Skip to main content
CodePilot IDE goes beyond a simple AI chat tool — it works through your entire real-world development workflow with you.

​
Multi-LLM and Hybrid Operation

Within a single IDE, freely switch between local and cloud models as you work.
  • Cloud models — High-performance commercial models registered by your Organization
  • Local models — Models installed on in-house PCs or servers; run without any external network
  • Real-time switching — Change models even in the middle of a conversation
For security-sensitive projects, use local models only to eliminate data leakage at the source, and switch models based on task difficulty to optimize costs.

​
Automatic Model Assignment by Purpose

Different models are used automatically depending on the type of task. Because expensive models aren’t used for simple tasks, operating costs go down. You can assign models by purpose yourself in Settings → AI Models.

​
Automatic File Creation, Modification, and Deletion

When you make a request via chat, the AI creates and modifies files directly. It edits multiple related files at once — such as controllers, services, and DTOs — and updates the call sites along with them.

​
Inline Diff Preview and Approval

Diff view Compare the code the AI modified right inside the editor and apply only the parts you want.
  • Inline diff — Changes are shown inline within your current code, with no separate window
  • Real-time display — The diff is drawn as the AI writes code, so you can review it even before it finishes
  • Selective apply — Keep / Undo on a per-change-chunk basis
  • Turn-level handling — Confirm or undo all of the current response’s changes at once
  • Persists across restarts — Changes you haven’t approved yet remain even after you restart the IDE
Secures both the AI’s automation and the developer’s control at the same time.

​
Automatic Terminal Command Execution

The AI handles repetitive terminal work such as building, testing, and deployment on your behalf. It automatically converts commands to the shell syntax appropriate for your operating system, then reads the output to determine whether they succeeded.

​
Automatic Error Correction

The AI detects errors that occur in the terminal and attempts to fix them on its own. It repeats “fix → re-run → verify” for the configured number of times, and reports back to you if the issue isn’t resolved.
Even when errors occur, it attempts an immediate fix without any web searching, so your development flow is never interrupted.

​
Multi-Agent

Complex requests are split into multiple subtasks and processed. Tasks that don’t depend on each other run concurrently, while tasks that need the results of an earlier step run in sequence. Once all tasks are done, the results are merged into one and verified with builds and tests. Task progress

​
Automatic Build/Test Verification

After writing or fixing code, it automatically runs builds and tests. It detects the project type to pick the right commands, and if they fail, it analyzes the cause, applies a fix, and verifies again.
Turns “a procedure people have to remember” into a procedure the system enforces.

​
Keyword-Based Automatic Command Execution (Hot Load)

Hot Load Register frequently repeated tasks as keywords, and simply typing a sentence containing that keyword in the chat runs a predefined command. It runs even when the wording isn’t an exact match, as long as the meaning comes through.

​
Security Guardrails

Security settings Before the AI acts, it pre-validates every operation to block dangerous executions.
  • Blocked commands — Prevents the execution of commands that could destroy the system
  • Protected files — Blocks modification or deletion of sensitive files such as .env and certificates
  • Hidden files — Conceals specific files so the AI cannot read or search them
  • Out-of-project access blocking — Blocks access to files outside the working folder
  • Credential protection — Inspects responses so that secret keys don’t slip out
No matter how autonomously the AI operates, it cannot cross the security boundary.

​
Inline Code Completion

The moment you type, the AI suggests the code that comes next as gray text. Accept it with the Tab key, or dismiss it with Esc. It considers not only the current file but also the context of other files open in the editor. Inline code completion
It’s OFF by default. Turn it on in Settings → General. If you assign a lightweight local model dedicated to completion, you can use it quickly with no API costs.

​
Leveraging Project Context and Internal Knowledge

RAG The AI figures out your project structure and the tech stack in use on its own, and searches the internal documents (RAG) registered by your Organization to incorporate them into its answers. It also references the syntax and type errors that VS Code detects.
It generates code “tailored to your project” rather than generic examples from the internet.

​
Next-Task Suggestions

Next-task suggestions When a task finishes, it suggests follow-up tasks that naturally come next. Click the button and that request is sent right away.

​
Clarifying Questions from the AI

AI question When a decision is needed mid-task, the AI presents options and asks you. Answer with a single click and it continues the work based on your choice.
Because the AI confirms rather than assumes, you get the result you want.

​
MCP Tool Integration

MCP Connect external tools such as GitHub, Jira, and in-house systems so the AI can operate them directly. MCP servers pre-registered by your Organization administrator can be used right away with no additional setup.