Every service has an assistant in the panel chat. It inspects your container, creates files, runs commands and verifies the result — so a new project is a conversation rather than a manual setup session.
What it actually does
- Scaffolds a project from a description: build a Discord bot with a !ping command, create a FastAPI endpoint that reads from Postgres.
- Inspects your container first, so it works with what is actually installed rather than assuming.
- Writes files, runs installs, and adjusts the runtime configuration.
- Verifies the result before reporting success — including an HTTP check against your app's port.
- Checkpoints every file change, so any step can be rolled back.
- Requires explicit confirmation before risky actions.
That last pair is what makes it safe to use on a service you care about. You can see what changed and undo it.
Requests that work well
The assistant works from a plain description, but gets better results with specifics. Useful things to include:
- The runtime and framework you want, if you have a preference.
- What it should do — the actual behaviour, not just a name.
- Any environment variables it needs, by name.
- Which port the service listens on.
“Build a Discord bot with a !ping command and deploy it” is a good request because it names the runtime, the behaviour and the outcome. “Make it better” is not, because there is nothing to act on.
Starting from a template
Ready-to-run templates are available for the common cases — a Discord bot, a web API, a static site, a Python script. Asking for one of these is faster than describing the project from scratch, and the result starts from correct dependency manifests rather than none.
Checkpoints and rollback
Every file change is checkpointed, and rollback is available from the chat. Treat a checkpoint like a backup: free, immediate, and the reason you can experiment on a live service.
If a change turns out to be wrong, roll back rather than trying to undo it by hand. Manual reversal of an AI-made change tends to leave something behind.
Token quota
Usage is metered with a quota per plan, over a rolling three-day window:
| Plan | AI tokens |
|---|---|
| Starter | 100K |
| Basic | 250K |
| Pro | 750K |
| Max | Unlimited (fair use) |
Once the quota is exhausted, work continues when it resets. Because it is metered rather than billed, heavy use does not add cost — it only becomes unavailable until the window rolls over.
Tokens are consumed by the conversation, including its back-and-forth. A long exploratory session costs more than a focused request, so being specific is cheaper as well as faster.
Full terms are on the panel's AI services section.