AI models in the Hub
Section titled “AI models in the Hub”Models & keys is the Hub page where you decide which AI answers on one of your computers. It covers the agent on that computer: the model your AI helper uses, the AI accounts it may use, Claude Code, and what AI use has cost. In the Hub, a computer is one of your computers, and each computer has its own settings.
A new computer needs none of this. Its AI helper and its Bots use Hoody Free, which is included and needs no key. Open the page when you want a different model, your own AI account, or your Claude plan.
The examples below use a computer named Studio. The page always shows your computer’s own name.
Page layout
Section titled “Page layout”Open AI in the Hub, then Settings → Models & keys. Pick the computer in the top bar. When a computer is in more than one realm, the page asks you to choose one, because the default model is set per realm: Default model says Choose a realm and asks you to pick it from the realm menu at the top.
The line under the title says what answers today:
- “Studio uses Hoody Free for everything.”
- “Studio uses GPT-5.5 for everything.” when you chose another model.
- “Studio uses GPT-5.5 for chat and Hoody Free for Bots.” when the Bots’ model differs from the helper’s.
- “Studio chats with Claude Code and runs Bots on Hoody Free.” when Claude Code is in use.
A problem with the default model comes first, in a Now using band with the button that fixes it:
| The page says | Buttons |
|---|---|
| ”GPT-5.5 needs a key from OpenAI before it can answer.” | Connect OpenAI, Use Hoody Free |
| ”The default model is no longer available.” | Choose a model, Use Hoody Free |
| ”Claude Code is selected but not ready.” | Fix Claude Code |
Below that, the page has six sections on one scroll, with a list of them at the side to jump between: Default model, Accounts, Claude Code, Hoody AI allowance, Spending and Advanced. Only Advanced starts closed, behind Show advanced.
Hoody AI allowance lists each server you own with its free AI allowance, shared by every computer on it. Each server shows its Chat, Quick decisions, Web search, Speech and Transcription use, as a percent used, “Limit reached for now” or “Not on this server”. Usage and plans opens your usage and plans.
Default model
Section titled “Default model”The default model is what your AI helper and the tasks your Bots start use, unless something has chosen its own model. The section shows two lines:
- AI helper and tasks
- Bots, which says “Bots follow the default.” or “Bots use their own default.”
The default belongs to the realm you picked. Another realm on the same computer keeps its own default.
To change it:
- In Default model, choose Change on AI helper and tasks or Bots.
- Pick a model in Choose a model. The list remembers your Show and Provider filters. Without a saved Show filter, it opens on Ready when any model is ready, otherwise All. Models are grouped by provider, with Hoody AI first. To reach models from accounts you haven’t connected, set Show to Needs setup or All. Search models and providers keeps the Show filter, so with Ready it finds only models that are ready. Choosing a model that needs setup sets up its provider inside the drawer. Your own endpoint under the list adds an endpoint.
- Read the confirmation, “Use GPT-5.5 as the default?”, and choose Use this model.
The confirmation says what changes and when. It depends on the row:
- AI helper and tasks: “Your AI helper and tasks switch to this model.” Sessions that start after the save and follow the default use it. Open sessions keep their current model; switch a session’s model to change it at once. The dialog names the Bots and the agents that have their own model, which keep it.
- Bots: “Bots that follow the default switch to this model, each when its conversation restarts.” The dialog names the Bots that follow the default and the Bots with their own model, which keep it.
- When the model runs on your own key, the dialog says its use is billed to that provider account.
Each save changes only that row’s setting. Choosing Change on Bots sets a separate model for Bots. Choosing Hoody Free explicitly selects hoody-ai/hoody-free for the row you change.
The row says “Saved. This computer now uses GPT-5.5.” only after the computer reports the new model back. When the page can’t tell whether the model runs there, it says “Saved. We can’t tell yet if it runs on Studio.” Show advanced lists the model set on the agent under Helper model and Bot model.
If the answer is lost, the row says “This may have gone through. Check again before trying a second time.” Choose Check again. It reads the saved setting without sending the change again; “Not there yet. Check again in a moment.” means the change has not shown up yet. Change stays unavailable for that row while the outcome is pending. Stop tracking stops waiting but does not undo the original request.
A Bot’s own model
Section titled “A Bot’s own model”A Bot can follow the computer’s default or use a model of its own for its replies. Its Hoody-model helpers run on the Helper model set under Helpers in the Bot’s settings. If none is set, they run on their agent’s own model, else the AI helper’s model. Routing set to Pick per task overrides that choice.
To change it, open the Bot’s page, then Settings, Model, and pick a model. Agent’s default follows the default model above; the Bot’s model then reads, for example, “Hoody Free (AI page)”. A model that Needs setup needs its account connected first; choosing it offers to set up the provider.
