Skip to main content
Hostwares

Is there an AI that fixes my site in production?

Yes, within limits you set: HW is an AI engineer with real access to your Hostwares account, so when a site fails it reads the logs, names the cause, and restarts, rolls back or redeploys. The industry pattern today is diagnosis first and remediation with a human in the loop, and that is how HW is built: reading status and logs needs no approval, while anything that changes a running site asks for your yes. Health checks restart a container that keeps failing without anyone asking; for everything beyond that, you tell HW once, in the dashboard, terminal or Claude Code, and it does the work and shows the evidence.

Step by step

  1. 1

    Know what runs without you

    Every site has health checks: an HTTP probe on the app's port and a process check. A container that fails several checks in a row is restarted. Set a specific health path in site settings if your app has one. This is the only unattended action; it never changes code or configuration.

  2. 2

    Ask HW what is wrong

    From the dashboard chat, the terminal or an MCP client: "my-app is returning 502, what happened?". HW calls diagnose_site, which returns database status, HTTP reachability, container state and recent logs in one shot, and explains the cause with the log line it is based on.

    Terminal

    hw ask "why is my-app returning 502?"
  3. 3

    Approve the fix

    HW proposes one action: restart_site for a wedged process, rollback_site to the previous build after a bad deploy, set_env plus redeploy_site for a missing variable, or a code change pushed and rebuilt. Each proposal shows what it will run; you approve it in the chat and HW runs it and verifies with verify_site_health afterwards.

  4. 4

    Give it a standing approval for routine work

    Over MCP, start_approval_session once at the start of a working session: one approval then covers the session's actions on your own account, invoice creation included. Deleting data or erasing a machine still returns an approval link every time, and revoke_approval_session ends it.

  5. 5

    Read the record

    Each action is recorded with what ran and when, and deployment_history lists every build. If HW got it wrong, rollback_site returns to the previous build and the record shows exactly what changed.

What it costs

  • Hosting: a Frontend pack from $1/mo; a Cloud Instance from $2/mo.
  • HW's work is billed in AI credits for what it does. Reading status and logs is a lookup and costs a fraction of a credit; a multi-step fix costs more because it does more.

App

$1/mo

Starter Frontend pack, 1 vCPU / 1 GB RAM / 10 GB

Database, users and files

$5/mo

Smallest dedicated Supabase pack; several projects fit on one

Questions

Does HW fix things without asking?+

No. Deleting, restarting, changing DNS and spending money stop and wait for your yes. Checking status and reading logs change nothing, so they need no approval and debugging stays fast. Health checks restart a failing container on their own.

Does it watch my site all the time?+

Health checks do, continuously, and restart a container that keeps failing. HW itself acts when you ask it to; a page or chat you send is what starts an investigation. Outbound alerts to a URL you choose are on the roadmap.

What can it actually do to a site?+

Restart, start, stop, roll back to the previous build, redeploy, set environment variables, attach a domain, run one shell command on the site's own instance (site_terminal), take a backup, and push a code fix to the repository.

Is it safe to give an AI access to my servers?+

HW works only on the sites, databases and domains in your account and cannot reach anyone else's. Every action is recorded. Risky actions need approval, and an approval session is scoped to one account and can be revoked.

How is this different from asking ChatGPT?+

A general chatbot gives you commands to paste. HW has the tools: it reads your real logs, runs the fix on your real site, and shows the result. The answer comes with the log line it is based on.

Where does HW run?+

In the dashboard chat, in the terminal as hw, in VS Code as the HW Dev extension, and inside Claude Code, Cursor or any MCP client. Same account, same tools, same approvals.

What does it cost?+

Hosting is a flat pack from $1/mo. HW's work is billed in AI credits for what it does; no retainer.