agent-native cloud control plane

Humans and AI agents —
run production with the same one line.

A single Go binary that brings Fly.io-grade DX to GCP. Deploy, infra, env, cost, and diagnostics in one command. Guardrails built in.

Install $ curl -fsSL devops.oqmenu.ai/install | sh
what a human sees
$ oq deploy api --env prod
  ✓ image    oq-api:20260531 (digest verified)
  ✓ revision 00312 ready
  ✓ traffic  ████████████ 100% → 00312
  ✓ health   /health 200
  ✓ audit    logged
  Deployed in 41s. ← auto-rollback on failure.
what an agent reads
harness → oq deploy api --env prod --json
{
  "service": "api", "env": "prod",
  "new_revision": "00312",
  "traffic_percent": 100,
  "health": "200",
  "result": "success"
}
exit 0  // typed exit code
Green checks for humans. JSON for agents. Same one command, same safety.
The Pain

Why is deploying straight to the cloud
still this hard?

Hand-edit an image tag → push → wait on a CI runner. Bootstrap infra by running scripts in order, by hand. Memorize gcloud incantations. The deploy says "success" while traffic stays on the old revision. You ship without knowing what will change. So you end up needing a dedicated DevOps specialist.

oq removes the need for that role. Developers and AI agents both deploy safely with one command.

Now · Next · Vision

Works today. More soon. This is the future.

NOW ✓
Proven reality
  • 833 tests passing
  • Production-ready 90/100
  • 13 live services
  • deploy · plan · cost · diagnose
NEXT →
Coming up
  • Agent deploy (--json)
  • Traffic-shift verification
  • Self-heal loop
VISION ★
Declared goals
  • End of the DevOps role
  • Agent-native
  • Self-evolving infra
  • One interface, every cloud
Proof

Not claims. Numbers.

833
tests passing
90/100
production-readiness score
13
live services
~40
CLI commands

Every package builds, vets, and tests green. gcloud, Cloud SQL, GLB, Secret Manager, and Cloud Build are already production-grade on real API calls. This site was deployed by oq itself.

Use cases

From in-house operations to a general-purpose product.

▌In-house — running oqoq.ai

  • Multi-service DAG deploy/rollback across 13 services
  • Festival burst scaling / stress testing
  • Repo ↔ central env single-source-of-truth sync
  • Per-service TCO cost attribution (shared split)
  • kakao-bridge / redroid specialized automation

▌General — any GCP team

  • Fly-like one-command deploy (oq deploy)
  • mini-terraform plan / apply
  • env/secret management + drift detection
  • TCO monitoring + anomaly alerts
  • Natural-language oq ask · typed exit codes
How it works

Four layers, one binary.

Config Model
Type-safe config loading. Typos and missing fields error at load time — no silent failures.
Resource Abstraction
Describe / Diff / Apply. Desired vs current state compared consistently, idempotency guaranteed.
GCloud Runner
Unified execution, timeouts, logging, dry-run. Every infra command in one pattern.
Orchestrator
DAG pipeline. Dependencies resolved automatically, parallel execution, rollback plan.
Agent-native

Not a CLI for humans —
a control plane for humans and agents alike.

Designed so AI agents can invoke deploys directly: non-interactive mode, structured JSON output, typed exit codes. A harness decides success or failure mechanically.

agent → oq
$ oq deploy api --env prod --non-interactive --json
{ "result": "success", "traffic_percent": 100,
  "health": "200", "audit_ref": "2026-05/deploy.jsonl#41" }
exit 0   # 0 success · 2 prod-gate · others typed

Autonomy is only as smart as its guardrails. Prod gate, traffic-shift verification, auto rollback, and audit logs are why you can hand an AI the prod keys.

Roadmap

Eight bets — where we're headed.

1 · End of the DevOps role
Every developer their own SRE
2 · Agent-native control plane
AI as a first-class operator
3 · Self-evolving infra
Fixes itself while you sleep
4 · Natural language as interface
oq ask "faster and cheaper"
5 · Safety unlocks autonomy
Deterministic guardrails
6 · One mental model, every cloud
GCP · AWS · K8s
7 · FinOps autopilot
Just set the budget
8 · Verifiable operations
A credit score for infra

Install in one line. Look around in a minute.

The one control plane you can hand the prod keys to an AI with.

Install $ curl -fsSL devops.oqmenu.ai/install | sh