“Saved changes keep the current model until its conversation restarts.” To switch at once, use Restart the conversation on the Bot’s page.
Connected accounts
Section titled “Connected accounts”A connected account is an AI company’s account that the agent on a computer may use: an API key you pasted, a ChatGPT sign-in, or an endpoint of your own. An account belongs to the computer, not to a realm: every Bot and helper on the computer can use it. The Accounts table has one row per provider, with Provider, How it connects, Status and Used by. How it connects reads Included with Hoody, Your key, Key set outside Hoody or Your plan. Status reads Ready or No working key, and Used by counts the Bots that use the account.
A model’s tag says who pays: Included (Hoody AI), Your key, Your plan or AI credit. Needs a key marks a model whose account isn’t connected, and Not checked one the page could not check.
Connect an account opens the providers: Needs attention, Connected, Suggested (OpenAI, Anthropic, Google Gemini, OpenRouter and Xiaomi MiMo) and All providers, then Other ways with Your own endpoint. Choose the provider: Set up on its card, or its row under All providers. For OpenAI, pick Sign in with ChatGPT or Use an API key under How to connect OpenAI. When Anthropic offers Use your Claude plan, Claude Code setup opens there; otherwise Claude Code appears under Other ways.
For OpenRouter, you can use an API key or, when the computer’s agent offers it, choose Sign in with OpenRouter. A successful sign-in says “Your OpenRouter account is signed in.”
API key
Section titled “API key”- Choose the provider, then choose Use an API key if a connection-method choice appears. For Anthropic, this switches from Use your Claude plan to the key form.
- Paste the key into OpenAI API key and choose Check and save. A current agent tests the key with the provider before it keeps it.
| Answer | What happened |
|---|---|
| ”OpenAI is connected” and “The key was checked and works.” | The key is saved and the account is connected. |
| ”OpenAI is connected” and “The key is saved and confirmed.” | The answer was lost, and the page confirmed the save by reading it back, without a test. |
| ”OpenAI key saved” and “Not checked: the check could not finish.” | The key is saved. The provider didn’t answer the test. |
| ”The key was refused. Nothing was saved.” and, under the field, “OpenAI did not accept this key.” | The provider refused the key. |
| ”OpenAI key saved” and “Not checked: this computer can’t check keys yet.” | The computer’s agent is older and saves keys without a test. |
| ”This may have gone through. Check again before trying a second time.” | The answer was lost. Check again reads what the computer has; Stop tracking this save lets you type the key again. |
“Stored on Studio only. It is never shown again.” The field clears when you press Check and save, and the Hub never shows the key again. After connecting, choose Make it the default to pick a model, or Done to close.
Anthropic connects with an API key only. To use a Claude plan, use Claude Code.
Custom endpoint
Section titled “Custom endpoint”Your own endpoint connects an AI service with its own address, such as a gateway your company runs:
- Name: how the account appears in the Hub.
- Address: a public
https://address, for examplehttps://api.example.com/v1. - Models: one model name per line, as the endpoint names it (1 to 500).
- Choose Add endpoint.
- On the next step, paste the endpoint’s key and choose Check and save.
The endpoint is expected to work like OpenAI. When it speaks another protocol, choose it under Advanced, Protocol: OpenAI chat (default), Anthropic messages or OpenAI responses. The agent connects only to public addresses.
When the endpoint is added but its key fails, its row says No working key: use Replace key or Disconnect in the row’s menu. Edit changes the name, the address or the models, then Save. A new address asks first: “Your saved key will be sent to the new address. Tick the box to allow it.”, with Send my saved key to the new address.
ChatGPT sign-in
Section titled “ChatGPT sign-in”With a ChatGPT plan, OpenAI models run on your plan instead of an API key.
- Choose OpenAI, then pick Sign in with ChatGPT under How to connect OpenAI.
- The Hub shows a code. Choose Open the sign-in page, or open the ChatGPT sign-in page on any device, and enter the code. The code works for 15 minutes.
- When you finish, the page says “OpenAI is connected” and “Your ChatGPT plan is signed in.”
When both a ChatGPT sign-in and an API key are stored, the agent’s credential preference decides which it tries first. Change that preference in the agent’s Models tab; choosing a connection method in the Hub does not change it. A plan serves only the models it includes. The ChatGPT sign-in does not cover image generation or a provider’s web search: those tools use an API key. The default web search runs on Hoody AI and needs no key.
Closing the drawer, or Cancel sign-in, cancels the sign-in. A sign-in you finished before the cancel arrived is kept. If the page can’t tell, the step says “The sign-in may have gone through”. When you left before the cancel was confirmed, a notice says “The ChatGPT sign-in may have gone through” with Check again. On a computer whose agent is older, sign in from the agent’s TUI on that computer instead. To add another ChatGPT account, open the agent’s TUI, go to Models, and use + Add account in the provider’s account list. To sign out, choose Disconnect in the OpenAI row’s menu while How it connects reads Your plan. When it reads Your key, Disconnect removes the API key instead. To remove the ChatGPT sign-in directly, send DELETE /api/v1/agent/providers/openai/auth/oauth.
