Connect your tools
FastGPT
Add a model under FastGPT's System Models, then add an OpenAI-protocol channel for it.
About 8 minFor:Admins who self-host FastGPT
Before you start
- A FastGPT admin account (recent versions ship with AI Proxy)
- A key and a model name
Step 1 · STEP 0x01
Add the model
Model type
Model provider
Model id (model)
Display name
- Go to Admin → System resources → System models and click Add model at the top right of the Model config tab.
- The model id must match the plaza exactly — channels match on it. Fill capabilities (context, function calling, vision) from the plaza.
- If a built-in entry with the same name exists, you can simply enable it.
Step 2 · STEP 0x02
Add a channel
Channel name
Protocol
Models
Proxy address
API key
- Switch to the Model channels tab and click Add channel. Set the protocol to OpenAI and tick the model you just added.
- Per the docs the proxy address must be the BaseUrl, not the full request URL — check whether /v1 is needed. For our service use
https://api.suanlimao.shop/v1. - Leave model mapping empty, or use it to expose the model under a different name in apps.
Step 3 · STEP 0x03
Test and use it in an app
Result
Each model shows pass / fail and latency
App → AI settings → Model
- Run Model test; once it passes, the model is selectable in an app's AI settings.
llmModels in config.json and OneAPI environment variables; the current docs only describe the System models route and offer a OneAPI channel import. This tutorial follows the current version.✓ How to tell it worked
- Model test passes, app chats answer and the Usage page shows the requests.
FAQ
Model test fails?
Check the proxy address is https://api.suanlimao.shop/v1 (not a full request URL), the ticked model id matches the plaza, then the key and balance.
The model is not selectable in apps?
The model must be added / enabled under Model config and ticked in a channel — both are required.
Related tutorials
Still stuck?
Read the troubleshooting tutorials first. If that does not help, open a ticket with the request_id, time, model and the exact error (never paste your key).