IndirectTek Vibe Engine 🤖⚡

> "Autonomous Local AI Agent. Maximum Badassery."

IndirectTek Vibe Engine turns your VS Code into an autonomous development powerhouse. It writes code, creates files, and executes commands—running entirely on your infrastructure.

---

🚀 Quick Start (Direct Mode)

Prerequisites:
- Ollama installed and running (http://localhost:11434).
- Model: qwen2.5-coder:14b (Recommended). Pull it via ollama pull qwen2.5-coder:14b.

$3

1. Install "IndirectTek Vibe Engine" from VS Code Marketplace.
2. Open the IndirectTek AI chat panel.
3. Type: "Create a hello world python script".
4. Important: Check your Settings! The default configuration might point to a specific IP (192.168...). Change partnerBot.ollamaUrl to http://localhost:11434 (or your server IP) if running locally.

_Note: The agent connects to http://localhost:11434 by default._

---

🛠️ Advanced: Vibe Controller (Optional)

For Telemetry, Dashboards, and Enterprise Features, run the standalone Vibe Controller:

1. Run: npx indirecttek-vibe-engine@latest
2. Copy the Token.
3. In VS Code Settings:
- Set Use Controller to true.
- Paste the Controller Token.

---

✨ Key Features (v2.6)

- 100% Data Privacy: Your code never leaves your network.
- Permission System (Yellow Card): Interactive approval for sensitive actions:
- Shell: run_command requires manual allow (Always/Once/Deny).
- Visual: generate_image triggers approval.
- Files: create_file, edit_file, delete_file check permissions.
- Capabilities: Command/install/delete and file/delete are gated separately.
- Cybernetic Vision: The agent can now Scan Directories (LIST_DIR) to understand your project structure instantly.
- Studio Mode: Generate assets using generate_image (requires Fooocus API).
- Intent Router: Automatically switches models based on task (Chat vs Refactor vs Plan).
- Autonomous Editing: Strict "Hunk Only" editing to prevent creating massive diffs.
- Working Memory: Keeps goal/last tool/next step to maintain momentum.

⚙️ Configuration (The "Router")

You can map distinct models to different "Intents" in your settings.json:

``json
"partnerBot.models.fast": "qwen2.5:7b", // Chat, Explanations
"partnerBot.models.default": "qwen2.5-coder:14b", // Refactoring, Editing
"partnerBot.models.deep": "qwen2.5:32b", // Planning, Complex Logic
"partnerBot.allowedCommands": ["npm install", "cd", "ls", "grep"], // Whitelist
"partnerBot.allowDangerousCommands": false, // Override hard-deny list (rm/sudo/etc)
"partnerBot.toolEnforcement": "soft" // off | soft | strict
`

The agent will automatically route "Fix this" requests to the Default (Coder) model and casual chat to the Fast model.

🚨 Troubleshooting


- Command 'cd' not in allowed list: Update to v2.6.39+.
- Image Generation fails: Ensure Fooocus is running and partnerBot.fooocusUrl is correct.
- Connection Refused: Check your ollamaUrl`. If using WSL or Docker, use the host IP.
- New VSIX not taking effect: Uninstall and reinstall the VSIX (reload alone may keep the old extension host).

---

📜 License

MIT License. Built with ❤️ and ☕ by IndirectTek.