API Key User Manual

Each key starts with 100M (1x-equivalent) quota. Different models consume quota by multiplier. Higher multiplier means faster quota usage.

1) Purchase & Delivery

• Payment: USDT (TRC20)
• Auto key delivery after on-chain confirmation (one order = one new key)
• No delivery for unpaid/timeout orders
• Issued keys are non-refundable

2) Connection Settings

Base URL: https://gpt-agent.cc/v1
API Key : sk-xxxxxxxxxxxxxxxx

3) Tutorial: How to Use

Use in VS Code Plugins

Install a plugin that supports custom API endpoint or Claude/Anthropic-compatible format.

• base_url: https://gpt-agent.cc/v1 or https://gpt-agent.cc
• API Key: your key
• Protocol format: Anthropic-compatible
• Model name: claude-sonnet-4-6 or other supported model name

If your plugin defaults to OpenAI protocol, switch it manually to Claude/Anthropic-compatible format.

Use in Claude Code (with CC Switch)

CC Switch is a community open-source desktop tool for unified provider switching across Claude Code, Codex, Gemini CLI, OpenCode, etc.

• Open CC Switch and add provider: provider name, API key, Base URL, model name, target app (Claude Code)
• Switch provider to Claude Code via main UI or tray menu
• Confirm it is active (typically hot reload, no restart required)
• Use Claude Code normally in terminal

Use in OpenClaw (Separate Setup)

• Model Provider: Custom Provider
• Base URL: https://gpt-agent.cc/v1
• API Key: your key
• Model: choose your target model ID

OpenClaw should be configured separately in its own provider/model settings.

Use in Cherry Studio

• Model Provider: Custom Provider
• API Base URL: https://gpt-agent.cc/v1
• API Key: your key
• Endpoint Compatibility: Anthropic-compatible
• Model Name: claude-sonnet-4-6 or supported domestic model names

Rule: choose Claude-compatible and Anthropic protocol.

Connectivity Check

curl https://gpt-agent.cc/v1/models -H "Authorization: Bearer sk-xxxxxxxxxxxxxxxx"

Extra Notes Requested

• If Lobster OpenClaw crashes, you can install OpenCode, Claude Code, or Codex, then use local AI workflows to repair the AI agent.
• Now supports gpt-5.4, Codex, and Gemini 3.1 Pro. API is callable, with multiplier-based billing.

4) Billing Formula

Total usage = (Input Tokens + Output Tokens) 脳 Model Multiplier

5) Full Model List & Multipliers

Provider	Model	Multiplier	Effective Tokens
Lobster	claude-sonnet-4-6	1x	100M tokens
MiniMax	MiniMax-M2.5	1x	100M tokens
MiniMax	MiniMax-M2.5-highspeed	1x	100M tokens
Kimi	kimi-for-coding (Kimi 2.5)	2x	50M tokens
Qwen	qwen3.5-plus	1x	100M tokens
Qwen	qwen3-coder-plus	1x	100M tokens
StepFun	step-3.5-flash	1x	100M tokens
Google	gemini-3.1-pro	8x	12.5M tokens
OpenAI	gpt-5.3-codex	5x	20M tokens
OpenAI	gpt-5.4	5x	20M tokens
OpenAI	gpt-5.3-codex-spark	4x	25M tokens
Anthropic	claude-opus-4-6 (kiro)	4x	25M tokens

6) Quick Troubleshooting

• Call /models first to verify your key
• Ensure Base URL is exactly https://gpt-agent.cc/v1
• Ensure key starts with sk- and has no extra spaces