🚀 Connect AI: Autonomous Code Agent

The open-source, elite-tier AI coding assistant that brings Ollama Cloud, Gemini, Grok, Deepseek, Glm, Minimax, Gemma, Claude and GPT-4o features directly into your VS Code environment.

⚡ Why Connect AI?

We were tired of paying $20/month for proprietary IDE forks or heavily-gated extensions that don't let you bring your own keys. Connect AI is a native VS Code extension built to give you absolute, unrestricted access to top-tier agentic coding capabilities using Ollama (Local and Cloud), Deepseek, Gemini, Grok, GPT-4o,Claude.

We built it with the hardcore power features you expect from elite AI tools.

🔥 Killer Features

👻 1. Native FIM "Ghost Text" Autocomplete

As you type, Connect AI actively predicts your next lines using strict Fill-In-The-Middle (FIM) boundary targeting (<|fim_prefix|>). With a hyper-aggressive 1.5s timeout threshold, if the suggestion isn't lightning fast, it gets out of your way. Press Tab to accept instantly.

🎯 2. @file Explicit Context Pinning

Tired of the AI hallucinating? Type @file directly into your chat. Connect AI will instantly surface a dropdown of your workspace buffers. Click a file, and we seamlessly parse the local byte-buffer and pin it to the absolute top of the System Prompt perfectly.

🎼 3. Multi-File Composer

When you ask the AI to refactor an entire React component pipeline (e.g. Update index.ts, styles.css, and App.tsx), you don’t get broken code blocks. You get the Composer Widget. Click "Review in Composer" to launch a massive, tabbed Side-by-Side Diff UI covering all modified files. Hit "Apply All" to batch-mutate your disk.

🧠 4. Autonomous Codebase RAG (@codebase)

Type @codebase -> Connect AI instantly sweeps a local Vector Graph database constructed from your source code using internal embeddings. It grabs the exact architectural context needed to answer questions without you opening a single file.

💻 5. Instant Terminal Scraping (@terminal)

Did your compiler crash? Don't copy-paste the error. Type @terminal. Connect AI silently triggers the OS clipboard, sweeps VS Code's active terminal buffer, cleanly truncates the stack trace, and injects exactly what went wrong.

💡 6. Diagnostic Quick Fixes

When VS Code throws a Red Squiggle (TypeScript / ESLint error), click the Lightbulb 💡. Select "Fix with Connect AI". The agent reads the exact compiler trace, rewrites your function immediately, and clears the error.

🧩 The Connect AI "Swarm" Architecture

Type /swarm in the chat to unleash the brutal Multi-Agent Swarm. Instead of begging a single LLM to write code, you trigger a microservice pipeline:

1. ⚙️ Executor Agent: Churns through the roadmap and outputs code replacements.
2. 🔬 Reviewer Agent: Sweeps the output for vulnerabilities or logical fallacies.
3. 🩺 Fixer Agent: Closes the loop, finalizing the perfect diff buffer.

🛠️ Bring Your Own Models (BYOM)

Connect AI doesn't lock you down. Use the custom Provider API interface:

• Cloud (Unlimited Power): Connect to the world's best models including Ollama Cloud, Claude, Gemini, Grok, and GPT-4o using native API integration or OpenRouter.
• Local (100% Privacy): Hook Connect AI to deepseek-coder, starcoder, or qwen-coder using native Ollama detection.

📦 Installation & Setup

1. Add Your API Keys:

   • Cloud Providers: Open the settings gear in the Connect AI Chat Sidebar or run the command Connect AI: Set Provider API Key.
   • Ollama Cloud (NEW):
     • Go to ollama.com/settings/keys to generate your API key.
     • Select ollamaCloud as the provider and paste your key.
     • Choose from elite models like deepseek-v3.1:671b-cloud, qwen3.5:latest, or minimax-m2.7.
   • Local Ollama: Ensure Ollama is running locally (ollama serve). Select ollama as the provider; no API key is required for local instances.
   • Other Providers: Supports OpenAI, Anthropic (Claude), Google (Gemini), xAI (Grok), and OpenRouter.

2. Start Chatting: Open the Connect AI Chat Sidebar and start asking questions about your code. Use @file, @git, or @codebase for rich context.