02
Approve before it runs
You see the exact commands before anything executes — nothing runs on your server without your one-tap approval.
00Features
Everything the AI agent does. Nothing you didn't approve.
Infraa connects to any Linux VPS with one command, then runs it through chat — installing services, watching for trouble, and proposing fixes. You stay in the loop on every write.
- ~30s
- to connect a server
- none
- root access handed over
- 0
- writes without approval
one curl command
agent runs non-root
every action signed
01Install
One command, then it's connected.
Run a single curl one-liner on any supported VPS. The agent downloads, verifies its own signature, and registers with the gateway over an outbound WebSocket — no inbound ports, no SSH keys to hand out.
- Ubuntu 22.04 / 24.04 and Debian 12 · arm64 + amd64
- Outbound-only connection — nothing to expose
- Ready and reporting metrics in about 30 seconds
perplesmile@laptop ~
$ curl install.infraa.sh | sh
→ downloading agent
→ verifying signature ✓
→ registering with gateway…
✓ connected as srv_8f3a2c
ready in 28s
Works onubuntudebianarm64amd64
02Operate
Chat your way through the work.
Ask for what you want. The AI reads live server state and drafts a plan — but it never executes. It proposes; you approve; the agent runs exactly the approved steps and logs it.
- Read-only tools inspect state before proposing
- One-tap approval — inspect raw commands first
- Full audit trail of every approved change
srv_8f3a2c · prod-api-01
live
Install Postgres 16 with nightly backups
infraa
Proposing 4 steps. You approve before anything runs.
1Install postgresql-16 (apt)
2Configure postgresql.conf + tune memory
3Schedule nightly pg_dump backups
4Run first backup + verify freshness
ApproveReview commands
03Watch
It's watching while you sleep.
The agent tracks disk, memory, CPU, and running services on a cadence. When something drifts toward trouble, it surfaces in your dashboard — before it becomes an outage.
- Per-module health checks every 60 seconds
- Disk, memory, CPU, and service signals
- Surfaced in the dashboard with context, not noise
srv_8f3a2c · monitor
live
CPU
14%
Memory
2.1 / 8 GB
Disk
62 / 80 GB
recommendation
/var/log growing 1.2 GB/day. Rotate?04Install services
A catalog of services, one click each.
Postgres, Docker, and a growing catalog install through the same propose-and-approve path. Infraa detects what's already running on a fresh server and offers to set up the rest — with sane defaults baked in.
- PostgreSQL 15/16 with a nightly backup action
- Docker CE with Buildx + Compose plugins
- First-run inventory detects what's already installed
catalog · available services
PostgreSQLdetected
15 · 16
Dockerone-click
CE · Compose
nginxone-click
+ Let's Encrypt
Redisone-click
cache · queue
05Recommend
It tells you what to do next.
Rule checks run around the clock — plus a deeper nightly AI pass — and surface recommendations: rotate a growing log, tune a config, close an exposed port. Accept one and it becomes a proposal you approve like any other.
- Rule-based checks around the clock, deeper AI pass nightly
- Accept → a normal proposal you review and approve
- Dismiss with a cooldown so it doesn't nag
srv_8f3a2c · recommendations
Rotate /var/logaccept · dismiss
Logs are growing 1.2 GB/day with no rotation configured.
Enable automatic security updatesaccept · dismiss
unattended-upgrades is available but not enabled.
Restrict Postgres to localhostaccept · dismiss
Port 5432 is listening on 0.0.0.0 — no remote clients seen.
06Bring your own AI
Your AI subscription works here too.
Infraa is an MCP server. Connect Claude Code, claude.ai, Codex CLI, Gemini CLI, or Cursor with one command and drive your servers from the agent you already pay for — the thinking runs on your existing plan, the guardrails stay Infraa's.
- Works with any MCP-capable client
- One command, then a browser OAuth sign-in
- Writes still wait for your signed approval in the dashboard
claude — ~/work
mcp
> why is prod-api-01 slow?
read_metrics(server_id: srv_8f3a2c)✓
list_processes(top_n: 10)✓
propose_capability(service.restart, postgresql)✓
proposal prop_4e12 waiting for your approvalinfraa.sh/app
Works withClaude Codeclaude.aiCodex CLIGemini CLICursor
07And the rest
The details that make it trustworthy day to day.
01
Full audit trail
Every proposal, approval, and capability call is written to Postgres. Nothing runs off the record.
03
Multi-server
Connect one VPS or fifty. Each server gets its own agent, metrics, and recommendation feed.
04
Bot & abuse protection
When an app's traffic turns mostly automated, Infraa flags it and recommends the fix — rate limits, fail2ban, CrowdSec, or Cloudflare at the edge.
05
Slack & email alerts
Route health alerts where your team already is — coming soon.
06
Live metrics
CPU, memory, and disk stream to the dashboard over the same connection the agent registered on.
07
Crash-safe execution
In-flight runs survive a gateway or agent restart — approvals resume, unspent tokens are revoked.
Ask, don't SSH.
Connect your first server and drive it from chat — or read exactly how the approval model keeps you in control.