HW: the AI DevOps agent
for your production site.
HW is an AI engineer with real access to your hosting. Tell it what you want - it builds the app, puts it online, connects the domain, reads the logs and fixes what breaks in production. You approve anything risky.
Then type hw— it signs you in and deploys whatever you describe.
What HW is
HW is the Hostwares AI DevOps agent: an AI engineer with real tools on your own sites, databases, instances and domains, reachable from the dashboard chat, the hw CLI, the VS Code extension and any MCP client such as Claude Code or Cursor. It is for builders who ship with AI and do not want to become the on-call engineer: HW builds, deploys from GitHub, connects domains and SSL, and after launch reads the real logs, diagnoses a 502 or a failed deploy and applies the fix with your approval. It costs AI credits billed only for what you use with no minimums; hosting is a separate flat monthly pack. Unlike a chatbot, it runs the tool calls itself and shows them to you.
HWLiveAction Log
Full transparencyActions appear here as they complete...
Every action HW takes on your real infrastructure is logged with a timestamp, what changed, and who approved it - visible anytime from your dashboard.
Four everyday requests. Pick one to see the steps HW takes.
Detects your stack - no config files to write
One engineer, wherever you work.
The same HW, the same account and the same tools in all four places.
Say it. HW does it.
Not advice to copy and paste - the work itself, from the first line of code to the live site, with the result shown to you.
Build the app
you › build a booking page with a contact form
- Writes the code in your project
- Runs it and opens a preview for you
- Fixes the errors it finds before saying done
Change it later
you › add dark mode and a pricing section
- Reads the code that is already there
- Makes the change across the files it touches
- Checks it still builds
Put an app online
you › deploy github.com/me/storefront
- Works out the framework and how to build it
- Builds it and starts it on its own address
- Turns on HTTPS and shows you the link
Connect your domain
you › put it on mystore.com
- Adds the DNS records for you
- Issues the certificate
- Checks the domain really answers before saying done
Find out why it broke
you › why is my site returning 502?
- Reads the real logs, not a guess
- Tells you the cause in plain words
- Offers the fix and applies it when you say yes
Give it more room
you › it's slow - give it more memory
- Shows what the app is using now
- Proposes the new size
- Resizes after you approve
Real tools, not suggestions.
Every step HW takes is a call to a real tool. A few of the 95+ it can use:
Deploy
Repo to running app
Databases
Create, check, back up
Domains & SSL
DNS and certificates
Diagnose
Logs, health, metrics
Run & scale
Day-to-day operations
On your machine
Terminal and editor only
Real access only works if it can't surprise you.
HW can do a lot, so the rules about what it does without asking are short and fixed.
It asks before anything risky
Deleting, restarting, changing DNS, spending money - HW stops and waits for your yes.
Reading is instant
Checking status and reading logs changes nothing, so there is nothing to approve and debugging stays fast.
Only your own account
HW works on the sites, databases and domains in your account. It cannot reach anyone else's.
Everything is written down
Each action is recorded with what ran and when, so you can always see what changed.
Pay for what HW does. Nothing else.
Billed only for what you use - no minimums. A quick question costs a fraction of a credit; longer work costs more because it uses more.
AI Starter
Basic AI DevOps assistant
100 credits · Fast models
- ~130 messages/mo
- Fast coding & chat models
- All DevOps tools
- Email support
AI Pro
Full AI DevOps with advanced models
300 credits · Advanced models
- ~130 advanced messages/mo
- Full model range — fast + advanced reasoning
- Priority support
- Action history
AI Business
Enterprise AI automation
600 credits · Max reasoning
- ~130 Opus messages/mo
- All models incl. Opus
- 24/7 support
- Higher daily cap
AI Power
1,600 credits · Max reasoning
- ~340 Opus messages/mo
- All models incl. Opus
- Highest daily cap
- Batch processing
How HW stacks up
Against a freelancer, a locked-in platform, a general chatbot and a full-time hire.
