Bring your own model.
Pick a provider in Settings, then AI. Each chat remembers its own provider and model, and you can switch in the middle of a chat without losing the conversation.
- Your own API keyyou pay the provider, works as a guest
- OpenAI
- Anthropic
- Gemini
- Grok (xAI)
- Groq
- OpenRouter
- Amazon Bedrockus-east-1
- Z.AI
- Kimi (Moonshot)
- Meta Model API
- A connected accountyour own plan, needs a Shard account
- Codex (ChatGPT)
- Claude Code
- Shard-hosted modelscredits, needs a Shard account
- Shard: GPT-6 Lunathe hosted list can change
Three ways to connect.
Guests can use any provider that takes their own key. The other two ways need a Shard account.
- Your own API keyworks as a guest
- Paste a key from the provider’s site. You pay them directly at their normal rates.
- Codex (ChatGPT)needs a Shard account
- Already pay for ChatGPT? Press Connect ChatGPT and enter the code on OpenAI’s sign-in page. Requests use your plan’s own usage limits.
- Claude Codeneeds a Shard account
- Already pay for Claude? Run
claude setup-tokenon your computer and paste the token. Requests use your plan’s own usage limits. - Shard-hosted modelsneeds a Shard account
- No key and no subscription. Pick the Shard provider and a model whose name starts with “Shard:”. Hosted usage is paid with credits. Credit prices are not set yet.
Which one should I pick?
If you already pay for ChatGPT or Claude, connect that. If you want a specific model, bring a key. If you have neither, a Shard-hosted model needs nothing else.
| Question | Own API key | Connected account | Shard-hosted |
|---|---|---|---|
| Needs a Shard account | No | Yes | Yes |
| Who you pay for the AI | The provider | Your ChatGPT or Claude plan | Shard, in credits |
| Counts toward Shard’s limits | Yes | Yes | Yes |
| Setup | Paste a key | Sign in once | Pick a model |
| Good for | Choosing any model you like | Using a plan you already pay for | Trying Shard with no other accounts |
Worth knowing before you choose.
Small things that change what you pay or what you see in Settings.
- The model list updates itself
- Providers and models come from Shard’s online catalog, so new models can show up without a plugin update.
- Reasoning levels
- When a model can think before it answers, Settings shows a Reasoning picker. More reasoning usually helps on hard tasks but is slower and uses more tokens.
- Where your key goes
- Your key is saved in Studio’s plugin settings on your computer, and sent with each request to Shard’s server, which calls the provider for you.
- Prompt caching on Anthropic
- With Anthropic, a Prompt caching switch (on by default) lets repeated input cost less.
- Amazon Bedrock
- Uses a Bedrock API key and calls the us-east-1 region. Make sure your AWS account has access to the models you want.
- Agents use models too
- Agents use the chat’s provider and model, or another model from the same provider that suits their task.
Pick one and send a message.
You can change provider and model at any time, even in the middle of a chat.
- Limits
- The same for every way of connecting during the beta. Pricing and limits
- Your data
- What is sent where, in plain words. Privacy and data