The best free SillyTavern proxy is a free model backend: OpenRouter's free models, Google AI Studio's free tier, AI Horde or a local model. OpenRouter listed 12 free models on November 7, 2026, capped at 50 requests a day until you buy 10 credits. Skip the shared reverse proxies posted online: their operators can read every message.
In SillyTavern, "proxy" names the server that writes your characters' replies. The fields that take it are labeled Reverse Proxy and Custom Endpoint. The other kind of proxy, the one that changes your IP address, plays no part in getting a free model. Below are the four free routes, the exact fields, and the fixes for the four errors everyone meets.
Why "proxy" means a backend here
SillyTavern's documentation is direct about it: "Since SillyTavern is only an interface, you will need access to an LLM backend to provide inference." Its API menu offers Text Completion, Chat Completion, NovelAI, AI Horde and KoboldAI Classic. Under Chat Completion, two options carry the word proxy. The Reverse Proxy block takes a Proxy Server URL and a Proxy Password, and the Custom (OpenAI-compatible) source takes a base URL and an optional key.
Both are addresses of a model server, and neither changes your IP. SillyTavern's own docs use the term once more, for a third thing: their reverse proxying guide covers HTTP servers placed in front of your own SillyTavern, and it says it "does not refer to OpenAI/Claude reverse proxies".
The "free proxies" traded in roleplay communities are servers other people run, relaying your requests on their key. SillyTavern's interface warns that such a proxy "is a risk to your data privacy" and that "ANY support requests will be REFUSED if you are using a proxy." So the best free SillyTavern proxy is a free backend you reach with your own key. Four qualify.
The free backends that hold up
OpenRouter free models
OpenRouter puts many providers behind one key and tags some models :free. Its public model list carried 12 of them on November 7, 2026, down from 18 on August 24 and 17 on September 28. They include nvidia/nemotron-3-ultra-550b-a55b:free and google/gemma-4-31b-it:free. OpenRouter also lists openrouter/free, a router it describes as one that "selects free models at random". SillyTavern offers OpenRouter as a Chat Completion source, so you paste a key and pick a model.
The catch is the cap. OpenRouter's published limits allow 20 requests a minute on free models. An account that has bought fewer than 10 credits gets 50 free requests a day; at 10 credits or more, the daily cap rises to 1,000. Free ids also come and go, as the counts above show. If a name stops working, open OpenRouter's models page and take a current one.
Google AI Studio
Google's Gemini API pricing page invites you to "Start building free of charge with generous limits". SillyTavern lists Google AI Studio as a Chat Completion source, so you paste a key and pick a Gemini model. Two caveats come from the same page. On the free tier, Google marks your "Content used to improve our products", which the paid tier does not. And some models, such as Gemini 4 Argon, are "not available on the Free tier".
AI Horde
AI Horde is the free route SillyTavern's own docs name first: "You can use AI Horde for instant out-of-the-box chatting." It is "Free · community run · open source", and its home page explains how: "Volunteers share their computers so anyone can make images and text here, for free." On November 7, 2026 at about 17:15 UTC, its status page counted 35 text workers online and 138 text requests waiting. Replies come from strangers' machines through a queue, so expect waits and treat what you send as shared.
A local model
KoboldCpp, llama.cpp and Ollama run an open-weights model on your own computer and serve it to SillyTavern. Nothing leaves your machine and there is no daily cap; the limit is the model your hardware can hold. For local inference SillyTavern recommends "a 3000-series NVIDIA graphics card with at least 6GB of VRAM". KoboldCpp answers at http://localhost:5001 by default. Ollama's OpenAI-style address is http://localhost:11434/v1, and the key it asks for is "required but ignored".
DeepSeek is the cheap option rather than a free one. Its API charges $0.15 to $0.30 per million input tokens for deepseek-flash (DeepSeek's price list, November 7, 2026), and our free DeepSeek proxy for Janitor AI guide covers it in full. In SillyTavern it goes in the Custom source with https://api.deepseek.com as the base URL and deepseek-flash as the model.
What each free backend costs you
| Backend | Free? | The catch |
|---|---|---|
| OpenRouter | yes, :free ids | 50 a day before 10 credits |
| Google AI Studio | yes, free tier | Google uses the content |
| AI Horde | yes | a volunteer queue |
| Local model | yes | needs a strong GPU |
| DeepSeek | no, cheap | $0.15 per million in |
The exact fields to fill in
OpenRouter Chat Completion > OpenRouter model: a current :free id
Google AI Studio Chat Completion > Google AI Studio model: a Gemini model on the free tier
AI Horde API: AI Horde model: one from its list
KoboldCpp API: Text Completion address: http://localhost:5001
Ollama Chat Completion > Custom address: http://localhost:11434/v1
DeepSeek Chat Completion > Custom address: https://api.deepseek.com
model: deepseek-flash
Two SillyTavern rules save most of the support threads. The Custom source takes the base URL only; its documentation says to add /v1 if the connection fails and to "NOT add the /chat/completions suffix", because SillyTavern appends it. And the model id is case sensitive, :free suffix included.
