# Built-in AI tools

ByteHaus's customer control panel has an **AI Tools** section with **Coding Agent** and **OpenClaw**. These are separate ways to work. Customers using the built-in Coding Agent do not need to pass a documentation file to an external agent first.

This guide records panel HELP and dialogs reviewed on a Scale account on November 2, 2026. Check the customer's actual panel for availability, limits, monthly credits, and supported AI accounts. Availability on every plan, exact credit quantities, and every AI subscription tier have not been verified. The interface and access descriptions were inspected; a complete Coding Agent project build was not tested.

## Build with Coding Agent

1. Create or sign in to your ByteHaus hosting account. Add the intended domain under **My Domains → Hosted Domains**, check its document root, and point DNS to the values in the panel. See [Account, domain, and SSL setup](https://bytehaus.co/.well-known/skills/bytehaus/references/onboarding.md).
2. Turn on **SSL** in the hosted domain's settings and save. Request and activate a Let's Encrypt certificate under **My Domains → SSL Certificates**, unless a working certificate already exists. Verify HTTPS. Coding Agent cannot perform these panel setup steps for you.
3. Open **AI Tools → Coding Agent**. If connecting your own AI account, open **AI Accounts** and follow the provider sign-in and confirmation flow shown there. The reviewed panel offered Claude and ChatGPT connections; supported accounts and use are subject to each provider's current options and terms. The reviewed flow did not accept API keys. A new project also offered **Our AI (credits)** using the hosting account's allowance.
4. Create a project, choose the correct hosted domain and AI account or credit option, then describe what you want to build. The domain cannot be changed after project creation. The inspected Scale account allowed eight projects; use the live panel's limit for the customer's account.
5. For an existing site, back up files and databases first. Ask the agent to inspect the project, explain its proposed changes, and confirm before destructive work. Review its changes and test the live HTTPS site and main functions yourself.

The account connection flow may ask for an authorization code or a localhost callback URL. Treat those values as credentials. Enter them only into the authenticated panel's connection dialog. Do not paste them into public chats, tickets, screenshots, or deployment notes. The panel help says it stores an access token rather than your provider password; disconnecting an AI account removes its stored token.

### Coding Agent's access

- It works on the selected domain's project files: creating and changing sites or apps, investigating errors, and optimizing code.
- It can inspect access/error logs and run supported shell commands. Shell access does not mean root access.
- It can access databases whose credentials are present in the project. Files containing secrets can therefore expose those secrets to the connected AI provider. Keep unrelated credentials out of the project.
- It cannot manage DNS, nameservers, mailboxes, or hosting account settings. The panel help also says it cannot drop an entire database. This does not protect against all destructive file or data changes; maintain backups.
- Projects have separate histories. Multiple projects for one domain can still work on the same site's files. Do not treat separate chats as separate copies of the site or a guarantee of security isolation.
- It runs within hosting resources and needs no separately configured Supervisor process.

## Run OpenClaw

OpenClaw is a background assistant inside the hosting account, separate from Coding Agent's domain-based projects. It can use browser chat, skills, plugins, channels, and automations.

1. Review what is stored throughout the hosting account, including application configuration and database credentials. OpenClaw has the account user's access to files and databases across its sites. Use it only where that scope fits your needs.
2. Open **AI Tools → OpenClaw** and enable it. Wait for the panel to report it ready.
3. Connect a supported AI account using the options and instructions currently shown in the panel. Confirm provider eligibility and usage terms. Do not assume a subscription or API key supported elsewhere works here.
4. Open the **Control UI** from the hosting panel. Approve the browser in the panel if prompted. Control links contain access tokens; do not save or share them publicly.
5. Review Automations, Skills, Plugins, and any connected channels before use. Scheduled activity, including heartbeat and memory tasks, can consume model allowance even without a new chat message. Keep only the activity you intend to run.
6. Start with a small task, review the result, then add integrations as needed. Keep backups and obtain the customer's authorization before sending messages, deleting data, or taking other consequential actions.

OpenClaw can access files and databases throughout the hosting account, including readable configuration secrets. It does not control the hosting panel UI or account settings. It uses the account's hosting resources; model usage may incur separate charges or consume a subscription/credit allowance. Its own automations are separate from panel cron jobs, and it does not require a separately configured Supervisor process.

Disabling OpenClaw stops the service but may retain its settings, conversation data, and installed files. Disconnect AI accounts and follow the panel HELP or support instructions if you also want stored data removed. Do not delete account directories blindly.

## Choose the other path when appropriate

An existing project in a local editor can still be deployed using an external coding agent. Give it **https://bytehaus.co/SKILL.md** and let it fetch the references it needs. Native panel AI and external deployment can coexist, but coordinate changes to avoid overwriting each other's work.