Mercury Code

Interactive AI coding assistant powered by the Mercury-2 diffusion model from Inception Labs. Like Claude Code, but for Mercury-2.

Installation

npm (recommended, all platforms)

npm install -g mercury-code

After installation, the mercury and mercury-code commands are available globally.

Windows Installer (EXE)

Download the latest mercury-code-vX.X.X-windows-setup.exe from Releases. The installer:

• Installs to Program Files
• Adds mercury and mercury-code to your PATH
• Creates Start Menu shortcuts
• Includes an uninstaller

Standalone Binaries (no Node.js required)

Download the appropriate binary from Releases:

| Platform | Architecture | File | |----------|-------------|------| | Windows | x64 | mercury-code-vX.X.X-win-x64.zip | | Windows | ARM64 | mercury-code-vX.X.X-win-arm64.zip | | macOS | Intel | mercury-code-vX.X.X-macos-x64.tar.gz | | macOS | Apple Silicon | mercury-code-vX.X.X-macos-arm64.tar.gz | | Linux | x64 | mercury-code-vX.X.X-linux-x64.tar.gz | | Linux | ARM64 | mercury-code-vX.X.X-linux-arm64.tar.gz |

Extract and move to your PATH:

# Linux / macOS
tar xzf mercury-code-v*.tar.gz
chmod +x mercury-code-*
sudo mv mercury-code-* /usr/local/bin/mercury-code

Install from Source

git clone https://github.com/LHAMNS/Cloud-code-for-mercury-2.git
cd Cloud-code-for-mercury-2
npm link
# or: ./install.sh (Linux/macOS)
# or: powershell -ExecutionPolicy Bypass -File install.ps1 (Windows)

Quick Start

# 1. Set your API key (from https://api.inceptionlabs.ai)
export INCEPTION_API_KEY=your_key_here

# 2. Start interactive session
mercury

# Or use single-shot mode
mercury -p "Explain this codebase"

Usage

mercury                              # Interactive REPL
mercury -p "your question"           # Single-shot mode
mercury --plan                       # Plan mode (read-only analysis)
mercury --verbose                    # Verbose logging
mercury --trust-mode aiSafetyDecide  # AI evaluates safety of each operation
mercury --help                       # Full options

Trust Modes

| Mode | Level | Description | |------|-------|-------------| | open | 0 | Auto-allow everything in workspace | | aiSafetyDecide | 0.5 | AI judges each operation's safety | | acceptEdits | 1 | Auto-approve file edits, ask for bash | | approval | 2 | Ask before writes and commands (default) | | dontAsk | 3 | Allow reads, deny everything else | | readonly | 4 | Read-only, no writes/bash/agents |

Interactive Commands

| Command | Description | |---------|-------------| | /help | Show all available commands | | /clear | Clear conversation history | | /reasoning <level> | Set reasoning effort (instant/low/medium/high) | | /trust <mode> | Switch trust mode | | /context | View context usage | | /export <format> | Export conversation (md/json/txt/html) | | /agents | List available agent types | | /settings | View/modify settings | | /exit | Exit Mercury Code |

Built-in Tools

| Tool | Description | |------|-------------| | Read | Read files | | Write | Create/overwrite files | | Edit | Precise text replacements | | Bash | Execute shell commands (sandboxed) | | Glob | Find files by pattern | | Grep | Search file contents with regex | | Fetch | HTTP requests | | Diff | Compare files | | ListDir | List directory contents | | SubAgent | Spawn sub-agents for parallel work | | SubAgentTeam | Run multiple agents concurrently | | AgentTeams | Collaborative agent teams |

Security

• Sandbox isolation: Workspace boundary enforcement, path traversal protection
• Content scanning: Detects secrets (API keys, private keys, tokens) in file writes
• SSRF protection: Blocks private IPs, DNS fail-closed, credential detection
• Rate limiting: Per-minute limits for bash and fetch operations
• AI Safety Judge: Optional trust mode where AI evaluates each operation

Configuration

Global config: ~/.mercury/config.json

{
  "model": "mercury-2",
  "max_tokens": 50000,
  "temperature": 0.75,
  "reasoning_effort": "medium",
  "stream": true
}

API

Mercury-2 uses an OpenAI-compatible endpoint at https://api.inceptionlabs.ai/v1/chat/completions.

Set INCEPTION_API_KEY in your environment. Context window: 128K tokens. Max output: 50K tokens.

Building from Source

# Run tests
make test

# Build standalone binaries for all platforms
make build

# Build for specific platform
make build-win    # Windows x64 + ARM64
make build-mac    # macOS Intel + Apple Silicon
make build-linux  # Linux x64 + ARM64

# Full release (test + build)
make release

License

MIT