Connecting it, step by step
- Open the API Connections panel (the plug icon) and choose the API type: Chat Completion for OpenRouter, Google AI Studio and Custom sources, AI Horde or Text Completion for the others.
- Pick the source from the list.
- Paste the key from the provider. Local servers usually need none.
- For a Custom source, paste the base URL exactly as above, with nothing after the host except
/v1if needed. - Select or type the model. For OpenRouter, filter the list on
:free. - Click Connect. If a working endpoint keeps failing the automatic check, the docs offer "Bypass API status check" to stop SillyTavern from warning about it.
- Open a character and send a test message.
Treat the key like a password: anyone holding it can spend against your account.
Fixing the four errors everyone hits
401 Unauthorized
OpenRouter defines 401 as "Invalid credentials". The key is wrong, revoked or pasted with a stray space. Create a new one, paste it clean, and on a paid backend confirm the account has credit.
429 Too Many Requests
You hit a limit, almost always OpenRouter's free cap. Wait a minute, switch to another free model, or buy 10 credits once to lift the daily cap from 50 to 1,000. A network proxy changes nothing here, because the limit follows your account; our post on LLM API rate limits and proxies explains why.
No endpoints found
OpenRouter found no provider to serve the model for your request. Its errors page defines 503 as "There is no available model provider that meets your routing requirements". Usually the id is misspelled or retired, the model is down, or your privacy settings exclude every provider that offers it. OpenRouter documents which providers may capture data and lets you control it in those settings. Check the id, then the settings, then pick another :free model.
Connection error or the status check fails
The request never reached a model. For a Custom source, recheck the base URL: no /chat/completions, and try /v1. For a local server, confirm it is running and listening on the port you typed. If the URL is right, the endpoint may be down for the moment or your network may block the host.
The honeypot problem: shared reverse proxies read your chats
All over Discord and Reddit you will find ready-made "proxies" for SillyTavern: a Proxy Server URL and a Proxy Password that route through somebody else's OpenAI or Claude key. They look like the easiest free option. Avoid them.
Chub's documentation describes them plainly. Such proxies "are created by unauthorized third parties who gain access to your data and IP address. They may enable logging, leak your information, and use stolen API keys which you have no control over." The same page adds that the owners pay the costs and "could shut down access at any time". Every message and reply passes through their server, and SillyTavern refuses support to anyone using one.
Use your own key with an official endpoint, or a free tier you signed up for yourself. It keeps the chat between you and the provider.
The same trust problem exists on the network side: a free proxy of any kind puts a stranger's machine in the middle of your traffic, as are free proxies safe sets out.
If you actually needed a network proxy
None of the free backends above needs one. SillyTavern does support the IP kind for one case: its config.yaml has a requestProxy setting that "Enables the use of a proxy for outgoing requests", over HTTP or SOCKS, with a bypass list of hosts. That fits a machine on a network that sends all outbound traffic through a proxy, such as a company network. Our guide to proxies for SillyTavern covers that setup, and HProxy's residential proxies start at $0.23/GB.
If a provider does not serve your country, read its terms; we give no steps around that. For the free-backend question that brought most readers here, pick a backend from the table, use your own key and get the base URL right.
What this page could not check
We ran no model through any backend and bought no key, so every limit and label comes from each project's own documentation. AI Horde's speed depends on its volunteers, and one status reading is a snapshot. Google's per-model free limits are not quoted, because its terms page did not load from our server. MegaNova and similar hosts advertise free daily quotas on "eligible models" without naming them, so they are not on this list. OpenRouter's free ids changed three times in eleven weeks: check every name and number here by December 7, 2026.
Sources
- SillyTavern documentation: introduction, Chat Completions, config.yaml and reverse proxying, read November 7, 2026.
- SillyTavern interface source, public/index.html on the release branch, read November 7, 2026.
- Chub documentation, API connections, read November 7, 2026.
- OpenRouter documentation: API Credit & Rate Limits, Errors and Data collection; OpenRouter models API, read November 7, 2026.
- Google, Gemini Developer API pricing, read November 7, 2026.
- AI Horde, home page and status API, read November 7, 2026.
- KoboldCpp README and Ollama documentation (OpenAI compatibility), read November 7, 2026.
- DeepSeek API documentation: first call and Models & Pricing, read November 7, 2026.