Account actions
Section titled “Account actions”The actions in a row’s menu depend on how the account is connected:
| Account | Actions |
|---|---|
| Hoody AI (Included with Hoody) | No menu |
| A key saved in the Hub | Replace key, Test, Disconnect |
| A ChatGPT sign-in | Disconnect |
| A key set outside Hoody | Test. Change or remove that key where it was set on the computer. |
| Your own endpoint | Edit, Replace key, Test, Disconnect |
What each action does:
- Replace key opens “Replace the key for OpenAI”. Replies already running finish with the old key; the new key applies from each conversation’s next turn.
- Test checks the key saved through Hoody and shows Key works, Key refused or Could not tell. It cannot test a key set outside Hoody; without a saved key, it asks you to connect the provider first.
- Disconnect asks first: “Remove the key for OpenAI?” with Remove key, “Sign out of OpenAI?” with Sign out, or, for your own endpoint, Delete endpoint. The dialog says “Anything using this account stops working. Nothing switches to another AI by itself.” and lists what uses the account: your AI helper, the number of Bots, and the agents with their own model. When another credential for the same provider is still stored, such as a key set outside Hoody or the other of a key and a sign-in, models on that provider keep working with it. Deleting an endpoint makes its models unavailable.
Claude Code
Section titled “Claude Code”Claude Code lets you use your Claude plan in the Hub. The Claude Code program on the computer signs in to your Claude account itself, with its own sign-in. The Hub never asks for a Claude password or token.
The card shows where the computer stands:
| Claude Code on this computer says | Button |
|---|---|
| ”Claude Code is not installed” | Open a terminal to install |
| ”Claude Code needs an update” | Open a terminal to update |
| ”Not signed in to Claude Code” | Open a terminal to sign in |
| ”Sign-in could not be confirmed” | Open a terminal to sign in |
| ”Signed in to Claude Code” | Use Claude Code |
| ”Not checked yet” or “The check did not finish” | Check again |
| ”Update the computer’s agent to use Claude Code" | |
| "Claude Code is not available on this computer” |
“Signed in to Claude Code” is as far as the computer can tell before Claude answers. Once Claude confirms the sign-in, it reads “Signed in to Claude Code and verified”.
Setup in a terminal
Section titled “Setup in a terminal”Open a terminal to install, Open a terminal to update and Open a terminal to sign in open a setup terminal on the computer below the card, which you can resize from its bottom edge, and run the command immediately. Follow any interactive steps there. The card checks progress while setup runs, for up to 10 minutes; after that, choose Check again.
The terminal opens under Setting up Claude Code; Open in a new tab opens it on its own. After installing, the card shows the next step: Open a terminal to sign in runs the sign-in command. Follow Claude’s sign-in steps there; the page sees when you are signed in.
Check which account the terminal runs as before you complete the Claude sign-in.
Manual setup
Section titled “Manual setup”When the agent offers a supported standard setup command, Do it myself shows the command (Run the command yourself) in a Command field with Copy. Run it in any terminal on the computer:
# Install or update Claude Codesudo npm install -g @anthropic-ai/claude-code
# Sign in with your Claude accountclaude auth loginThe setup buttons and Do it myself appear only when the agent’s setup command matches the standard one. If they are missing for a custom installation, update Claude Code and sign in the way it was installed there.
Other computers
Section titled “Other computers”Sign-in on other computers (Copy sign-in…) copies Claude Code’s sign-in from a computer where it works to another computer of this realm. Settings on other computers (Copy settings…) copies Claude Code’s settings, agents, commands and skills. Neither copies chats. You see what changes before anything is written.
Effect on new conversations
Section titled “Effect on new conversations”Use Claude Code (“New chats on this computer will use your Claude plan.”) selects Claude Code for new Hub conversations in the realm. If the card then says “Selected, but Claude Code is turned off”, choose Turn it on. Once it runs, the card reads “New chats use Claude Code”. Conversations already open keep their AI. The Agent app, Bots, Explain and script writing keep using the models you chose.
A Bot’s own replies use its selected chat model. Its tasks can run in Claude Code when the Bot allows Claude Code helpers. Those helper settings are separate from the backend selected for new Hub conversations.
The card says the selection is in use only while Claude Code is signed in, turned on, selected and able to start. Otherwise it says what is missing: “Selected, but Claude Code is not ready yet”, “Selected, but Claude Code is turned off” with Turn it on, or “Selected, but Claude Code could not start” with Check again. Go back to my models undoes the selection.
