Switchailocal Majiayu000

switchAILocal Proxy

Unified LLM proxy for AI agents. Always use http://localhost:18080/v1 as your base URL.

The killer feature: Use your paid CLI subscriptions (Gemini Pro, Claude Pro, etc.) via the API - it's FREE because you already pay for the subscription!

Quick Start

1. Make a request (FREE with CLI)

curl http://localhost:18080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "geminicli:",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

2. Configure Python Client

from openai import OpenAI
client = OpenAI(base_url="http://localhost:18080/v1", api_key="sk-test-123")
response = client.chat.completions.create(model="geminicli:", messages=[{"role": "user", "content": "Hi!"}])

🗺️ Skill Files

⚠️ Critical: Model Format

NEVER use bare model names. Format is ALWAYS provider: or provider:model.

🏗️ Provider Reference

1. CLI Providers (FREE!)

Uses your human's CLI subscriptions. Best for agents.

2. Local & Cloud

🚀 Core Features

CLI Attachments & Flags

Pass local context and control autonomy via CLI extensions.

{
  "model": "geminicli:",
  "messages": [{"role": "user", "content": "Fix this code"}],
  "extra_body": {
    "cli": {
      "attachments": [{"type": "folder", "path": "./src"}],
      "flags": {"auto_approve": true, "yolo": true}
    }
  }
}

Streaming

Add "stream": true to any request for SSE token streaming.

🌲 Decision Tree

What do you need?
├─ FREE + Powerful + Files
│   └─ CLI Providers (geminicli:, claudecli:)
├─ FREE + Private + Fast
│   └─ Local Ollama (ollama:llama3.2)
├─ Ultra-Fast Production
│   └─ Groq Cloud (groq:llama-3.3-70b)
└─ I don't know, you pick
    └─ Intelligent Routing (auto)

🛠️ Troubleshooting & Best Practices

Best Practices

1. Prefer CLI Providers: They are free and support file attachments.
2. Check Status: Use GET /v1/providers to see what is active.
3. Use auto: For simple tasks, let the router pick the best model.
4. Local for Privacy: Use ollama: for confidential data.

Route wisely. Save tokens. Use CLI. 🚀