Welcome to KageOps

What is KageOps?

KageOps takes an idea — a sentence, a brief, a half-formed prompt — and runs it end-to-end through a team of AI agents. You describe what you want; the agents research, design, build, test, and ship.

Three things that make it different from other AI coding tools:

• It's a team, not a single chat. Eight specialists each own a part of the lifecycle, coordinated by an orchestrator called Sensei — nine agents in all.
• It runs on your machine. Your code, your database, your project files — all local. Cloud only handles auth and billing.
• You set the budget. Every project run has a hard USD cap. The system kills the run before it overspends.

Getting started

1. Install

Download the latest build from kageops.ai/downloads for macOS, Windows, or Linux. Open the installer and follow the prompts.

2. Sign in

On first launch, KageOps opens a sign-in window. You can sign in with email + password, magic link, Google, or GitHub. Enterprise customers can use SSO (Microsoft Entra ID, Okta, Google Workspace).

3. Pick a plan

After sign-in you'll see the plan selection window. Free is enough to evaluate the product (1 active project, 20 agent runs/month). You can upgrade any time from Settings → Billing.

4. Configure an AI provider

KageOps doesn't ship with a built-in AI key — you bring your own. Open Settings → Providers and add at least one of:

• Claude CLI — uses your Claude.ai Pro/Max subscription (no per-call cost). See "Using Claude CLI" below.
• Codex CLI — uses your ChatGPT Plus/Pro subscription (no per-call cost). Same pattern as Claude CLI — see "Using Codex CLI" below.
• Claude API — premium quality, pay-per-token (set ANTHROPIC_API_KEY)
• OpenRouter — pay-as-you-go across many models (recommended for budget runs)
• Ollama — fully local, free, slower (great for dry runs)
• OpenAI, Gemini — also supported

5. Run your first project

Click + New Project in the Command Center. Give it a name, write a prompt describing what to build, set a budget cap, and pick which phases to run. Click Create & Start.

The Autonauts

KageOps coordinates Sensei (orchestrator) plus eight specialist agents — nine in total. Each owns a slice of the product lifecycle.

Project lifecycle

Every project flows through up to six phases. You pick which ones to run when you create the project — skip what you don't need.

1. Discovery — Scout researches the problem space.
2. POC — Smallest possible thing that proves the idea works.
3. Business viability — Costs, market fit, pricing assumptions.
4. Design & Planning — Pixel designs, Blueprint architects.
5. Development — Forge, Cipher, and Aegis build. Vigil tests.
6. Launch & Growth — Herald handles announcement, copy, growth.

Phase gates run between stages. Depending on your trust level, gates either pause for your approval or auto-advance:

• Low trust — approve every gate (best for first-time use)
• Medium trust — auto-advance routine gates, pause for risky ones
• High trust — autonomous after design phase

AI providers

You bring your own AI keys. Configure them in Settings → Providers.

You can mix providers per agent. For example: Claude for Forge, OpenRouter+Haiku for Scout, local Ollama for Vigil's test scaffolds. KageOps ships with several presets — pick one from the Command Center dropdown to apply a sensible mix.

Using Claude CLI

If you have a Claude.ai Pro or Max subscription, the Claude CLI provider is the most cost-effective option — KageOps spawns the official claude binary and it runs against your subscription quota with no per-call charge.

Prerequisites

• An active Claude.ai Pro or Max subscription
• Node.js 20+ already installed (you almost certainly have this)

1. Install the CLI

One npm command, globally:

npm install -g @anthropic-ai/claude-code

Reference: Claude Code quickstart · GitHub repo.

2. Authenticate

From any terminal, run:

claude

It opens a browser window, you sign in to Claude.ai once, and the session is cached in ~/.claude/. KageOps reads from this same cache — you don't need to do anything else.

3. Verify KageOps can find it

Open Settings → Providers → Claude CLI and click Test connection. If KageOps can't locate the binary on its own (rare; we check Homebrew, ~/.local/bin, npm-global, and Windows Program Files), set the KAGEOPS_CLAUDE_CLI_PATH environment variable to the absolute path.

