Open beta. Rin Chat is under active development. Expect rough edges, and please tell us what breaks.

Skip to content

Quickstart

Which AI provider should I use?

Rin Chat does not ship a model. You connect a provider, the service or program that writes the replies, and pay it directly if it charges at all. Rin Chat takes no cut.

You can add as many as you like and switch between them any time, so the first choice is not a commitment.

Setting up for the first time? Start with Your first ten minutes

The short answer

If you wantUse
To try it for free, with no cardZ.AI. Three of its models cost nothing.
Nothing leaving your computerLM Studio or Ollama, running a model on your own graphics card.
Good roleplay without researchOpenRouter or NanoGPT with a few dollars of credit, and a popular mid-size model.
To use an API account you already haveOpenAI, DeepSeek or Grok (xAI): pick its preset and paste the key.

Two things that trip everyone up

  • A ChatGPT or Claude subscription is not an API key. API access is a separate, pay-per-use account with the provider.
  • Credit-based services need credit first. A fresh OpenRouter or NanoGPT key returns errors (a 402) until you add credit to the account. Do it before your first message.

What it costs

Hosted models charge per token, and at roleplay scale that is small: on a mid-priced model an evening of chatting is typically cents, and $5 of credit lasts most people weeks. Big premium models cost more. The provider's own pricing page has the exact numbers.

Free models cost nothing but have limits, and a local model costs nothing beyond the computer it runs on.

Free: Z.AI

Z.AI, the company behind the GLM models, offers GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash free.

  1. Sign up at z.ai/model-api.
  2. Create a Standard API key. A Coding Plan key does not work here.
  3. In Settings → LLM Providers, click + Add provider, set Type to Z.AI and paste the key.
  4. The config Rin Chat makes starts on GLM-4.7-Flash. The free models are marked Free in the model list.

GLM-4.7-Flash is the strongest of the three for chat; it thinks before answering, so replies start a little later. GLM-4.6V-Flash can see pictures. Z.AI runs one request at a time per account on most of them, and Rin Chat queues your requests so they wait their turn instead of failing.

Local: LM Studio, Ollama and KoboldCpp

A local model runs on your own computer, so nothing you write leaves it and nothing is billed. How good the model can be depends on your graphics card's memory. The Local app card described below has See which models fit this computer, a model finder that checks for you.

AppAddressBefore you connect
LM Studiohttp://localhost:1234/v1Start its local server and load a model.
Ollamahttp://localhost:11434/v1Pull a model, or let Rin Chat install Ollama and download one for you.
KoboldCpphttp://localhost:5001/v1Run it with a model, or let Rin Chat download and run it.

None of them needs a key. Pick the app as the provider Type and its page gets a Local app card that shows whether it is installed and running, and can start it for you.

"Failed to fetch" with the server running?

The server is blocking the app's requests. Turn on Enable CORS in its settings (LM Studio: Developer tab, server settings).

Hosted: OpenRouter and NanoGPT

Both put hundreds of models behind one key and one balance, so you can try a new model without signing up anywhere else.

  • OpenRouter: make an account at openrouter.ai, add credit, create a key under Keys, and pick the OpenRouter preset. Its roleplay rankings are a good place to find a model.
  • NanoGPT: pick the NanoGPT preset and paste your key. With a NanoGPT subscription, the models it includes are marked with a star in the model list and cost nothing per token. Our referral link takes 5% off your API costs and helps support Rin Chat.

Switching later

A provider is only the address and key. The model belongs to a generation config (Settings → Generation Configs), and you can make as many as you like, one per model. Making a config active switches provider, model and settings in one move, and the model name in a chat's top bar switches mid-chat.

Give a config a fallback model and a reply that hits a busy provider is retried once on the fallback.

Checking it works

The provider form's Test fetches the model list, which tells a wrong address from a wrong key. The config's Test sends a tiny message to that exact provider and model. If either fails, the quickstart's error table says what each error means.

Setting up for the first time? Start with Your first ten minutes