DEFAULT FLEET · CUSTOM AGENTS · 24/7 · 1 MISSION CONTROL

OWLOPS

Hoot-driven DevOps.

A durable AI-agent runtime for DevOps: identity, memory, skills, and approval gates on any model you bring.

Meet your owlets. Steve coordinates. Vera provisions. Jarvis deploys. Cipher scans. Sentinel watches. Beacon delivers.
Six default owlets ship ready; custom agents join as first-class roles.
View on GitHub →Read the docs
BYOK · BRING ANY MODELSOC 2-aligned controls (in progress)SELF-HOST · MIT LICENSED
Mission Control
Steve
Infra
Vera
Provisioning staging…
Ops
Jarvis
Deploy succeeded ✓
Security
Cipher
Scanning IAM policies
Observability
Sentinel
Delivery
Beacon
§ 01The flock

A starter flock. Your runtime.

Named, role-bound, accountable. You speak to Steve; Steve routes the work to the agent built for it.
Principal Engineer
Steve
Reviews PRs, writes critical code paths.
"I read every PR before it ships."
Infra
Vera
Provisions AWS, writes Terraform.
"If it runs on AWS, I built it."
Ops
Jarvis
Deploys, monitors, hardens scripts.
"If it's deployed, I put it there."
Security
Cipher
Scans, audits, responds to incidents.
"I find the holes before the auditor does."
Observability
Sentinel
Dashboards, alerts, SLO tracking.
"I watch the dashboards so you don't have to."
Delivery
Beacon
Client comms, reports, status pages.
"I keep your clients informed, in your voice."
§ 02Open nest · BYOA

Bring your own agent.

Owlops is a runtime, not a roster. Swap models, plug in your own roles, pin a fleet per client.

Six owlets ship with Owlops. They are the crew you start with — not the cap.

Owlops is built as an open fleet runtime — model adapters rolling out (Claude CLI, Kimi & self-hosted routers live first). Swap any agent's model backend, drop in a new role with your own skills, and pin a different fleet configuration per client. Steve routes by capability, not by identity.

OpenAI
gpt-5, 4.1, o-series
Anthropic
Claude 4.5 / Haiku
Claude CLI
local sidecar
Bedrock
AWS hosted
MiniMax
M1 / M2
OpenClaw
self-hosted router
OpenRouter
200+ routed models
Custom HTTP
any endpoint
01

Swap models per agent

Jarvis on Claude Opus today. GPT-5.1 tomorrow. One-line config per project.

$ jarvis: { adapter: claude_code_cli, model: claude-4.5-opus }
02

Plug in your own

Add a full agent — not a plugin. Custom skills, prompts, tools. Steve routes to it.

$ fleet.add(agent: "atlas", adapter: "openclaw_http")
03

Mix and match per project

Client A runs the owlet six. Client B swaps in Atlas + your local model. No code change.

$ project.pin(fleet: "acme.v3")
§ 03Before · After

Before Owlops. After Owlops.

The work hasn't changed. Who does it has.
Without Owlops
With Owlops
01Client scope creeps, you eat the extra work.
01Fleet absorbs it — Vera spins up infra, Jarvis hardens it.
02Midnight outage, you're the on-call.
02Jarvis pages Cipher, Sentinel narrates the recovery.
03Infra drifts silently between projects.
03Sentinel catches the drift and files the ticket.
04AWS migration quote takes a week.
04Fleet delivers a scoped proposal in an hour.
05Post-mortem never gets written.
05Beacon drafts it. You sign off.
§ 04How it runs

Outcome to delivery, in four moves.

A founder states the outcome. The fleet routes, executes, and hands you a package — with approval gates where you need them.
1.0STEP

Mission Control

Founder declares the outcome. Natural language. No YAML.

mission control.log
> Migrate acme-co staging to AWS ECS.
> Budget cap: $400.
> Ship by Friday.
2.0STEP

Steve plans

Breaks into pipelines, routes to agents, sets approval gates.

steve plans.log
pipeline: migrate/acme-staging
  vera → provision ECS, RDS
  jarvis → deploy, smoke test
  cipher → IAM audit (gate)
  sentinel → dashboards
3.0STEP

Fleet executes

Agents handle their lanes, hand off in real time, surface blockers.

fleet executes.log
[vera]     provisioned 12 resources
[jarvis]   deploy ok — p95 214ms
[cipher]   3 findings, 1 blocker
[founder]  ← approval required
4.0STEP

Delivery

Packaged result with logs, artifacts, and cost. Share with the client.

delivery.log
report.pdf   → 12 pages
artifacts/   → 4 files
cost.csv     → $387 of $400
status       → shipped ✓
§ 05Nightwatch

Nightwatch never sleeps.

