Features

Brand-new UI (some screens are still being updated)

Multi-language support (work in progress)

Supports image generation APIs such as Midjourney-Proxy (Plus). Make sure you comply with the third-party service's licensing, content safety requirements and terms of use.

Internal balances, cost allocation or enterprise customer billing in lawfully authorized deployments

• ☑️ EPay

Query usage or balance information for lawfully authorized channels

• Works with API keys in the EasyAPI console, so you can query usage with an API key

Choose how many items to show per page

SQLite database storage: lightweight and ready to use out of the box

Internal cost accounting or enterprise customer billing metering (configure it in System settings → Operation)

Weighted random channel selection

Data dashboard (console)

Restrict which models an API key can call

Telegram sign-in:

1. System settings → Login & registration → Allow Telegram login
2. Send the /setdomain command to @Botfather
3. Select your bot, then enter http(s)://your-site-address/login
4. The Telegram bot name is the bot's username without the leading @

Suno API support (before use, confirm the licensing, content safety and terms of service requirements of the third-party project and the upstream service)

Rerank models, currently compatible with Cohere and Jina, and usable with Dify

OpenAI Realtime API — supports OpenAI's Realtime API, including Azure channels

Set reasoning effort with a model name suffix:

• OpenAI o-series models: add the -high suffix for high (e.g. o3-mini-high)
• Add the -medium suffix for medium (e.g. o3-mini-medium)
• Add the -low suffix for low (e.g. o3-mini-low)

Claude thinking models: append the -thinking suffix to the model name to enable thinking mode, e.g. claude-3-7-sonnet-20250219-thinking

Thinking to content: the thinking_to_content option (default false) is available under “Channels → Edit → Extra channel settings”. When enabled, the reasoning_content returned by the model is converted into text wrapped in <think> tags and appended to the response content.

Model rate limiting: set rate limits per model under “System settings → Rate limits”. You can limit both total requests and successful requests.

Cache billing: when enabled, cache hits are billed at a specified ratio.

1. Configure the Prompt Cache Multiplier under “System settings → Operation”
2. Set the ratio (0–1) on the corresponding channel; for example, 0.5 bills cache hits at 50%

• ☑️ OpenAI
• ☑️ Azure
• ☑️ DeepSeek
• ☐ Claude