SillyTavern Custom API Setup (Claude, GPT)

Connect QCode to SillyTavern via the Claude reverse proxy or a custom OpenAI-compatible endpoint: field names, models, and image-generation limits (2026).

Need an API key? Sign up →·Endpoints & base URLs →

Last verified: 2026-10-20 · 📄 Per official docs (SillyTavern 1.19.0, released 2026-09-14; the GitHub Releases page still lists 1.19.0 as Latest on 2026-10-20)

Connect QCode in SillyTavern through either Chat Completion entry: pick Claude and fill Proxy Server URL and Proxy Password under Reverse Proxy, or pick Custom (OpenAI-compatible) and fill Custom Endpoint (Base URL) together with your API Key — Claude and GPT models both work for chat. The image generation extension has no custom OpenAI-compatible image endpoint, so gpt-image-2 cannot be connected.

At a glance¶

SillyTavern is a popular local LLM frontend focused on role-play and multi-turn conversations. It supports connecting to all kinds of chat models and ships with extensions for image generation, TTS, and more. This guide explains how to use QCode.cc's models in SillyTavern.

Bottom line first¶

1. Chat with QCode's Claude models¶

SillyTavern's Claude integration ships a collapsible "Reverse Proxy" section, and that is what you point at QCode:

1. Open API Connections (the plug icon) → Chat Completion.
2. Set Chat Completion Source to Claude.
3. Expand the Reverse Proxy section below it. It is not a checkbox and there is no toggle — leaving the proxy URL empty simply means "use SillyTavern's built-in official address".

4. Fill in two fields:

5. Proxy Server URL: https://api.qcode.cc/api/v1 (in mainland China, prefer https://asia.qcode.cc/api/v1)

6. Proxy Password: your QCode API Key (starts with cr_)
7. Pick claude-opus-5, claude-sonnet-5, claude-haiku-4-5-20251001 etc. in the Claude Model dropdown (4.x such as claude-sonnet-4-6 remains on sale).
8. Click Connect. Because a proxy URL is set, the first click shows a "connect to this proxy URL?" confirmation dialog — that is expected.

🔴 The proxy URL must include /v1: fill in https://api.qcode.cc/api/v1 and stop there. SillyTavern appends only /messages after it and will not add /v1 for you — https://api.qcode.cc/api would hit /api/messages and return 404. No trailing slash either (it would produce //messages). Source: SillyTavern 1.19.0 src/endpoints/backends/chat-completions.js, the Claude branch runs fetch(apiUrl + '/messages').

The official release notes for 1.19.0 (published 2026-09-14) list new Anthropic support for Claude Fable 5 and 5.1, Claude Opus 4.8 and 5, and Claude Sonnet 5 (checked 2026-10-20).

Two things to know up front:

• The Claude Model dropdown is a fixed list; you cannot type a model ID. The 1.19.0 list already contains claude-opus-5, claude-sonnet-5, claude-haiku-4-5-20251001 and claude-sonnet-4-6, which overlaps with the Claude models we sell. If we later add a Claude model that is not in SillyTavern's list, you will not be able to pick it here — wait for a SillyTavern update, or use Claude Code instead.
• What the official docs say about proxies: using a proxy you do not run yourself is a privacy risk, and SillyTavern refuses support requests from proxy users. Here the URL points to your own QCode account and key — proceed with that in mind.

2. Chat with QCode's GPT models¶

Use the OpenAI-compatible endpoint:

1. API Connections → Chat Completion.
2. Set Chat Completion Source to Custom (OpenAI-compatible).

3. Fill in:

4. Custom Endpoint (Base URL): https://api.qcode.cc/openai/v1 (in mainland China, prefer https://asia.qcode.cc/openai/v1)

5. API Key: your QCode API Key
6. Type gpt-6-sol, gpt-5.6-terra etc. into Enter a Model ID.
7. Click Connect.

Same rule here: fill in only /openai/v1 and do not add /chat/completions yourself (SillyTavern appends it). One API Key works on all the protocols; the path decides which one. See Endpoints and API formats.

The same 1.19.0 release notes also list new OpenAI support for GPT-5.6 models and GPT-6 Astra (checked 2026-10-20).

3. About image-2 generation (an honest note)¶

Many users ask: Can I connect QCode's gpt-image-2 to SillyTavern for character portraits / illustrations?

Not for now. Why:

• SillyTavern's image generation extension (formerly the Stable Diffusion extension) supports only a fixed set of backends: 24 entries in the 1.19.0 dropdown, verbatim including ComfyUI, Stable Diffusion Web UI (AUTOMATIC1111), SD.Next (vladmandic), stable-diffusion.cpp server, DrawThings HTTP API, NovelAI Diffusion, Stability AI, OpenAI, OpenRouter, TogetherAI, Pollinations, Z.AI. Only the ones you run on your own machine (SD Web UI / SD.Next / ComfyUI / stable-diffusion.cpp / DrawThings) have a URL field.
• It does not have a "custom OpenAI-compatible image endpoint" option — meaning you cannot, the way you do for chat, give image generation a custom base_url pointing at QCode's /v1/images/generations.
• There is already a community feature request #4851 (official title verbatim: [FEATURE_REQUEST] Custom openai compatible image generation enpoint (image generation extension) — the misspelling is theirs), still OPEN and not implemented as of 2026-09-18.

Note carefully: the built-in OpenAI image source does list gpt-image-2 in its model dropdown, so the name looks like it will work. But its request target is hardcoded in the source to api.openai.com/v1/images/generations with no base URL field, so it still cannot reach QCode.

Alternatives¶

If you want to use QCode's gpt-image-2, you can:

• Use the official OpenAI SDK / any tool that supports a custom images endpoint to call it directly: base https://api.qcode.cc/qcode-img/v1, model gpt-image-2, reusing the same QCode Key. See gpt-image-2 image generation.
• Track the progress of SillyTavern issue #4851; once a custom OpenAI-compatible image endpoint is supported, you can fill in QCode's image endpoint the same way as for "chat" on this page.

FAQ¶

Q: Connection returns 401 / 403? A: Check that the API Key is correct and not expired. Claude uses the "Proxy Password" field, while OpenAI-compatible uses the "API Key" field — don't put it in the wrong place.

Q: Unstable network in mainland China? A: Switch the BASE_URL domain to asia.qcode.cc (Asia node, near Korea / Taiwan / Hong Kong); if that is unstable, switch back to api.qcode.cc (global routing).

Q: Can I look up my request records? A: Yes. Requests across all domains are reported to probe.qcode.cc; enter your API Key to view them.

Want one plan that covers Claude Code, Codex, and third-party clients like this all at once? Check out QCode.cc pricing — one API Key works across all three protocols.

Enterprise →