How to Use DeepSeek API with SillyTavern

In SillyTavern

1. Click the plug icon (API Connections) in the top bar.
2. Select Chat Completion from the API dropdown.
3. Choose OpenAI as the API type (or "Custom" / "OpenAI-compatible" depending on your version).

Fill in these fields

Important: The base URL must end with /v1. SillyTavern appends /chat/completions automatically. If you forget /v1, requests will go to the wrong path and you will get 404 Not Found.

Use exactly: deepseek-v4-flash

Do not add prefixes like openai/ or text-. Just the bare model ID.

Available models on this gateway:

• deepseek-v4-flash — Fast, great for chat/roleplay. Recommended for SillyTavern.
• deepseek-v4-pro — Higher capability, more expensive in tokens.

Click Connect or send a test message in your chat. If everything is configured correctly, you should get a response within 2-5 seconds.

401 Unauthorized

Problem: Invalid or missing API key.

Fix: Check that your sk-gateway-... key is copied correctly — no extra spaces, no missing characters. Contact your admin if the key is new and still failing.

404 Not Found

Problem: The path does not exist.

Fix: Make sure your Base URL ends with /v1. SillyTavern will call /v1/chat/completions. If your URL is just http://IP without /v1, it will call the wrong path.

402 Payment Required / Token Quota Exceeded

Problem: Your key has used all allocated tokens.

Fix: Contact your admin to add more tokens. This is a per-key quota — each key has a set number of tokens and stops when exhausted.

429 Too Many Requests

Problem: You exceeded the rate limit (default: 30 requests per minute).

Fix: Wait 60 seconds and try again. Reduce auto-retry frequency in SillyTavern settings.

5xx Server Error

Problem: The gateway or upstream DeepSeek service is temporarily unavailable.

Fix: Wait 1-2 minutes and retry. If it persists for more than 5 minutes, contact your admin.

Streaming (SSE) is supported. Characters in SillyTavern will appear token-by-token as they are generated. If streaming seems broken, check that the Streaming toggle is enabled in your API connection settings and that your character card does not have excessive prefill length.

Reasoning Content

This gateway filters reasoning_content by default. You only receive the final content field — exactly what SillyTavern expects. No stray thinking tokens in your character replies.