Unlocking AutoDev’s AI Agent: Custom Actions to Supercharge Your IDE

Background

General large language models lack internal domain knowledge; Retrieval‑Augmented Generation (RAG) can extend them but is unsuitable for constantly changing data inside an IDE. Therefore a server‑side AI Agent is needed, exposing APIs that the IDE can call.

Custom AI Agent Design

AutoDev adds a customizable AI Agent that can be configured with different response actions. Three response types are demonstrated.

1. Direct Result – Internal API Integration

Typical scenarios: knowledge the model does not know, internal domain knowledge, API specifications.

{
  "name": "内部 API 集成",
  "url": "http://127.0.0.1:8765/api/agent/api-market",
  "responseAction": "Direct"
}

Using Direct returns the raw response from the specified endpoint.

2. TextChunk Result – Component Library Query

Typical scenario: knowledge retrieval where a list of components is needed for further coding.

{
  "name": "组件库查询",
  "url": "http://127.0.0.1:8765/api/agent/component-list",
  "responseAction": "TextChunk"
}

Using TextChunk returns a chunk of retrieved text that can be used in subsequent dialogue.

3. WebView Result – Low‑Code Page Generation

Typical scenarios: front‑end page generation or low‑code UI creation where the AI returns a full page URL.

{
  "name": "页面生成",
  "url": "http://127.0.0.1:8765/api/agent/ux",
  "auth": {
    "type": "Bearer",
    "token": "eyJhbGci"
  },
  "responseAction": "WebView"
}

Using WebView returns a URL that the IDE can display inside a web view component.

Conclusion

In the IDE, AI Agents act as modular plugins that accelerate development by delegating specific tasks to external services. The three response actions—Direct, TextChunk, and WebView—cover common use cases such as internal API calls, knowledge retrieval, and low‑code UI generation.

For JetBrains users awaiting marketplace approval, the plugin can be downloaded directly from GitHub: https://github.com/unit-mesh/auto-dev/releases.