Workspace isolation

When KageOps spawns the Claude CLI for a project, it sets the subprocess working directory to that project's workspace — not the KageOps install directory. The CLI reads the project's own CLAUDE.md (if any) and writes files into the project, never into KageOps itself.

Using Codex CLI

If you have a ChatGPT Plus or Pro subscription, the Codex CLI provider is the OpenAI equivalent of Claude CLI — KageOps spawns the official codex binary and it runs against your subscription quota with no per-call charge.

Prerequisites

• An active ChatGPT Plus or Pro subscription
• Node.js 20+ already installed

1. Install the CLI

npm install -g @openai/codex

Reference: OpenAI Codex CLI on GitHub · Codex CLI docs.

2. Authenticate

codex

It opens a browser window, you sign in to your ChatGPT account once, and the session is cached locally. KageOps reads from this same cache — no extra setup.

3. Pick the preset

In Settings → Providers → Preset, choose OpenAI Codex CLI (subscription). The preset routes every agent (Sensei, Scout, Forge, etc.) through Codex.

Tuning

Set KAGEOPS_CODEX_CLI_PATH if the binary lives somewhere unusual. Set KAGEOPS_CODEX_CLI_ARGS to pass extra Codex flags (e.g. --approval-mode auto-edit). Pin a specific model by editing the seeded preset JSON: change "model": "codex-cli" to "model": "codex-cli/gpt-5-codex".

Cost controls

The biggest fear with AI agents is "what if it spends $200 while I'm getting coffee?" KageOps stops that from happening.

• Per-run budget cap — hard USD limit. Set it on the New Project form. The run is killed the moment cumulative spend hits the cap.
• Per-task call cap — each agent task can make at most N AI calls. Default: 8.
• Per-agent token cap — each agent has a max-tokens-per-call ceiling so a single response can't run away.
• Zombie guard — if no progress is made within 60s, the run aborts.
• Live cost meter — top-bar shows current run cost in real time.
• Dry-run mode — preview the entire plan without making any AI calls.

Plans & billing

Annual plans bill once for 10 months = 2 months free. All paid plans include a 30-day free trial — full features, card required at checkout so trial-to-paid is automatic if you don't cancel. Manage your subscription from Settings → Billing — that opens the Stripe customer portal in your browser.

BYOK note: KageOps is the orchestration layer. Your AI provider bill (Anthropic, OpenAI, OpenRouter, Ollama, Gemini) is on your own account. The prices above are KageOps only — separate from whatever you spend on inference.

Privacy & data

KageOps is local-first. The desktop app is the only place your project data lives.

• Your code — never leaves your machine. Git, files, build artifacts: local only.
• Your prompts — sent to whichever AI provider you configured. KageOps doesn't proxy them.
• Your AI keys — stored in your OS keychain, never in plain-text files.
• Sign-in — handled by Clerk; only a short-lived session token is kept.
• Billing — handled by Stripe; KageOps never sees your card.
• Telemetry — opt-in only; anonymised crash reports.

Embedded Postgres data lives at ~/.kageops/pgdata on your machine. You can inspect, back up, or delete it with normal filesystem tools.

Troubleshooting

Sign-in window doesn't load

The sign-in flow opens a window that connects to api.kageops.ai. Check your firewall isn't blocking it. If you see a "Worker not deployed" error, KageOps is unreachable — try again in a few minutes or contact support.

Splash screen hangs

The splash is timed (5s by default). If it stays open longer, the embedded database may have a stale lock from a previous crash. Quit the app from the tray, then delete ~/.kageops/pgdata/postmaster.pid and relaunch.

Project run starts but no agents do anything

Most likely your AI provider key isn't valid, or the model isn't available. Open Settings → Providers, click Test connection, and check the response.

Run hits the budget cap immediately

Some models charge by token; a single Claude Opus call can cost $0.50+. Either increase the cap or switch to a cheaper preset (try openrouter_budget).

"Plan window keeps appearing on every login"

This is fixed in v0.5.0+. If you're on an older build, your plan choice didn't sync to the cloud. Update the app from /downloads.

Support

Need help?

• Email: [email protected]
• Bug reports: include the version (top-right corner of Command Center) and your platform.
• Feature requests: we read every one — be specific about the use case.
• Enterprise / custom deployment: [email protected]