Overview
Code Assistant is a local AI-powered tool that lets you chat with your codebase using RAG (Retrieval-Augmented Generation). Index your code, ask questions, and get intelligent answers with relevant context.
100% Private
Your code stays on your machine. Use local LLMs for complete privacy.
Multiple Providers
Connect to Gemini, OpenAI, OpenRouter, Z.AI, LM Studio, or Ollama.
E2E Encrypted Sync
Optional cloud sync with military-grade encryption. Even we can't read your data.
Getting Started
Get up and running in three simple steps.
Create a Project
Go to Projects page, click "New Project", and give it a name. Each project has its own indexed files, chat history, and settings.
Configure Your LLM
In Project Settings, choose your AI provider (local or cloud), select a model, and save. For cloud providers, add your API key in Global Settings first.
Upload & Index Your Code
Go to the Ingest tab, drag and drop your files or folders, then click "Start Indexing". Once complete, start chatting!
Installation
Code Assistant runs as a desktop application with bundled Python runtime. No additional dependencies required.
System Requirements
Minimum:
- 8 GB RAM
- 2 GB disk space
- Windows 10, macOS 11+, or Linux
Recommended:
- 16 GB RAM
- GPU for local LLMs (optional)
- SSD for faster indexing
Windows
Download the installer (.exe) from GitHub Releases and run it. Python is bundled - no separate installation needed.
macOS & Linux
Download the appropriate package from GitHub Releases. Docker deployment is also available for server environments.
LLM Providers
Choose the AI provider that fits your needs - from free local options to powerful cloud APIs.
Local Providers (Free & Private)
LM Studio
User-friendly GUI for running local models. Great for beginners.
- Download models with one click
- Built-in local server
- No API key required
Ollama
CLI-based tool for power users. Fast model switching.
- Command-line interface
- Easy model management
- No API key required
Cloud Providers
Gemini
Google AI with generous free tier.
Free tier availableOpenAI
GPT-4o and other premium models.
PaidZ.AI
GLM-4.7 and other GLM models.
Free & paid modelsOpenRouter
Access 100+ models from one API.
Free & paid modelsUsage Guide
Get the most out of your code conversations.
Example Questions
"What does this project do?"
"Explain the authentication flow"
"Where is the database connection handled?"
"Why might this function return null?"
Tips for Better Results
- Be specific - "How does the login function validate passwords?" beats "How does login work?"
- Index relevant files - Only index the code you want to query for faster, more accurate results.
- Try different models - Some models handle code better than others. Experiment to find your best fit.
Security & Privacy
Your code security is our top priority. Choose your privacy level.
Maximum Privacy (Local)
Use LM Studio or Ollama - your code never leaves your machine.
- All processing happens locally
- No internet required after setup
- Zero data transmission
Cloud with E2E Encryption
For cloud sync users, all data is encrypted before leaving your device.
- XChaCha20-Poly1305 encryption
- Argon2id key derivation
- 24-word recovery key backup
Data Handling by Provider Type
| Provider | Code Destination |
|---|---|
| LM Studio / Ollama | Stays on your machine |
| OpenAI / Gemini / Z.AI / OpenRouter | Sent to APIs (encrypted in transit) |
Supported File Types
Code
.py, .js, .ts, .tsx, .jsx, .java, .cpp, .go, .rs, .rb
Web & Config
.html, .css, .json, .yaml, .toml, .env
Documentation
.md, .txt, .rst
Excluded
Binary files, images, videos are automatically skipped
API Key Setup
Troubleshooting
Chat Not Responding
- Verify your LLM provider is running (for local providers)
- Ensure a model is selected in Project Settings
- Check API key validity for cloud providers
- Try refreshing the page or restarting the app
Search Results Irrelevant
- Be more specific in your questions
- Verify files are indexed (check Ingest tab)
- Increase retrieval count in Settings
- Re-index if you've made major code changes
Embedding Model Mismatch Error
This occurs when trying to index with a different embedding model than originally used. Go to Ingest tab, clear indexed files, then re-index with your preferred model.
Frequently Asked Questions
Is my code sent to the cloud?
It depends on your provider choice. Local providers (LM Studio, Ollama) keep everything on your machine. Cloud providers send code to their APIs, encrypted in transit.
Which provider should I use?
For maximum privacy, use LM Studio or Ollama. For free cloud access, try Gemini. For best quality, OpenAI GPT-4o or Z.AI GLM-4.7. For variety, OpenRouter offers 100+ models.
Can I change the embedding model after indexing?
No, the embedding model is locked per project. To change it, you must clear all indexed files and re-ingest from scratch.
Do I need a GPU?
Not required. Cloud providers work without a GPU. For local LLMs, a GPU helps but CPU-only mode works for smaller models.