Nexus

Your machines.
Your cloud.

Scroll

Five machines. One cluster. Your cloud.

Inside

Scroll through it. Or click.

Every release, with its build log. Secrets redacted.

mac-mini:8080

Deployments

Recent deploys in this workspace.

storefront● live7c1e04b3m 12s2m ago
storefrontsuperseded4f2a91c2m 58s9d ago
assistant● live1c88f0e1m 40s14m ago
checkoutfailedb03e7aasoak failed1h ago
mailer● live377e91158s1h ago
Build log · storefront 7c1e04bsecrets redacted

→ git clone acme/storefront@7c1e04b

#6 [build 4/5] RUN pnpm build

✓ image nexus-storefront:7c1e04b (41s)

health GET /health 200 · 200 · 200 …

soak held 2m · promoting

router switched to 7c1e04b · 4f2a91c retired

AI can run your servers now. Nexus decides what it may touch, and asks you before anything it can’t take back.

Instead of

a platform

Vercel · Coolify · Railway · Kubernetes

doing it yourself

ssh root@prod · docker compose up -d · .env in Slack · crontab -e · the 3 a.m. page

Your agent, over MCP. With your OK.

Your agent

It does the work. You hold the keys.

One click from you. The rest, it did.

storefront at 700% CPU

storefront is at 700% CPU. What's going on?

Thought for 3s

deploy.inspectnexus-prodread

appId: "storefront"

live · cpu 702% · up 9 days

deploy.logsnexus-prodread

appId: "storefront", tail: 1000

1000 lines · secrets redacted

POST / 500 next-action: x ×1,204

cat: /root/.claude/.credentials.json: No such file ×142

sh: wget: not found

./.cache_rt --pool stratum+tcp://… cpu 702%

Thought for 7s

A scanner got in 9 days ago through a framework exploit.

It hunted for AI-agent keys 142 times and left a miner running.

I'll stop it, rotate what it could read, and ship the patch.

deploy.stopnexus-produpdate

appId: "storefront"

approval_required · confirm to run it

deploy.stopnexus-produpdate

appId: "storefront", confirm: true

stopped · the miner is gone

deploy.rotate_secretnexus-produpdate

key: "GATEWAY_TOKEN", confirm: true

rotated · the old value is retired

deploy.deploynexus-prodexternal

appId: "storefront"

approval_required · an operator must grant this

Shipping needs a grant from you. Issue one in Nexus and I'll go on.

You granted it. Deploying.

deploy.deploynexus-prodexternal

appId: "storefront"

grant used · deployed 7c1e04b · live

Done. storefront is patched and back. Nothing else needed you.

Reply to Claude…

+nexus-prodClaude↑

storefront › Grants

AI deploy authorizationWithout a grant, the AI's deploy.deploy stays blocked.

1 grant · single use · used

audit ai · deploy.deploy storefront · grant g-7f2 issued by you · used once

Shipping

Push. It bakes in the dark. Then it switches.

Never settles? It is thrown away. The live one never stopped.

router:443storefront4f2a91cretiredstorefront7c1e04b● live

Didn't settle? 7c1e04b is removed. 4f2a91c never stopped serving, so there's nothing to roll back.

What else

The boring parts, done.
So nobody has to remember them.

01

A database in one call.

pgvector, qdrant, neo4j, redis, minio. Bind it to an app and the connection string flows in, as a secret.

02

Set it once, for the group.

A shared value lives on the group and reaches every member. Rotate it in one place.

03

More machines, one cluster.

Install a box with one line, approve it from another. One control plane for all of them.

04

The database goes. The apps stay.

Running apps keep serving. Writes wait, a banner says why, and peers reconnect on their own.

05

Backups you can put back.

Config and Postgres on a schedule, to disk or any S3. Restore from the UI, not from a script.

06

A token per connection.

Every caller gets its own. Rotate one without touching the others.

In the works

It moves your apps.

Drain a machine and its apps land on another.

nexus

Your machines. Your cloud. Your call.