CodePilot AI
🌐 Live Website: https://codepilot-website.vercel.app
CodePilot AI is a VS Code extension that brings a local AI coding assistant directly into the editor.
It can explain code, detect errors from VS Code diagnostics, generate fixes, answer questions about the current project, and use RAG to understand files across the workspace. The current version runs with local Ollama models, so no paid API is required.
Features
CodePilot AI currently includes:
Ask AI directly from the Explorer sidebar
Fast, Normal, and Detailed response modes
Current-file and cursor-aware answers
Function-level context using VS Code document symbols
Short conversation memory for follow-up questions
Project-wide RAG search using LangChain
Automatic project indexing on startup
Incremental re-indexing when files are saved
Source attribution for project-aware answers
Clickable source files that open directly in VS Code
Automatic VS Code error detection
Screenshots
Project-Aware AI Chat
CodePilot can answer questions using the current project context and show the relevant source files.
Error Detection and AI Fixes
CodePilot detects VS Code diagnostics, generates a proposed fix, and lets the user review it before applying changes.
Explain Error
Generate Fix with preview
Apply Fix with safety checks
Fix All Errors with a separate preview before applying changes
Explain Selected Code
Fix Selected Code
Improve Code
Copy Response and Copy Fix actions
Example workflow
If VS Code detects an error, CodePilot can show it inside the sidebar:
Detected Error ↓ Generate Fix ↓ Review Proposed Fix ↓ Apply Fix
For multiple related errors:
Fix All Errors ↓ Review Proposed Fix All ↓ Apply All Fixes
This keeps code changes visible before they are applied.
Project-Aware AI
CodePilot can index the current workspace and retrieve relevant code before answering a question.
For example:
button logic kis file me hai?
CodePilot can search the indexed project, identify the most relevant files, answer the question, and show the source files underneath the response.
The RAG pipeline currently uses:
LangChain
RecursiveCharacterTextSplitter
OllamaEmbeddings
nomic-embed-text
in-memory vector stores
Requirements
Before running CodePilot AI, install:
Visual Studio Code
Node.js
npm
Ollama
The current setup uses these Ollama models:
ollama pull qwen3:4b ollama pull nomic-embed-text
Ollama should be running locally before starting the extension.
The extension connects to:
Running the Extension
Install the project dependencies:
npm install
Compile the extension:
npm run compile
Open the project in VS Code and press:
F5
A new Extension Development Host window will open.
Open the Explorer sidebar and expand:
CodePilot AI
Using CodePilot AI
Ask about the current function
Place the cursor inside a function and ask:
What does this function do?
CodePilot uses the current file, cursor position, and detected function as context.
You can then ask a follow-up such as:
Can you optimize it?
The recent conversation is kept in memory during the current extension session.
Ask about the project
Ask a project-level question such as:
Which file contains the button click logic?
If the project has been indexed, CodePilot retrieves relevant chunks and shows the source files used for the answer.
Fix a detected error
When VS Code reports an error:
Open CodePilot AI.
Review the detected error.
Click Generate Fix.
Review the proposed code.
Click Apply Fix only if the preview looks correct.
Fix multiple errors
When a file contains several related diagnostics:
Click Fix All Errors.
CodePilot creates a combined correction preview.
Review Proposed Fix All.
Click Apply All Fixes to apply the preview.
The extension does not automatically apply the Fix All result without showing the proposed code first.
Quick Actions
Select code in the editor and use:
Explain Selected Code
Fix Selected Code
Improve Code
These actions send the selected code directly to the local model.
How It Works
The main flow is:
VS Code │ ├── Diagnostics API │ └── Error detection and fix workflow │ ├── Active Editor │ └── Current file / cursor / function context │ ├── CodePilot Sidebar │ └── Questions, actions, previews and sources │ └── RAG Service ├── Workspace files ├── Text chunking ├── Local embeddings └── Similarity search │ ▼ Ollama / Qwen3
Tech Stack
TypeScript
VS Code Extension API
Ollama
Qwen3
LangChain
nomic-embed-text
MemoryVectorStore
RecursiveCharacterTextSplitter
esbuild
Current Limitations
CodePilot AI is still under active development.
Current limitations include:
RAG data is stored in memory and is rebuilt after the extension restarts.
Conversation history is session-based.
Very large repositories may take longer to index.
AI-generated fixes can still be incorrect, so previews should always be reviewed.
Function detection depends on VS Code language support and document symbol providers.
Fix All is designed around the active file, not multi-file refactoring.
Planned Improvements
Future versions may include:
Persistent vector storage
Git-aware context
Inline editor suggestions
Diff-based fix previews
Test generation
Terminal and runtime error awareness
Multi-file refactoring
Configurable local models
Extension settings for model and indexing preferences
Persistent project conversations
Release Notes
1.0.0
Initial development release of CodePilot AI.
Included:
Local Ollama integration
AI chat sidebar
current-file and function-aware context
project RAG
automatic and incremental indexing
source navigation
error detection
Generate Fix
Apply Fix
Fix All preview workflow
selected-code quick actions
conversation memory
loading states and response modes
Privacy
The current version is designed around local inference.
Code and project context are sent to the locally running Ollama server configured at:
The current implementation does not require a paid cloud AI API.
Notes
CodePilot AI is being built as a practical developer tool and as a project for exploring how VS Code extensions, local LLMs, RAG, diagnostics, and safe code editing can work together in one product.
AI-generated code should always be reviewed before applying it to important projects.