AI Chat

The AI assistant is built into the panel. It is not a generic chatbot: it is connected to your account, so it can look at your deployments, read your files and logs, and make changes for you when you ask.

Where to find it

• The chat drawer: open it from any page of the dashboard. It follows you around, so you can ask a question without leaving the page you are on.
• The AI assistant page (Account > AI assistant): your AI Token balance, your usage of the last 14 days, the list of tools the assistant may use, and the place to add your own provider key.

What it can do

The assistant works through the same operations the panel itself uses. Depending on what you allow, it can:

• list and inspect your deployments, read live resource usage, start, stop and restart them
• read and search the console logs
• browse, read, edit, upload and delete files in a deployment
• read and set environment variables (the value of a secret variable stays masked, always)
• add and remove packages
• list, create and restore backups
• run commands in the deployment's container
• search the Bot-Hosting documentation
• browse public community templates

It only ever reaches resources that belong to your own account.

How it decides to act

The assistant is deliberately cautious:

• it asks for confirmation before anything irreversible, such as deleting files or a deployment
• it asks before touching a deployment you did not point at
• if you only asked a question, it will not silently change something

Choosing its tools

Every tool is on by default. On the AI assistant page you can switch off the ones you would rather handle yourself. Fewer tools also means shorter, sharper and cheaper answers, because the assistant has less to consider.

Help offered automatically

On a deployment page, the panel reads the console you are already watching. When it recognises a known failure, it offers help before you have to ask: out of memory, a missing module, an invalid Discord token, disallowed intents, a Discord rate limit, a port already in use, a failed clone, a syntax error, a missing environment variable, a crash loop.

Accepting the offer sends a ready made question to the assistant. Nothing is sent, and nothing is spent, unless you accept.

What it costs

The assistant is metered in AI Tokens. See [Paying for AI Tokens](ai-tokens-payments.md) for prices and packs.

• Everyone gets a free weekly allowance that refills every Monday. It is currently 200,000 AI Tokens per week on a free account and 500,000 per week on a paid plan.
• Context the assistant has already seen counts as a tenth, so a long conversation costs far less than its size suggests.
• A short question costs a few thousand AI Tokens. A full coding task in Studio costs tens of thousands.
• Purchased AI Tokens never expire and are only used once the weekly allowance is empty.

When both are empty, the assistant pauses until the allowance refills on Monday. Nothing is ever charged automatically.

Using your own provider key

You can bring your own key from OpenRouter, OpenAI, Anthropic and others, or point the assistant at a custom endpoint.

1. Go to Account > AI assistant.
2. Add the key in the "Your own key" section.
3. Activate it.

While your own key is active, usage is billed by that provider and no AI Tokens are spent, so it is effectively unlimited. You can switch back to Bot-Hosting AI at any time. You can store several keys and the assistant falls back to the next one if a key fails.

Frequently asked

Does the assistant see my secrets? Secret environment variables stay masked, whatever the assistant is allowed to do.

Can it work on a deployment that is stopped? Yes. It can read and edit files and start the deployment when you ask it to.

Why did an answer cost more than the previous one? A conversation costs what it reads. A long thread, large files or long log extracts all add up. Starting a fresh thread for a new subject is the cheapest habit.