DOC / 04 User-defined endpoint
OpenAI-compatible.Your server, explicit destination.
Connect LM Studio, vLLM, LocalAI, or another compatible service without silently moving a saved credential to a new endpoint.
01 Setup
Connect OpenAI-compatible without guesswork.
Make these changes in the installed Palshell desktop application. This website never collects provider credentials.
- 01Start a compatible server
Run LM Studio, vLLM, LocalAI, or another service that implements the required OpenAI-compatible routes.
- 02Enter the destination
Open Settings → AI Providers, select OpenAI-compatible, and enter the server’s Base URL.
- 03Add a key only when required
A key is optional for local endpoints. When the server requires one, Palshell sends it as a Bearer token. Never put the key in the URL.
- 04Choose the exact model
Select Discover models. If the result is empty or unhelpful, enter the server’s exact case-sensitive model ID in the Model field.
- 05Save and verify
Select Save, then Test connection. Optionally select Set default.
02 Exact configuration
Use these values. Keep credentials out of URLs.
http://localhost:1234/v1POST /v1/chat/completionsModel discoveryGET /v1/models03 API cost
The endpoint owner decides the cost.
OpenAI-compatible describes an API shape, not one billing system. Palshell does not add a per-token fee or markup.
- Palshell provider surcharge
- $0
- Local self-hosted server
- Usually no API bill
- Hosted compatible service
- Provider-defined
Palshell passes requests to the endpoint you configured.
You still supply the computer, GPU or CPU time, memory, and electricity.
Check that service’s token, request, subscription, or infrastructure rates.
There is no honest universal calculator for this category. Use the configured server’s own rate card. If it bills per token, apply its input and output rates using the same per-million-token formula shown in the hosted-provider guides.
The destination stays visible. The secret stays native.
Plain HTTP is permitted only for loopback addresses. Non-loopback and remote endpoints must use HTTPS, and URLs cannot contain usernames, passwords, keys, or tokens. When the endpoint changes, Palshell requires the key again because the credential is cryptographically bound to the provider identity and normalized endpoint.
04 Troubleshooting
If OpenAI-compatible does not connect, start here.
- 01The required routes are missing
- Confirm the server implements GET /v1/models and POST /v1/chat/completions with compatible response shapes. Otherwise use a compatible server or adapter.
- 02A remote endpoint uses HTTP
- Move the service behind a properly configured HTTPS endpoint. Palshell permits plain HTTP only for loopback addresses on the same computer.
- 03The endpoint changed
- Enter the API key again, even if the same server uses it. The old credential binding cannot silently move to another destination.
- 04Discovery returns no useful model
- Enter the exact case-sensitive model ID documented by the server, then select Save and Test connection.
05 Official references