Choose your trust boundary
Cloud frontier models when you want the smartest answer. Local Ollama models when the session simply cannot leave your network. Same tool, same workflow — you decide where the data goes.
Now with fully-local model support via Ollama Tell it the server. Tell it the goal.
Tell it the server. Tell it the goal.
It SSHes in and does the rest.
Agentic SSH is an AI operator for your servers. It connects through your existing SSH config, diagnoses and fixes — and when the data can’t leave the building, it runs entirely on local models. Every command is logged. Nothing runs without your rules.
OpenSSH · ProxyJump · your keys zero install on the target Claude · OpenAI · Gemini · Ollama (local)
How it works
From “something’s wrong” to “it’s fixed” — in three steps.
- 01
Point it at a target
Pick any host from your SSH config. No agent to install on the server, no ports to open. It reuses your keys, ProxyJump and jump hosts.
- 02
State the goal in plain language
“Disk is full on web-02, find what’s eating it.” It plans, runs read-only checks, and shows you exactly what it’s about to do.
- 03
Approve, and it acts
In Confirm mode it pauses before anything that writes. In Auto mode it runs end-to-end. Either way, every command lands in an auditable log.
Why Agentic SSH
An agent you can actually put in front of production.
It can’t go rogue
Three safety modes — Auto, Confirm, and Confirm-on-write. Destructive and write operations are gated by default. You see the plan before it runs.
Provable, not just promised
Every command, output and decision is written to a structured JSONL transcript you can export, diff, and hand to an auditor.
Safety model
It only goes as far as you let it.
Destructive and write operations are gated by default. Pick the mode that matches the host — and the agent shows you the exact command before it touches anything.
- Auto
- Runs end-to-end without stopping. For trusted, low-stakes hosts.
- Confirm
- Pauses before every command and waits for your approval.
- Confirm-on-write
- Read-only commands run freely; anything that writes is gated.
agentic-ssh web-02.prod — ssh via bastion · OpenSSH config
you Disk is full on web-02, find what’s eating it.
agent Planning — read-only checks first.
$ df -h /
read-only · auto
/dev/sda1 40G 39G 0 100% /
$ du -sh /var/log/* | sort -rh | head
read-only · auto
18G /var/log/app/debug.log
agent debug.log is 18G. I’d truncate it to reclaim space.
Confirm before write
$ truncate -s 0 /var/log/app/debug.log
This modifies the target. Nothing runs until you approve.
Approve Skip Edit
Every command, output and decision is appended to an exportable JSONL transcript.
Use cases
The 3am work, handled.
“Disk full at 3am”
Incident triage without typing twenty commands half-asleep.
“Why is this service flapping?”
Log correlation across hosts, summarized in plain language.
“Harden this fresh box”
Guided baseline checks before it goes anywhere near traffic.
“Audit what changed”
Export the transcript and hand it straight to compliance.
For regulated & air-gapped teams
Built for the rooms cloud AI can’t enter.
Healthcare, finance, government, manufacturing — for you, piping a production shell into someone else’s API isn’t a feature, it’s an incident. Agentic SSH runs the entire loop on local models with Ollama. No tokens leave your network. You still get the agent; the cloud gets nothing.
- Fully-local inference — sessions never leave your perimeter
- Offline license activation (ed25519) — no phone-home, air-gap friendly
- Exportable JSONL audit trail for every action
Local mode Nothing leaves
Your machine → Ollama (local) → Target host
No external API. No tokens off-network.
Cloud mode You opt in
Your machine → Your API key → Frontier model
BYOK. We never proxy or store your key or sessions.
Pricing
Start free. Upgrade when it saves you a night’s sleep.
You bring your own API key, so there’s no usage to meter. You’re paying for the software — not the tokens.
Free
For homelab & evaluation
$0
- Single target
- Local Ollama backend
- Confirm mode
- Community support
Most popular
Pro
For independent SRE & developers
$79
one-time · includes 1 year of updates
- Unlimited targets
- All backends (Claude / OpenAI / Gemini)
- Auto mode
- Audit log export
- 3 devices
Team
For teams & MSPs
from $250
per seat / year
- Everything in Pro
- Shared connection configs
- Central license management
- Priority support
Need SSO, an offline license server, or a signed liability agreement? Talk to us →
Give an AI shell access you can actually trust.
On your keys, your models, your machine. Confirm-gated, fully auditable, and free to start.
macOS Windows Linux signed & notarized on macOS