What it does
TokenFerry provides an OpenAI-compatible endpoint at https://tokenferry.com/v1. Developers point an existing OpenAI client at that endpoint. They change the baseURL and the model field. One key then reaches Claude, GPT, DeepSeek, Qwen, Kimi, GLM, MiniMax, Hunyuan and Doubao models. Developers can declare a custom fallback queue. On rate limits, quota exhaustion or upstream errors, the gateway switches to the next model in the queue. Every response carries tracing headers. The headers record the actual model used and the routing logic. Users top up a balance and spend it across models. Per-model rates are published on the dashboard.
Capabilities
- Setting the model to tf/auto lets the gateway route each request to the optimal model based on request complexity.
- Creators can publish custom Agents to the TokenFerry marketplace and claim a 45% revenue split on every API call.
- TokenFerry says every charge carries a SHA-256 evidence chain that lets users audit any call independently.
- The example response for an Agent call includes X-TF-Agent, X-TF-Cost-USD and X-TF-Creator-Earned headers.
- Paying users get a dedicated high-speed channel with no rate limits and no concurrency caps.
When to use it
- You already use the OpenAI SDK and want to reach several model families by changing one baseURL and the model field.
- You need long-document analysis, such as summarizing a 200-page PDF with a 128K-context model.
- You need multilingual translation, such as translating East Asian and Romance languages with Qwen.
