Enfuse Sidecar - AI-Powered App Builder

A local-first AI development assistant that generates and executes code using your on-premises LLM infrastructure.

🚀 Quick Start

# Install globally from npm
npm install -g @enfuseio/sidecar

# Or install from source
cd apps/sidecar
npm install
npm link

# Initialize a project
mkdir myproject && cd myproject
enfuse init

# Describe what you want to build
enfuse "Create an Express API with user authentication"

# Review and approve the plan
enfuse plan approve -a

# Execute and auto-commit
enfuse plan execute --commit

✨ Features

Core Features

• 🤖 Local LLM Integration - Uses your local LLM (default: localhost:8000)
• 📋 Plan Generation - AI generates structured, executable plans
• ✅ Step Approval - Review diffs before any changes
• 📁 File Operations - Creates, modifies files automatically
• 🔧 Command Execution - Runs npm install, tests, etc.
• 📝 Git Integration - Auto-commit changes after execution

Nice-to-Have Features

• 💬 Interactive Mode - REPL with continuous chat
• 🔄 Plan Refinement - Iterate on plans with feedback
• 🧪 Test Running - Integrated test execution
• 🖥️ Live Preview - Start dev servers automatically

📖 Commands

Task Execution

# Generate a plan from task description
enfuse "Create a React dashboard with charts"

# Run in interactive mode (REPL)
enfuse

Plan Management

# View current plan
enfuse plan

# Show detailed plan with diffs
enfuse plan show -v

# Approve all pending steps
enfuse plan approve -a

# Approve specific step
enfuse plan approve 2

# Refine plan with feedback
enfuse plan refine "Use TypeScript instead of JavaScript"

# Execute approved steps
enfuse plan execute

# Execute with auto-commit
enfuse plan execute --commit

# Clear current plan
enfuse plan clear

Development

# Start dev server
enfuse dev --preview

# Run tests
enfuse test

# Run tests in watch mode
enfuse test --watch

# Run tests with coverage
enfuse test --coverage

System

# Check system health
enfuse status

# Debug health check
enfuse debug health -v

# View budget/usage
enfuse budget

# Initialize project
enfuse init

🎮 Interactive Mode

Start interactive mode with just enfuse:

$ enfuse
✓ Connected to openai/gpt-oss-20b

enfuse ➜ How do I add authentication to Express?
[AI responds with explanation]

enfuse ➜ do Create JWT middleware
[Generates executable plan]

enfuse ➜ approve
✓ Approved 3 steps.

enfuse ➜ execute
✓ Execution complete!

enfuse ➜ help
━━━ Interactive Commands ━━━

▸ Chat
• Just type to chat with the AI
• "reset" - Clear conversation history

▸ Tasks
• "do <task>" or "create <task>" - Generate executable plan
• "plan" - View current plan
• "approve" - Approve all pending steps
• "execute" or "run" - Execute approved steps

▸ System
• "status" - Check system health
• "clear" - Clear screen
• "exit" or "quit" - Exit

📁 Project Structure

After enfuse init, your project will have:

myproject/
├── .enfuse/
│   ├── plans/          # Saved plans
│   │   └── current.json
│   ├── cache/          # LLM response cache
│   ├── logs/           # Execution logs
│   └── checkpoints/    # Rollback checkpoints
├── enfuse.config.json  # Configuration
└── .gitignore

⚙️ Configuration

enfuse.config.json:

{
  "environment": "dev",
  "platformApiUrl": "https://platform-dev.enfuse.ai",
  "localLlmUrl": "http://localhost:8000/v1",
  "loraLlmUrl": "http://localhost:8001/v1",
  "logLevel": "info",
  "autoCommit": true,
  "autoRunTests": true,
  "dailyBudgetLimit": 10,
  "sessionBudgetLimit": 5
}

Environment Variables

ENFUSE_LLM_URL=http://localhost:8000/v1
ENFUSE_LORA_URL=http://localhost:8001/v1
ENFUSE_API_URL=https://platform-dev.enfuse.ai
ENFUSE_API_KEY=your-api-key

🔄 Workflow Example

Create a TypeScript Express API

# 1. Start fresh
mkdir my-api && cd my-api && git init

# 2. Initialize Enfuse
enfuse init -y

# 3. Describe your goal
enfuse "Create an Express API with GET /users endpoint"

# Output:
# ▸ Generated Plan:
#   ○ 1. [+] Create package.json
#   ○ 2. [cmd] Install Express
#   ○ 3. [+] Create index.js

# 4. Want TypeScript? Refine the plan
enfuse plan refine "Use TypeScript with proper types"

# Output:
# ✓ Plan refined!
# ▸ Updated Steps:
#   ○ 1. [+] Create package.json (with TS deps)
#   ○ 2. [cmd] Install dependencies
#   ○ 3. [+] Create tsconfig.json
#   ○ 4. [+] Create src/index.ts

# 5. Review the plan
enfuse plan show -v

# 6. Approve all steps
enfuse plan approve -a

# 7. Execute with git commit
enfuse plan execute --commit

# 8. Start the dev server
enfuse dev --preview

# 9. Test your API
curl http://localhost:3000/users
# [{"id":1,"name":"Alice"},{"id":2,"name":"Bob"}]

🧪 Supported Test Frameworks

• vitest (recommended)
• jest
• mocha
• ava
• tap

Auto-detected from package.json.

🖥️ Supported Project Types

For enfuse dev --preview:

• Next.js - npm run dev on port 3000
• Vite - npm run dev on port 5173
• Create React App - npm start on port 3000
• Express - npm start on port 3000
• FastAPI - uvicorn on port 8000
• Static - serve on port 3000

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                    enfuse CLI                               │
├─────────────┬─────────────┬─────────────┬──────────────────┤
│   init      │   plan      │   test      │   dev            │
│   status    │   refine    │   debug     │   budget         │
└─────────────┴──────┬──────┴─────────────┴──────────────────┘
                     │
        ┌────────────┴────────────┐
        │                         │
┌───────▼───────┐    ┌────────────▼────────────┐
│ Plan Manager  │    │ Plan Generator (LLM)   │
│ (.enfuse/     │    │ → Structured JSON      │
│  plans/)      │    │ → File content         │
└───────┬───────┘    └────────────────────────┘
        │
┌───────▼───────┐    ┌─────────────────────────┐
│ Plan Executor │───▶│ Local LLM (localhost)   │
│ → Files       │    │ openai/gpt-oss-20b      │
│ → Commands    │    │ ~19-50ms latency        │
│ → Git commit  │    └─────────────────────────┘
└───────────────┘

📊 Plan Step Types

| Type | Icon | Description | |------|------|-------------| | create | [+] | Create a new file | | modify | [~] | Modify existing file | | delete | [-] | Delete a file | | command | [cmd] | Run shell command | | test | [test] | Run tests |

🔒 Security

• Local-first: Your code never leaves your network
• On-prem LLM: Uses your local LLM infrastructure
• Approval required: No changes without explicit approval
• Diff preview: See exactly what will change
• Git integration: Every change is trackable

🐛 Troubleshooting

LLM not responding

enfuse status
# Check "Local LLM" status

enfuse debug health -v
# Detailed health check

Plan execution fails

enfuse plan
# Check step statuses

enfuse plan show -v
# See detailed step info

Reset everything

enfuse plan clear -f
# Clear current plan

rm -rf .enfuse
enfuse init
# Reinitialize

📚 More Resources

• Sidecar App Builder Report
• Platform Documentation

🏷️ Version

v0.5.0 - Published to npm as @enfuseio/sidecar