v0.5.0 · AGPL-3.0 · local-first

Sensei plans.
Autonauts ship.

KageOps is an open-core AI engineering team you run yourself. One orchestrator, eight specialised agents, six product phases — from a one-line brief to a built, tested app, with a hard budget cap you set.

Run the quickstart View source
  • LicenseAGPL-3.0
  • RuntimeLocal + BYO keys
  • SurfaceCLI · Desktop
  • ModelsClaude · OpenRouter · Ollama

01 · Quickstart

Three commands. No account.

Clone the repo, install, and preview your first build. The dry-run never makes an AI call — it shows the plan and a cost estimate, free. No signup, no telemetry, no lock-in.

~/code — bash 
$ git clone https://github.com/hmanoor/kageops-core
$ cd kageops-core && npm install
$ npx tsx src/cli/headless-runner.ts --dry-run --name "MyApp" --description "..."
Three shell commands: clone the repo, install dependencies, and dry-run your first build.

Node 20+ only. Workspaces live in ~/.kageops. Bring your own keys ($ANTHROPIC_API_KEY, $OPENROUTER_API_KEY, or local Ollama). A real run needs an explicit KAGEOPS_MAX_RUN_USD budget cap. Never phones home.

02 · The engine

A single orchestrator. Eight specialists. Six phases.

KageOps is not a chat wrapper. It is a deterministic execution graph that plans, builds, reviews, and ships — with hard budgets at every hop.

01

Sensei — the orchestrator

One planner that reads intent, decomposes work into a typed task graph, assigns Autonauts, and arbitrates conflicts. No nested-agent recursion, no token sprawl.

Plans
typed DAGs
Memory
per-project, local
Failure mode
explicit replans
02

Eight Autonauts

Named specialists with fixed contracts — each owns one slice of the lifecycle and one toolbelt. Composable, swappable, individually evaluable.

  • Scout — strategy
  • Blueprint — architecture
  • Pixel — design
  • Forge — engineering
  • Cipher — data
  • Aegis — platform
  • Vigil — quality
  • Herald — marketing
03

Six-phase lifecycle

Every task moves through the same explicit phases. No hidden state, no improvised tool calls — the trace is the spec.

  1. 1 Discovery
  2. 2 POC
  3. 3 Viability
  4. 4 Design
  5. 5 Development
  6. 6 Launch
04

Cost guardrails

Per-task, per-phase, and per-run budgets in tokens and dollars. Sensei prefers the cheapest model that meets the contract; over-budget tasks halt loudly.

03 · Built with KageOps

Things the engine has shipped.

Real projects, fully traced. Click through to read the run log.

Featured · live

Prism

An interactive CSS design toolkit — gradients, shadows, glassmorphism, palettes, and easing curves, all client-side. Designed and built end-to-end by Sensei and the Autonauts.

  • 5studios
  • $4.17total spend
  • 0human edits
Open prism.kageops.dev
Full-stack SaaS

DevPulse

An agent-built SaaS proof — Clerk sign-in, test-mode Stripe subscriptions, and a Neon Postgres database. Built end-to-end to show the engine ships real backends, not just static pages.

Stack
Next.js · Clerk · Neon
Billing
Stripe (test mode)
Shape
auth + billing + DB
Open DevPulse

04 · Open-core

Free at the core. Honest at the edge.

The engine is AGPL-3.0 — clone it, run it, ship with it. We sell the things teams genuinely cannot self-host.

AGPL-3.0 · Open

KageOps Core

  • Sensei orchestrator and all eight Autonauts
  • Full six-phase lifecycle engine
  • Build & acceptance gates + cost guardrails
  • Embedded local database — no Docker to start
  • BYO API keys — Claude, OpenRouter, Ollama, OpenAI, Gemini
  • Headless CLI runner. No telemetry. No account.
Clone the repo
Commercial

KageOps Hosted

  • Multi-tenant trace cloud with SSO
  • Shared model budgets across teams
  • Audit-grade run signing and retention
  • Managed Autonaut fleets with autoscaling
  • Priority response on regressions
  • License clarity for non-AGPL distribution
Talk on GitHub