Connect your tools
Raycast AI (Custom Providers)
Add an OpenAI-compatible provider in providers.yaml (requires Raycast Pro).
About 6 minFor:Mac and Windows users of Raycast
Before you start
- A Raycast Pro subscription (custom providers are Pro-only per the docs)
- A key and a model name
Step 1 · STEP 0x01
Open providers.yaml
File location
~/.config/raycast/ai/providers.yaml (copy providers.template.yaml first)
Go to Settings → AI → Models & Providers, click Reveal Providers Config under Custom Providers and copy providers.template.yaml to providers.yaml. The docs stress the file must be named exactly providers.yaml.
Step 2 · STEP 0x02
Write the provider and model
$ cd ~/.config/raycast/ai$ cp providers.template.yaml providers.yaml
providers:
- id: myrelay
name: My relay
base_url: https://api.suanlimao.shop/v1
api_keys:
myrelay: MYRELAY_API_KEY
models:
- id: <MODEL_NAME>
name: <MODEL_NAME>
provider: myrelay
context: 128000
abilities:
tools:
supported: true
vision:
supported: falsebase_urlishttps://api.suanlimao.shop/v1— the docs describe it as “an OpenAI-compatible endpoint without the /chat/completions suffix”.api_keysholds environment variable names; the key itself lives in the variable. Each provider needs at least one model withidandname.- Fill
contextandabilities(temperature, vision, system_message, tools, reasoning_effort) from the plaza; see the template file for field details.
Step 3 · STEP 0x03
Set the key and pick the model in AI Chat
export MYRELAY_API_KEY=<YOUR_API_KEY>$ export MYRELAY_API_KEY=<YOUR_API_KEY>- The variable name must match
api_keys; add it to your shell profile and restart Raycast. - Open AI Chat and pick the model under your provider.
✓ How to tell it worked
- AI Chat answers with your model and the Usage page shows the request.
FAQ
“Invalid providers.yaml”?
Hover for the error. Common causes: wrong file name, bad indentation, a provider with no models or a model missing id / name.
401?
The variable is unset or not picked up (restart Raycast after setting it); the docs suggest testing the endpoint with curl first.
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).