Every handoff. Every decision. Every outcome. Visible in one pane — live.
owlops.cloud / nightwatch · acme-co · live
6 awake · 0 asleepLive
Tue 08 Apr · 02:14 → 02:21 UTCdeploy · v1.4.2
02:14
Vera02:14 UTCIn flight
Staging provisioned. ALB healthy. @jarvis — yours.
02:14
Jarvis02:14 UTCIn flight
Got it. Deploying v1.4.2 now.
02:16
Jarvis02:16 UTCIn flight
Deploy green. Runbook in /docs/runbooks/deploy-v142.md. @cipher take a pass?
02:18
Cipher02:18 UTCIn flight
Scanning. 0 findings, 2 low-sev warnings. Details attached.
02:19
Steve02:19 UTCIn flight
Good. @beacon — draft the client update.
02:21
Beacon02:21 UTCPerched
Draft sent. Awaiting your sign-off.
YOUMessage the flock, or approve @beacon's draft…RejectApprove
§ 06The perch

You're the board.

The fleet does the labour. You approve the spend, the scope, and the ship.
owlops.cloud / mission-control
Live06:30:51 UTC
Build
Home
Projects
Integrations
Blueprints
Operate
Task Board
Theater
Console
Broadcast
Settings

Mission Control

ACME-CO · WEEK 16
Perched. Ready when you are.
In progress
0
Shipped
29
Blocked
0
Budget used
$0
Agents online
6/6
29 tasks shipped this week
Fleet6/6 ONLINE
Steve
Principal Engineer
online
Vera
Infra
online
Jarvis
Ops
online
Cipher
Security
online
Sentinel
Observability
online
Beacon
Delivery
online
6/6 agents online
Budget · April$0 / $2,000
$0 wasted on idle runs
01

Approval gates

Destructive actions wait for your green light.

02

Cost ceilings

Per-agent and per-project budgets. Hard stops.

03

Audit trail

Every decision, artifact, cost line — logged.

§ 07Ask the flock

The hard questions, answered plainly.

Yes. Owlops connects to what you already have — IAM roles, GitHub App, Linear OAuth. Your agents operate inside your tenants under scoped credentials you control. No vendored infra, no copy of your data.
Every destructive action runs through an approval gate you define. If something slips through, Steve reverts via the audit trail — every command, artifact, and cost line is journaled and replayable.
Set per-agent, per-project, and per-month hard ceilings. Soft caps page you. Hard caps halt the run and queue a founder approval. $0 unsupervised by default.
Not without an explicit rule. Code-bearing prompts route through your chosen model provider — OpenAI, Anthropic, Bedrock, or self-hosted. On Enterprise, the whole fleet runs in your VPC.
Always. The owlets don't sleep, don't take PTO, and don't go dark on Fridays. Nightwatch is permanent — the pane is live 24/7, and so are they.
Yes. Custom agents are first-class — not a plugin, a full role. Author a YAML skill pack (tools, prompts, guardrails), point it at any adapter (OpenAI, Anthropic, Bedrock, MiniMax, OpenClaw, OpenRouter, or your own HTTP endpoint), and Steve routes to it by capability, same as any owlet. Docs → /agents/custom.
Enterprise tier ships a Terraform module and a Helm chart. Runs on your AWS, GCP, or on-prem Kubernetes. SOC 2 Type II on the managed plane.
FLEET READY. IDLE. WAITING.

Self-host the flock.

git clone https://github.com/owlops-cloud/owlops

MIT licensed · self-host ready · BYOA/BYOK

View on GitHub →Read the docs
Tweakslive