How AutoDev Boosts Coding Efficiency with AI-Powered IDE Integration

Design Goals for AutoDev

AutoDev is built to reduce the manual effort required when using large language models (LLMs) for software development. The core objectives are:

Provide end‑to‑end integration that covers prompt authoring, LLM communication, and insertion of generated code back into the IDE.

Automatically collect the surrounding code, project configuration, and other context needed to generate relevant code snippets.

Keep a human‑in‑the‑loop step for verification, correction, and quality assurance of AI‑generated output.

AutoDev 0.7 New Features

Built‑in Architecture and Code Standards

The plugin can enforce project‑wide conventions by generating code that follows a configurable specification. Users define rules for controllers, services, entities, repositories, and database DDL. When a new component is created, AutoDev applies these rules automatically.

{
  "spec": {
    "controller": "- Use BeanUtils.copyProperties for DTO‑to‑Entity conversion",
    "service": "- Prefer constructor or setter injection; avoid @Autowired",
    "entity": "- Annotate with JPA for database mapping",
    "repository": "- Extend JpaRepository for basic CRUD",
    "ddl": "- Apply NOT NULL constraints to ensure data integrity"
  }
}

Additional configurations can be added to tailor CRUD generation to specific project standards.

Deep Integration with Daily Programming Activities

One‑click generation of commit messages based on detected code changes.

Automatic creation of a CHANGELOG from a selected range of commits.

Instant error‑message analysis: the selected stack trace and surrounding code are sent to an LLM for diagnostic suggestions.

Automated generation of unit tests for newly created or modified code.

Multi‑Language AI Assistance

The architecture, originally focused on Java CRUD generation, has been refactored into a layered, extensible model (inspired by the IntelliJ Rust plugin). Extension points allow language‑specific data structures, prompts, and CRUD workflows, making it straightforward to add support for Kotlin, Rust, TypeScript, and other languages. Currently, Java and Kotlin have the most mature implementations.

Broader LLM Support

OpenAI API and compatible proxy endpoints (the most extensively tested configuration).

Azure OpenAI for compliant usage in regions where direct OpenAI access is restricted.

Placeholder interfaces for additional commercial or open‑source LLM APIs; users can configure custom endpoints.

Smarter Prompt Strategies

AutoDev adopts a richer prompt composition inspired by analyses of GitHub Copilot and JetBrains AI Assistant. The prompt includes the following elements where applicable: BeforeCursor – code preceding the generation point. AfterCursor – code following the generation point. SimilarFile – snippets from files with similar functionality. ImportedFile – content from imported modules. LanguageMarker – explicit language identifier. PathMarker – file path context. RetrievalSnippet – relevant UML or model excerpts.

Language‑specific adaptations:

Java + CRUD : gathers surrounding code, method calls, and UML snippets to provide a detailed context.

Other Java modes : uses a DtModel representation to embed class UML comments.

Python : extracts import statements and locates similar code fragments to build the prompt.

If the total token count exceeds the model’s limit, the strategy dynamically trims less critical context while preserving essential elements.

Repository

Source code, issue tracking, and community discussion are available at:

https://github.com/unit-mesh/auto-dev