Quickstart

Flux is a unified interface for large language models: call mainstream models across various clients through a unified API key.

For AGENTS

For bulk Markdown reads, use markdown-sitemap.xml. For a single page, append .md to any documentation URL that does not end with /, for example /en/docs/quickstart.md.

Getting Started

1. Choose a client below. In the model marketplace, confirm that the target model's group supports the required protocol and client.
2. Read Billing, check prices, then top up your balance.
3. Create an API key on the group you selected.
4. Follow the client guide below to enter the key, endpoint, and model ID.
5. Send a message and check the reply and usage log to confirm the integration.

For groups, account pools, balance-based billing, and other terms, see Core Concepts.

Client Options

Best for terminal workflows, programming tasks, or users who want AI Agent assistance while developing.

Follow the guide for your client. CC-Switch is an optional way to manage provider settings for clients such as Claude Code and Codex; it is not a prerequisite.

Client guide	Best for
Claude Code	Anthropic's official CLI, terminal coding
Codex	OpenAI's official CLI and desktop app
Codex++	Adds extra capabilities on top of Codex
OpenCode	Open-source terminal agent with model switching
Pi	Extensible terminal coding agent
Hermes	Lightweight terminal agent

Best for users who prefer graphical interfaces, mobile usage, or daily conversational workflows.

Client guide	Best for
Cherry Studio	Desktop (Windows / macOS / Linux), multi-model switching
RikkaHub	Android, OpenAI compatible

Add Flux as a custom provider in the client settings, enter your API key and an API endpoint, fetch the model list, and start chatting.

Confirm the Integration

1. Select the target model in your client, start a new conversation, and send "Reply with OK only". This incurs model call charges.
2. Confirm that you receive model-generated text, not just a model list or a "configuration saved" message.
3. Find the request in the usage logs, check the key, model, and charge, and confirm the group matches your key configuration.

This only confirms that this text conversation worked. Test tool calls, images, or other capabilities separately if needed. If there is no reply or the usage record differs from your expectations, see Troubleshooting.

Going Further

After integration, the following may be useful:

• Creative Studio - generate images directly on the web without client or API configuration
• Composite Key - bind one key to several groups and switch with prefixes
• Fast Mode - force the high-priority tier at the key level
• Team - share the owner's balance while everyone keeps their own keys
• Referral Rewards - earn rebate rewards by inviting friends

Troubleshooting

• Troubleshooting - find the problem by symptom, with a report template
• Error Codes - the full error reference
• FAQ - model detection and other common questions
• API Endpoints - choosing an endpoint and migrating from deprecated routes
• Usage Policy - what gets accounts banned