Builds the app too
Cost per task
Time to live
Actually executes
Reads YOUR logs & fixes
Framework coverage
Vendor lock-in
Safety — approval gates
24/7 monitoring
Your domain + SSL
| Metric | Winner HW Hostwares AI • 95 tools | Other DevOps AI Koyeb / Railway AI / Qodo | General AI ChatGPT / Claude | Vibe Builders v0 / Lovable / Bolt | PaaS Lock-in Vercel / Netlify | Manual DIY Docker + Nginx + CI/CD | Freelancer Fiverr / Upwork | Full-Time Hire In-house DevOps |
|---|---|---|---|---|---|---|---|---|
| Builds the app too | Yes — writes, runs, fixes | Mostly operations | Writes code, you run it | Yes — in their sandbox | Hosting only | You write it | Yes — days | Ops, not your app |
| Cost per task | By usage, no minimum | $2-5 + platform fee | $20/mo — zero execution | $20-40/mo build only | $20/mo + overages | Your time + server | $50-500/task | $8k-15k/mo |
| Time to live | 30s — GitHub → live | 2-10 min, limited stacks | ∞ — can't deploy | N/A — stops at code | 1-3 min, their PaaS | Hours — you wire it | Hours to days | 30 min — if staffed |
| Actually executes | 95 real tools | 10-30 tools, sandbox | 0 — talk only | Build ≠ deploy | Their PaaS only | Yes — if you script it | Manual per request | Full — if senior |
| Reads YOUR logs & fixes | Yes — live logs → fix | Generic advice | Explains 502, can't read | No ops access | Your logs, no AI fix | You tail & grep | Ask, wait, pay again | Yes — if on-call |
| Framework coverage | 72+ • auto-detect | 20-50 • Node-heavy | Knows, can't deploy | Their stack only | Their stack only | Anything you wire | Depends on hire | Whatever they know |
| Vendor lock-in | None — your infra | Medium — their cloud | N/A | High — export to exit | High — proprietary | None | None | None |
| Safety — approval gates | Asks first + audit log | Often auto-runs | N/A | N/A | You handle it | You script it | Trust the human | Process dependent |
| 24/7 monitoring | Yes — included | Basic or paid add-on | No | No | Metrics only | DIY — you build it | Extra cost | If staffed |
| Your domain + SSL | Your domain, auto SSL | Their subdomain first | No | Export + manual | Their domain or paid | You configure | If you ask | Yes |
Before you connect anything
Yes. HW is an AI DevOps agent for the production site, not only a build-time helper. Once the app is live it can check the site, read the real logs and status, tell you why a 502 or a failed deploy happened, and apply the fix after you approve it - from the dashboard chat, the hw CLI, VS Code or any MCP client.
Reading never needs approval: status, logs, metrics, DNS checks and health checks run instantly, so HW can diagnose on its own and come back with the cause. Anything that changes the system - a restart, a redeploy, a resize, a DNS record, a rollback - is proposed first and runs only after your yes.
Both. In the terminal and the editor, HW writes the code in your project, runs it, shows you the result and fixes what breaks - then puts it online when you say so. In the browser chat it can change and fix apps that are already hosted here. It is one engineer for building, deploying and running, not a deploy button.
HW always asks for explicit approval before any destructive or irreversible action - deleting a resource, restarting a production service, changing DNS. Non-destructive actions (reading logs, checking status, analysing errors) run instantly since there is nothing to approve. Every action is logged with a timestamp in your Action Log, so you can always see exactly what changed and when, and roll back from the dashboard.
AI credits, billed only for what you use - no minimums. A quick question costs a fraction of a credit; longer, multi-step work costs more because it uses more. Buy credits as you go or take a monthly plan for predictable pricing. Hosting is a separate flat monthly pack.
A general chatbot can explain what a 502 means. It cannot read your actual logs, restart your actual service, or issue an actual SSL certificate - it has no access to your infrastructure. HW has real tools across deployment, databases, domains, instances, diagnostics and billing, so it does not tell you what to do; it does it and shows the tool call.
Yes. The Hostwares MCP server exposes the same tools and the same approvals to Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, ChatGPT, Lovable and any MCP client, scoped to your own account.
HW only touches infrastructure you explicitly connect - your own sites, databases, instances and domains, never anyone else's. It uses the same scoped APIs a human admin would. Combined with approval-before-change and a full audit log, you are never one ambiguous prompt away from an outage. Scope is per-user, per-resource, enforced server-side.
HW tails the build log, parses the error (out of memory, missing env var, port mismatch) and proposes the fix - often more RAM, the missing variable, or the right port. One approval to apply and redeploy. Every deploy keeps the previous release for a one-click rollback.
You do. Your domain stays yours (DNS is standard A/CNAME, not proprietary), SSL is a standard certificate, and your app runs on a standard instance. Export your env config and redeploy anywhere - no Hostwares-specific format.
Compare, guides and docs
Guides
Where it runs
Ship the thing you built.
Install HW or open it in your browser, and tell it what you want live.