Tilcode

A clean, observable, and provider-agnostic CLI coding assistant based on Claude Code.

Features

• Works with any OpenAI-compatible API (MiniMax, DeepSeek, OpenRouter, etc.)
• No telemetry, no data upload, no login required
• Multi-model quick-switch via /model command
• Optional Langfuse observability integration
• Auto-discovery of skills from project directories
• File-based configuration (.tilcode/provider.json)

Install

Requires Node.js >= 18 and Bun >= 1.3.

# Install Bun (if not already installed)
curl -fsSL https://bun.sh/install | bash

# Install Tilcode
npm install -g @antaif3ng/tilcode

Quick Start

1. Create ~/.tilcode/provider.json:

{
  "provider": "openai-compat",
  "model": "your-model-name",
  "apiKey": "your-api-key",
  "baseUrl": "https://your-provider.com/v1"
}

2. Run:

tilcode

Configuration

Provider (~/.tilcode/provider.json)

User-level config applies to all projects. Project-level (.tilcode/provider.json in project root) overrides user-level.

{
  "provider": "openai-compat",
  "model": "MiniMax-M2.7-highspeed",
  "apiKey": "sk-xxx",
  "baseUrl": "https://api.minimaxi.com/v1",
  "models": [
    {
      "name": "MiniMax-M2.7-highspeed",
      "label": "MiniMax M2.7",
      "description": "Default model"
    },
    {
      "name": "deepseek-chat",
      "label": "DeepSeek V3",
      "apiKey": "sk-ds-xxx",
      "baseUrl": "https://api.deepseek.com/v1"
    }
  ]
}

Fields in models[] entries:

• name (required) - model name sent to API
• label - display name in /model picker
• description - description in picker
• apiKey - override top-level key for this model
• baseUrl - override top-level URL for this model
• maxOutputTokens - clamp max output tokens (auto-detected if not set)

Observability (~/.tilcode/config.json)

{
  "langfuse": {
    "enabled": true,
    "secretKey": "sk-lf-xxx",
    "publicKey": "pk-lf-xxx",
    "host": "https://cloud.langfuse.com"
  }
}

Usage

# Interactive mode
tilcode

# Pipe mode
echo "explain this code" | tilcode -p

# With a specific model
tilcode --model deepseek-chat

# Switch model at runtime
/model

License

MIT