AI Engineer

An interactive terminal coding assistant that talks to multiple LLM providers. Supports DeepSeek (with chain-of-thought display), OpenRouter, and any OpenAI-compatible endpoint. File operations are handled through function calling — the AI reads, creates, and edits files automatically as part of the conversation.

Demo

Providers and Models

DeepSeek

• deepseek-reasoner (default) — reasoning model with visible chain-of-thought
• deepseek-chat — faster, cheaper general-purpose model

OpenRouter

• google/gemini-2.5-pro-preview
• anthropic/claude-sonnet-4
• anthropic/claude-3.7-sonnet
• openai/gpt-4.1
• openai/o3-mini-high
• ...and any other model available on OpenRouter

Features

• Multi-provider support with runtime model switching (/model set <name>)
• Automatic file operations — mention a file in conversation and the AI reads it
• Recursive task completion — keeps iterating (up to 10 rounds) until the task is done, with linter feedback between iterations
• Multi-language linting — Python (flake8), JavaScript (ESLint), TypeScript (tsc + ESLint)
• MCP integration — connects to Model Context Protocol servers like Context7 for live documentation lookup
• Rich terminal UI — streaming output, color-coded feedback, model/provider indicator in the prompt

Installation

git clone https://github.com/FaustoS88/AI_Engineer
cd AI_engineer
cp .env.example .env
# add your API keys to .env

Install dependencies (pick one):

# uv (faster)
uv venv && uv run ai-engineer.py

# pip
pip install -r requirements.txt && python3 ai-engineer.py

For JavaScript/TypeScript linting support:

npm install -g eslint typescript

Environment

# .env
DEEPSEEK_API_KEY=your_key_here
OPENROUTER_API_KEY=your_key_here

You only need the key for the provider(s) you actually use.

Commands

Command	Description
/model list	Show all available models
/model current	Show current model
/model set <name>	Switch model
/mcp list	Show MCP server status
/mcp enable <server>	Enable an MCP server
/mcp disable <server>	Disable an MCP server
/mcp reload	Reload MCP config without restart
/add <path>	Preload a file or folder into context
exit / quit	End session

MCP Integration

MCP servers are configured in mcp.config.json. Example with Context7 (library docs) and Brave Search:

{
  "servers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp@latest"],
      "description": "Up-to-date library documentation",
      "enabled": true
    },
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-brave-search"],
      "env": { "BRAVE_API_KEY": "${BRAVE_API_KEY}" },
      "description": "Brave Search for web queries",
      "enabled": true
    }
  }
}

Servers are lazy-initialized and hot-reloadable via /mcp reload.

Function Calling Tools

The AI can call these automatically during conversation:

• read_file(file_path) — read a single file
• read_multiple_files(file_paths) — batch read
• create_file(file_path, content) — create or overwrite
• create_multiple_files(files) — scaffold multiple files at once
• edit_file(file_path, original_snippet, new_snippet) — precise snippet replacement

Troubleshooting

API key not found — check that .env exists and has the right key name.

Model not available — run /model list to see what's configured, and confirm you have the API key for that provider.

Import errors — run uv sync or pip install -r requirements.txt.

JS/TS linting not working — make sure eslint and typescript are installed globally (npm install -g eslint typescript). See MULTI_LANGUAGE_SETUP.md for details.

License

MIT — see LICENSE.