Cortex IDE

A multi-model AI-powered coding IDE that routes every task to the right model automatically. Users bring their own API keys - stored in their browser only, never on the server.

What it does

• Smart routing: DeepSeek for code, Kimi for long context, Gemini for multimodal, OpenAI for structured output, Claude as orchestrator only
• Monaco editor: multi-tab, syntax highlighting, 20+ languages, Ctrl+S save
• AI actions: Explain, Refactor, Fix Bug, Write Tests, Document, Complete - reads full file context automatically
• Browser file access: open any folder from your device directly (File System Access API)
• Integrated terminal: WebSocket terminal in the IDE (local mode only)
• 8 built-in agents: multi-step AI workflows (Code Review, Security Audit, Debugger, REST API Builder, etc.)
• 11 MCP tools: code linter, JSON validator, regex tester, token estimator, web search, diff, and more
• Cost analytics: real-time token/cost tracking per provider with Claude efficiency gauge
• Optimization settings: configure which model is the orchestrator, worker roles per provider, budget limits
• Per-user API keys: users paste their own keys on the Keys page - no account needed

Supported providers

Provider	Best for	Header
OpenAI	Structured output, security review	X-OpenAI-Key
Anthropic (Claude)	Orchestration, final approval only	X-Anthropic-Key
Google Gemini	Multimodal, architecture, long context	X-Gemini-Key
DeepSeek	Code generation, debugging, refactoring	X-Deepseek-Key
Kimi (Moonshot)	Long documents, repo analysis	X-Kimi-Key
OpenRouter	100+ models via one key (best for new users)	X-Openrouter-Key

Run locally

# Clone
git clone https://github.com/Sachin-NK/Cortex.git
cd Cortex

# Backend (Python 3.8+)
pip install -r backend/requirements.txt
python -m uvicorn backend.main:app --host 0.0.0.0 --port 8000 --reload

# Frontend (Node 18+)
cd frontend
npm install
npm run dev

Open http://localhost:3000 then go to http://localhost:3000/keys to add your API keys.

Or just run run.bat from the cortex/ directory on Windows.

Note: the integrated terminal requires a persistent backend (local) - not available on Vercel serverless.

How API keys work

Keys are stored in browser localStorage and sent as HTTP request headers on every call. The backend uses them only for that request - never stored anywhere.

User browser
  localStorage: { openai: "sk-...", gemini: "AIza..." }
       |
       | GET /api/chat
       | Headers: X-OpenAI-Key: sk-...
       |          X-Gemini-Key: AIza...
       v
  Vercel / Backend
       |
       | Uses key from header for this request only
       v
  OpenAI / Gemini / etc.

Architecture

frontend/          React + Vite + Monaco + xterm
  src/
    pages/         Dashboard, IDE, Chat, Agents, Costs, Keys, Optimization
    components/
      ide/         ActivityBar, FileExplorer, TerminalTabs, CommandPalette...
      AIPanel      Chat / Inline / Review modes

backend/           FastAPI (Python)
  main.py          All API routes + per-request key injection
  core/
    router.py      Smart provider scoring and fallback chain
    classifier.py  Task type detection (13 types)
    workflow.py    Multi-step agent execution + SQLite checkpointing
    memory.py      Session + scoped memory store
    security.py    Secret redaction, tool allowlists, privacy tiers
    token_accounting.py  Cost tracking + Claude budget enforcement
    observability.py     Execution traces
    file_manager.py      Workspace file operations
    user_config.py       Per-provider limit storage
  providers/       OpenAI, Anthropic, Gemini, DeepSeek, Kimi, OpenRouter, Mock
  agents/          8 built-in multi-step workflow definitions
  mcp/             11 developer tools (lint, diff, hash, regex, web search...)

vercel.json        Monorepo deploy: frontend service + backend service

Local .env

Copy .env.example to .env for local development with server-side keys. In production on Vercel, users supply their own keys from the browser.