In a new conversation’s options, Runs with offers Claude Code only while it is ready. Otherwise the choice is unavailable and the row says “Claude Code is not ready on Studio.” with a Set up Claude Code link to this section.
Guardrail checker
Section titled “Guardrail checker”A Bot with guardrails checks each new task against them before it starts, using a checker model in the Bot’s realm. A Bot in strict mode refuses every task while that checker can’t run; in lenient mode the task goes ahead unchecked. A Bot whose guardrail_check is off skips the check.
To see whether the checker can run, choose Show advanced. The Safety checker row says “On and ready.”, “Turned off.” or “Could not be checked.”, and its switch turns checking on or off. A line under it counts the Bots it guards, for example “3 Bots have guardrails the checker enforces.” Checker model picks the model; Hoody safety checker is Hoody Decide and needs no key. The picker appears once more than one checker model is usable. A computer whose agent is older says “This computer’s agent is too old to show the checker for a realm.”
Spending and AI credit
Section titled “Spending and AI credit”The Spending section shows recorded use for the selected period, next to your AI credit: 7 days, 30 days or All time. The default is 7 days, and a period selected in the URL survives a reload, so read the selected period before you read the total. The table lists one line per model with Model, Provider, Calls, Success and Recorded cost. A model with no recorded cost reads “No cost recorded”. The figures cover every realm on that computer, under a total such as “Recorded total $1.20. Your provider’s own bill is the final word.” Your AI credit shows next to the period, for example “$4.20 credit left”.
AI credit is your account’s, shared by all your computers. Adding Hoody AI credits is coming soon: Add AI credit, under Hoody AI allowance, is marked Coming soon. The free models work within your servers’ free limits.
Advanced
Section titled “Advanced”Advanced, behind Show advanced, keeps the details most people never need:
- Helper model and Bot model: a model set on the agent itself, as the agent’s API and CLI name it, or “Follows the default”. Hoody Free is
hoody-ai/hoody-free. - Rows such as “reviewer uses its own model”: other agents in the selected realm whose explicit model differs from the AI helper’s current model.
- Ensembles: an ensemble asks several models and combines their answers, and Bots can use one. The Hub shows them read only; change them in the agent’s TUI.
- Safety checker and Checker model: the guardrail checker for the realm.
- Account ids: “The names these accounts have on the computer.”
Access to these settings
Section titled “Access to these settings”Whoever can open a computer in the Hub can change its AI models and accounts, and every Bot and helper on that computer can use an account connected there. To limit who can reach a computer’s agent, set its proxy permissions.
Some computers are set up to accept AI account changes only from the agent’s TUI. On such a computer the page refuses the change with “Only an admin can do this.”; make it in the TUI instead.
The same settings over HTTP
Section titled “The same settings over HTTP”Hoody AI allowance and credit are account-level reads through the Hoody API. The computer’s model, account, Claude Code and recorded-use settings go through its agent’s HTTP API, documented in the API reference:
- providers, keys, ChatGPT sign-in, your own endpoints and models: Agent: Models & Providers;
- per-agent models: Agents;
- Claude Code status and selection: Settings;
- recorded use: Statistics.
The agent’s own TUI manages providers, keys and sign-ins in its Models tab; see Models and providers.
Troubleshooting
Section titled “Troubleshooting”| Message | What to do |
|---|---|
| ”Could not load the AI settings” | The computer didn’t answer. Check that it is running, then choose Try again. |
| ”OpenAI did not accept this key.” | Check on the provider’s page that the key is active, copy it again and paste it. |
| ”This may have gone through. Check again before trying a second time.” | Choose Check again. After Stop tracking this save, you can enter the key again. Saving it does not change the agent’s credential preference: a sign-in or a key set outside Hoody may still be used. |
| ”Only an admin can do this.” | The computer accepts AI account changes only from the agent’s TUI. Open the agent’s TUI on that computer and change the account there. |
| ”You do not have access to this computer’s AI settings.” | Your access to this computer doesn’t allow the change. See Access to these settings. |
| ”That name is taken. Choose another.” | Another account on the computer uses that name. Give your endpoint a different name. |
| ”The code ran out. Start again for a new one.” | The ChatGPT code works for 15 minutes. Start the sign-in again. |
| In a new conversation’s options, Runs with says “Claude Code is not ready on Studio.” | Choose Set up Claude Code, then follow the button the Claude Code section shows. |
| ”Not signed in to Claude Code” | Choose Open a terminal to sign in in the Claude Code section. |
| Safety checker says “Turned off.” or “Could not be checked.” | Checking is off, or the checker’s model has no key. Under Show advanced, turn Safety checker on and set Checker model to Hoody safety checker. |
| ”That model is no longer available.” | The account or endpoint behind it was removed. Choose another model. |