Custom Providers
Add and manage your own OpenAI-compatible API providers
Last updated: July 6, 2026
Overview
Settings → Custom Providers lets you add and manage your own OpenAI-compatible API providers. Any endpoint that speaks the OpenAI API format works — self-hosted inference servers, gateways, and proxy services.
Custom providers require Pro or higher. On the free plan the page shows a lock with an Upgrade button.
Adding a Custom Provider
-
Go to Settings → Custom Providers and click Add Custom Provider.
-
Fill in the dialog:
- ID — a unique identifier for the provider
- Name — just a name for your reference
- API Key (Optional) — leave empty if your endpoint doesn't need one
- Base URL — the endpoint's OpenAI-compatible base URL
-
Click Add Provider.
Managing Custom Providers
Each provider is a row in an accordion showing its Base URL and ID when expanded:
- Enable/disable — the switch on the row activates or deactivates the provider without deleting it.
- Edit — reopens the dialog to change the name, API key, or base URL (the button reads Update Provider when editing).
- Delete — removes the provider after a confirmation.
Deleting a provider cannot be undone — it is permanently removed from your settings along with its configuration.
Models
Custom providers start with no models. Expand the provider and click Add Model, then enter the Model ID exactly as your endpoint expects it, plus an optional display name, description, and modalities — the same dialog as for cloud providers.
Use cases
- Local inference servers: connect LM Studio, vLLM, or a LiteLLM proxy running on your machine or LAN and use those models everywhere in Slashspace.
- Corporate gateways: route requests through your company's internal AI gateway so usage stays compliant with internal policy.
- Aggregators and alternatives: plug in any OpenRouter-style aggregator or niche hosted provider that exposes an OpenAI-compatible API.
- Testing your own models: point a provider at a fine-tuned model you're serving yourself and compare it against commercial models side by side on a canvas.