Atelier · setup guide

Use Atelier for free

Atelier does not include an AI model. Connect a free service, or run one on your own computer. Provider limits change, so check each provider’s page before you rely on a number.

The provider information below was checked on 11 October 2026. Google says content sent through its free tier may be used to improve its products. Read its pricing and data-use details before sending private work.

Pick a free provider

Provider	What is free	Credit card needed	Get a key
Google AI Studio (Gemini)	Free tier with several Gemini models. Limits are shown in your account.	No	Open provider
OpenRouter	Models ending in :free, and openrouter/free. 20 requests per minute, 50 per day (1,000 per day after buying 10 credits).	Only to buy credits	Open provider
Groq	Free plan: 30 requests per minute, 1,000 per day, 8,000 tokens per minute and 200,000 per day on some models.	No	Open provider
Mistral	Free mode with a monthly allowance. Limits are shown in your account.	Not stated	Open provider
NVIDIA build	Free API endpoints for some hosted models. Limits are shown in your account.	Not stated	Open provider
Cloudflare Workers AI	10,000 Neurons per day on the free plan. Setup takes a few more steps.	Not stated	Open provider

Cerebras is left out because its one-time credit needs a payment method. GitHub Models ended on 30 July 2026. Hugging Face free accounts have no monthly credits.

No suitable graphics card and no confidential work? Start with Google AI Studio. Want another option if it reaches a limit? Add OpenRouter or Groq. For confidential work, choose a model on your own computer.

Route A

Add your own key

Choose a provider above, sign in and make an API key. Treat it like a password.
In Atelier open Settings > Model roles. On the Brain card, press Connect.
Set server type to Custom OpenAI-compatible. Enter that provider’s address, your key and a model name from the table below. Tick the image option if the model reads images.
Press Test. When it says Connected, press Save. Open a new chat, choose Vitruvius and send one message.

Provider	Address	Example model
Google AI Studio (Gemini)	`https://generativelanguage.googleapis.com/v1beta/openai`	`gemini-3.8-flash`
OpenRouter	`https://openrouter.ai/api/v1`	`openrouter/free`
Groq	`https://api.groq.com/openai/v1`	`openai/gpt-oss-120b`
Mistral	`https://api.mistral.ai/v1`	`mistral-large-latest`
NVIDIA build	`https://integrate.api.nvidia.com/v1`	`nvidia/nemotron-3-ultra-550b-a55b`
Cloudflare Workers AI	`https://api.cloudflare.com/client/v4/accounts/YOUR_ACCOUNT_ID/ai/v1`	`@cf/openai/gpt-oss-120b`

Route B

Ask your own AI assistant

Open Claude, ChatGPT, Gemini or Copilot in a new chat and paste the prompt below. Keep Atelier’s Model roles page open beside it. If you share a screenshot, hide any key. Paste your key only into Atelier, never into the chat.

Copyable setup prompt

You are helping me connect a free AI model to a desktop app called Atelier. I am an architect, not a programmer. Use plain English, short sentences, and one step at a time. Wait for me to confirm each step before the next.

RULES
1. Never ask me to paste, type or send my API key into this chat, not even part of it. I will paste the key only into the API key box in Atelier. If I paste a key here by mistake, tell me to delete it from this chat and create a new key.
2. If I send a screenshot, tell me to hide any key first.
3. Free tiers change. Do not trust your memory for limits, model names or addresses. Check the provider's own page, or ask me to send a screenshot of it.
4. Say before I sign up whether the provider needs a credit card.
5. Tell me that what I send to a cloud provider leaves my computer. If the work is confidential, suggest a model on my own computer (Ollama or LM Studio).

WHERE THE SETTING LIVES
Atelier > Settings > Model roles > the Brain card > Connect. Fields: Server type = "Custom OpenAI-compatible"; Address = the provider's base URL; API key = my key; Model name = the exact model name (after I press Test, the box suggests names); tick "This model reads images" if it does. Then Test (the card should say Connected), then Save, then one test message in a new chat.

FREE OPTIONS (checked 11 October 2026; verify first)
- Google AI Studio: key https://aistudio.google.com/apikey, address https://generativelanguage.googleapis.com/v1beta/openai, model gemini-3.8-flash. No card.
- OpenRouter: key https://openrouter.ai/settings/keys, address https://openrouter.ai/api/v1, model openrouter/free or one ending in :free (list: https://openrouter.ai/collections/free-models).
- Groq: key https://console.groq.com/keys, address https://api.groq.com/openai/v1, model openai/gpt-oss-120b. Check its current free limits.
- Mistral: key https://console.mistral.ai/api-keys, address https://api.mistral.ai/v1.
- NVIDIA build: key from a model page at https://build.nvidia.com/ ("Generate API Key"), address https://integrate.api.nvidia.com/v1.
- Cloudflare Workers AI: create an account at https://dash.cloudflare.com/sign-up, address https://api.cloudflare.com/client/v4/accounts/MY_ACCOUNT_ID/ai/v1. It needs the account ID and an API token.
- Fully local: Ollama at https://ollama.com/download (address http://localhost:11434/v1) or LM Studio at https://lmstudio.ai/ (address http://localhost:1234/v1). No key.

WHAT TO DO
1. Ask me: do I have a strong graphics card, is the project confidential, and do I already have a Google or OpenRouter account. Recommend one option and say why. Default to Google AI Studio if I have no graphics card and the work is not confidential.
2. Walk me through creating the key, one step at a time.
3. Ask me to open Atelier's Model roles page and send a screenshot with the key hidden. Tell me exactly what to type in each field.
4. After I press Test, ask for a screenshot of the card. If it says Failing, tell me the most likely fix.
5. When it says Connected, tell me to press Save and send one test message in a new chat.

Start with step 1 now.

No cloud service

Run it fully local

Nothing you send to the model leaves your computer. Install Ollama or LM Studio, then get a model that can read images. Choose that server on Atelier’s Brain card. No key is needed. llama.cpp works too. A computer without a graphics card can run a small model, more slowly.

Atelier sends no activity reports and contacts only the addresses you enter. A cloud provider receives what you send it; some providers may use it to improve their products. For confidential client work, use a model on your own computer.

Starts on as little as one 16 GB graphics card: on one card the models take turns, each loading when its job comes up; more cards or a model server keep them loaded side by side.