FastGPT

An open-source AI knowledge-base and customer-support platform that runs on any OpenAI-compatible model.

What FastGPT is

FastGPT is an open-source platform for building AI knowledge bases, customer-support bots and visual AI workflows. You upload your documents, and it answers questions from them. By pointing its model channel at TokenMix, you can run it on Claude, GPT or DeepSeek. There is an official cloud, so you do not have to self-host.

Before you start

• Use FastGPT cloud at cloud.fastgpt.io, or self-host it with Docker. Either way, sign in to the FastGPT dashboard.
• Sign up at tokenmix.ai, open the Console, go to API Keys, and create a key that starts with sk-tm-.

Connect TokenMix

FastGPT reaches models through a model channel. The one thing to get right: the channel's proxy address is the bare domain. Do not add /v1 here, because FastGPT appends the path itself.

1. In the FastGPT dashboard, open the model provider or channel settings and add a new channel.
2. Type: choose OpenAI.
3. Proxy address (the Base URL field): enter https://api.tokenmix.ai with no /v1.
4. API Key: paste your sk-tm- key.
5. Models: enter the TokenMix model ids you want, for example claude-sonnet-4.6.
6. Save, run the channel test, then enable those models in Model Config.

Check it works

Create a simple app or knowledge base, pick one of your TokenMix models, and ask a question. A sensible answer means FastGPT is connected to TokenMix. You can see the spend in the TokenMix Console.

Common questions

Test fails or returns 404: the proxy address must be the bare https://api.tokenmix.ai with no /v1, because FastGPT adds the rest of the path. If you instead route through the older OneAPI layer, its OPENAI_BASE_URL does need the /v1.

Cloud or self-host: the official cloud at cloud.fastgpt.io works without any install; self-hosting with Docker is optional.

Which models to pick: for knowledge-base answers, Claude Sonnet 4.6, DeepSeek V4 Pro and GPT-5.4 are good starting points.