Configuring AI in IntelliJ IDEA: Separate Completion and Agent Roles for Real Productivity Gains

The author shares a practical framework for configuring multiple AI plugins in IntelliJ IDEA, recommending one primary tool for inline code completion and a separate agent-capable tool for cross-file tasks, with advice on evaluating local vs. cloud models, verifying offline claims, and measuring real productivity gains through task-level tracking.

Java Architect Essentials
Java Architect Essentials
Java Architect Essentials
Configuring AI in IntelliJ IDEA: Separate Completion and Agent Roles for Real Productivity Gains

Problem: Overlapping AI Plugins Hinder Workflow

Installing several AI plugins in IDEA leads to clashing inline suggestions, and complex multi-file changes still require manual stitching. The author proposes a clear division of labor: keep one tool as the daily completion workhorse, and add a second tool with agent capabilities for cross-file tasks that can run commands.

Match Tools to Task Types

Listed tools include JetBrains AI Assistant, GitHub Copilot, Qoder CN (formerly 通义灵码), Continue (local), and Claude Code. The key is to decide per task: do you need a few lines completed, an explanation of a call chain, or a sequence of file edits plus command execution? This is more actionable than comparing "which understands the whole project best."

JetBrains AI Assistant: Cloud vs. Local Completion

AI Assistant works well for context-aware questions, generation, and edits within the IDE. For multi-file changes, use the diff view to review each chunk. Mellum powers cloud code completion by default; this does not make it inherently offline. IDEA's Full Line Completion can run locally; when a local model is required, verify the current version's local settings, supported languages, and licensing terms. Cloud chat and agent quotas are separate and should not be inferred from outdated screenshots showing free tiers.

IDEA与AI的补全和Agent分工示意
IDEA与AI的补全和Agent分工示意
JetBrains AI Assistant主题示意图
JetBrains AI Assistant主题示意图
代码补全与后续编辑建议示意
代码补全与后续编辑建议示意
JetBrains AI Assistant的历史对话界面
JetBrains AI Assistant的历史对话界面

Copilot and Qoder CN: Evaluate Current Capabilities

Copilot now supports JetBrains' Agent mode, enabling cross-project file edits and command invocation; it should no longer be categorized as single-file completion only. Actual performance depends on task, context, and model. The screenshot showing Copilot's next-edit suggestions comes from VS Code; IDEA's exact entry points depend on the installed plugin. Completion and agent features may follow different quota rules — check the current subscription plan before committing.

Copilot后续编辑建议示例,界面来自VS Code
Copilot后续编辑建议示例,界面来自VS Code

Qoder CN (formerly 通义灵码) still provides a JetBrains plugin covering completion, Q&A, multi-file editing, and agent features. Compare integration convenience; whether source code and data can leave the premises remains a company policy decision.

通义灵码的Java异常分析示例
通义灵码的Java异常分析示例

Continue and Claude Code for Specialized Needs

Continue + Ollama addresses on-premise requirements. Model tags must match the locally pulled names; chat and completion roles are configured separately. Also verify embedding models, telemetry, external tools, and network access — pointing only the chat model to a local endpoint does not automatically make the entire pipeline offline. Response times vary with model and hardware; do not copy benchmarks blindly.

Claude Code excels at chaining code reading, planning, editing, and testing. The official JetBrains plugin connects to the terminal CLI; both plugin and CLI must be installed. It supports paid Claude subscriptions or Console access, not exclusively token-based billing. Refer to the official JetBrains documentation for installation steps; other integrated Claude entry points in the IDE require separate authentication and billing confirmation.

Claude Code插件的历史安装界面
Claude Code插件的历史安装界面

Configure the Combination, Then Assign Tasks

Enable only one tool's inline completion first. If suggestions flicker or lag, troubleshoot plugins, network, and indexing step by step — do not immediately blame plugin conflicts.

For complex tasks, define a small, verifiable scope upfront. Example:

任务:为OrderService补超时处理
范围:只改服务实现及对应测试,不升级依赖
前提:先阅读现有事务与异常约定
交付:代码差异、实际执行的测试、未覆盖情况

Even with a scoped prompt, review the output. When the tool has command permissions, confirm which credentials it holds and which directories it can access.

日常补全和复杂任务分工及验收流程
日常补全和复杂任务分工及验收流程

Decide Retention by Recording Task Metrics

Try daily completion plus Claude Code. If an existing Copilot or Qoder CN workflow runs smoothly, there is no need to duplicate subscriptions for the sake of a new combination. Pick comparable small tasks and log time spent, rework count, cost, and failure rate. Inspect added dependencies, transaction boundaries, exception branches, then run tests. Keep the setup if daily completion stays unobtrusive and complex changes remain understandable.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

AI agentscode completionContinueIntelliJ IDEAGitHub CopilotAI coding assistantsClaude CodeJetBrains AI Assistant
Java Architect Essentials
Written by

Java Architect Essentials

Committed to sharing quality articles and tutorials to help Java programmers progress from junior to mid-level to senior architect. We curate high-quality learning resources, interview questions, videos, and projects from across the internet to help you systematically improve your Java architecture skills. Follow and reply '1024' to get Java programming resources. Learn together, grow together.

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.