AI in SourceCraft

The service provides the following AI functionality:

The SourceCraft UI provides detailed statistics on quota usage for neurocredits and code autocompletion. For more information, see Viewing Code Assistant usage statistics in SourceCraft.

AI models

You can select AI models in the chat, Code Assistant plugins and SourceCraft CLI. By default, the service automatically selects a model for the task. To select the model yourself, you need a paid subscription within the organization designated for neurocredit deductions, as well as permission from the organization's Administrator or Owner to use the model.

An organization administrator or owner can configure a set of internal and external models for organization members, regardless of the pricing plan. The setup and selection procedure is described in this guide.

Note

Model selection in VibeCraft will become available later.

AI coworkers

An AI coworker is an AI agent in SourceCraft that you can assign project tasks, such as analyzing code, fixing bugs, implementing new features, writing tests or documentation, or conducting code reviews. An AI coworker handles tasks in the repository context, can run and debug the code, and submits the result as a pull request. You can monitor the progress, answer the AI coworker’s questions, and ask it to refine the result.

Use system coworkers or create your own with the necessary skills and instructions. To get started, assign a coworker as the person responsible for the task, as the reviewer of the pull request, or mention them in a comment. You can also provide a task from the Code Assistant chat or Telegram.

For more information, see AI coworkers in SourceCraft and Working with AI coworkers in SourceCraft.

Code editing in the SourceCraft interface

Use the SourceCraft interface with built-in Code Assistant to edit your project code or add new files and directories. You can introduce changes directly into the repository branch or through a pull request.

The smart autocompletion mode explicitly supports code autocompletion for the following programming languages and frameworks:

  • C++
  • Go
  • Java
  • JavaScript/TypeScript
  • Kotlin
  • Python
  • Scala
  • SQL
  • Swift

Autocompletion may also work for other languages. For languages ​​that are less represented in public repositories, the number and quality of autocompletion prompts will be lower.

Code Assistant supports the following types of suggestions:

  • Inline completion: Text autocompletion as you type. Works for code, texts, and documents.
  • Full-function completion: Autocompletion of entire functions or methods.

When generating a suggestion, the loading icon will appear to the left of the editable code. If generation stops or there is no result, you will see the (no suggestion) icon.

For more information, see Editing code in SourceCraft Code Assistant.

Tip

For full-out coding, there is a Visual Studio Code-based cloud development environment called SourceCraft Spaces accessible via the browser.

Chat mode in the SourceCraft interface

The SourceCraft interface has a built-in chat with Code Assistant you can use to interact with the AI assistant.

The following scenarios are supported:

  • SourceCraft user help, FAQ.
  • Searching through the documentation.
  • Generating code snippets.
  • Explaining code in project files or individual file sections, including code in markdown syntax intended for documentation.
  • Help with debugging and troubleshooting.
  • Generating tests for your code.
  • Generating documentation.
  • Scanning code for potential issues.
  • Suggesting improvements and alternative solutions.
  • Generating charts.
  • Writing commit messages.

The page in the SourceCraft interface you view defines the context for Code Assistant, e.g., repository files, current open file, pull request or issue content, etc.

AI skills are also available in chat mode.

For more information, see Chat with Code Assistant in SourceCraft.

AI code review

SourceCraft is equipped with a built-in system for automated AI code review of pull requests. The AI assistant analyzes code changes, detects potential errors, architectural issues, and style inconsistencies, and provides code quality improvement recommendations.

With AI code review, you can save time on manual reviews, enhance code robustness and readability, and accelerate the change acceptance process.

For more information, see AI code review of a pull request in SourceCraft.

Working with an MCP server

With SourceCraft MCP, you can manage SourceCraft entities though the Code Assistant chat interface, e.g., search for and complete issues, create pull requests, etc.

For more information, see Working with a SourceCraft MCP server.

Generating descriptions for pull requests

With Code Assistant, you can automatically generate descriptions for new pull requests, as well as add and replace descriptions in previously created ones.

Tip

You can set the rules Code Assistant will follow to generate descriptions in the AGENTS.md file at the repository root.

For more information, see Generating a description for a pull request.

AI-powered vulnerability analysis and prioritization

For issues detected in the course of static application security testing, there is triage, i.e, AI-generated tips with risk and criticality assessment, false positive probability analysis, and possible fixes.

For issues detected through scanning of secrets and static application security testing, you can perform triage, i.e., bulk AI-powered alert severity assessment.

Tip

By using this feature you consume neurocredits. Check your balance and usage history on the Code Assistant statistics page.

For more information, see AI-powered vulnerability analysis and prioritization in SourceCraft.

Useful links

Previous
Integration with Yandex Tracker
Next
AI models