes side-by-side and edit them before accepting. To use it, install version 0.1.20 or later and run an initial setup command.
Other improvements include Custom Slash Commands for defining reusable prompts and a public roadmap for CLI v1. The goal is clear: to offer a more consistent, secure, and private development workflow, with frequent, community-driven iterations.
Basic setup and interaction
Getting started is easy: after installing the Gemini Code Assist extension in VS Code or JetBrains IDEs , you can activate the chat from the assistant icon and request help in two main ways: receive completed code or generate code directly in the editor , and chat in the sidebar with your questions or by selecting code to provide context.
Some prompts that work well on a daily basis: “Write unit tests for my code.","Make my code more readable." or "Help me debug my code.”. If you work with indexed remote repositories, remember the “@” prefix to orient context to one or more specific sources, and filter the list by typing the name of the repo.
For teams on Google Cloud, you can also explore Gemini Code Assist at Cloud Shell Editor, Cloud Workstations and Android Studio. In all cases, please refer to the detailed configuration guide if you need to integrate Single Sign-On, restrict access by perimeters (VPC-SC) or define exclusion rules with “.aiexclude".
If you want to prevent the wizard from suggesting snippets that match code from cited sources (for example, open source), there's an option to disable those suggestions. And if you encounter unusual behavior, keep in mind that there might be conflicts with other plugins that use the same shortcuts or IDE APIs.
Good practices, limits and responsibility
As an early-stage technology, Gemini Code Assist may generate plausible but incorrect outputs. Therefore, it's advisable to validate and test any changes before integrating them into production. The tool includes a citation system when reusing lengthy passages from other sources and provides documentation on how and when to cite.
In terms of security and compliance, the Standard and Enterprise editions rely on secure infrastructure, data governance, and network controls , along with IP indemnification in suggestions. Enterprise also enables customization with private codebases , which not only improves the relevance of responses but also helps reflect the organization's internal standards.
To preserve privacy, use “.aiexclude"And".gitignore” to exclude sensitive files from the local context used by the wizard. And if you detect a false positive in completions or transformations, You can always check the diff and not apply the proposed changes.
The individual edition is free and offers generous daily limits: 6.000 completed tasks , 240 chats, and code reviews. This is perfect for students, freelancers, or developers who want to evaluate the tool without worrying about early limits, with a chat allowance of up to 128.000 tokens to cover large files.
Finally, the team encourages users to submit feedback directly from the IDE to refine the experience. There are also sections dedicated to known issues and release notes, where updates such as Agent Mode, performance improvements, and new features in the CLI and its integrations are published.
With this full suite of features—including code generation and transformation, local and remote contextual chat, enterprise customization, and Google Cloud integrations —Gemini Code Assist is shaping up to be a powerful companion for programming with less friction, maintaining control over every change, and scaling from a personal project to a complex organization.
