Market research, competitive analysis, opportunity sizing, go-to-market strategy. First agent in every pipeline.
v0.1.2 — Early Access
Orchestrated AI for Product Teams.
KageOps is the multi-agent platform that takes your idea from concept to shipped product — through six automated phases, with hard budget guardrails and a specialist agent for every domain.
8 specialist agents
·
6-phase lifecycle
·
$0.25 default cap
·
zero-Docker embed
Built on tools you already trust
Electron
PostgreSQL
Claude API
OpenAI
OpenRouter
Ollama
Eight specialists. One job each.
Each Autonaut owns a domain. Sensei dispatches. You get full coverage without a single point of failure.
System design, tech stack decisions, dependency graphs, and technical documentation. Owns the architecture phase.
UI components, design tokens, brand-consistent screens. Produces production-ready HTML/CSS artifacts.
Code generation, feature implementation, API integration, and acceptance-fix retries. Ships the working product.
Testing, code review, acceptance gate verification. Blocks any phase gate until quality criteria are met.
Infrastructure, CI/CD pipelines, deployment, and security hardening. Gets your product live.
Database schema, query optimization, data pipelines, and pgvector semantic search. Owns the data layer.
Launch copy, SEO, go-to-market strategy, channel selection. Turns your product into a story people want to share.
The Orchestrator
Sensei sits at the centre.
Eight Autonauts orbit. Sensei coordinates. Live edges show which agents are running right now. The whole pipeline is observable from one window — and Sensei never does the task work directly. Pure coordination.
- Decomposes your prompt into a typed task graph
- Routes each task to the right specialist
- Holds the budget — kill-switch polled every 3 s
- Brokers approval at every phase gate
Six phases, one continuous flow.
- 01
Discovery
Scout maps the problem space.
- 02
POC
Forge proves the core hypothesis.
- 03
Business Viability
Sensei pressure-tests model + cost.
- 04
Design & Planning
Blueprint + Pixel finalize the shape.
- 05
Development
Forge + Vigil + Aegis build and harden.
- 06
Launch & Growth
Herald ships and signal-tracks.
Watch the orchestration in real time.
Run progress
0%
Live event rail
- discoveryScout: scoping segments…
- pocForge: scaffold ready
- buildVigil: 312/312 green
Cost guardrail
Every task, every agent, every gate.
The Mission Control project board groups every task by phase. Drag, approve, retry — Sensei reroutes in real time.
01Discovery3
scout
Map plant-blog segments
3 sources
scout
Competitive scan
4 refs
sensei
Phase gate · approve
auto-pass
02POC + Viability3
blueprint
SSG architecture decision
2 deps
cipher
Content schema
4 fields
sensei
Cost projection
$0.18 est
03Design2
pixel
Tokens + type scale
tokens.css
pixel
Component sketches
6 patterns
04Development3
forge
Scaffold pages + posts
14 files
vigil
Lighthouse + a11y
98 / 100
aegis
Cloudflare deploy
queued
05Launch2
herald
Launch copy + meta
waiting
herald
Distribution checklist
waiting
Not another chat wrapper.
Phase-gated
Hard gates between Discovery, POC, Design, Build, Launch.
Cost-capped
Per-run USD ceiling. Kill switch every 3s.
Local-first
Embedded Postgres. Your code never leaves your box.
Multi-provider
Claude, OpenAI, OpenRouter, Ollama, Gemini.
Auditable
Every agent step logged. Every prompt versioned.
Open
Inspect prompts. Override agents. Fork the matrix.
A factory, not a collaborator.
Claude Code, Copilot, Cursor and Devin are excellent at what they do. KageOps does a different job. Below: where each tool wins — and where an unattended, budget-capped, multi-agent pipeline beats a chat window.
Reach for Claude Code, Copilot or Cursor when you're driving — exploring a repo, debugging, refactoring with a model in the loop. They're the best collaborators available.
Reach for Devin or Replit Agent when you want a cloud-hosted agent on a ticket and don't mind your repo leaving your machine or paying in opaque credits.
Reach for KageOps when you want eight specialists to take a spec from Discovery → Launch on your laptop, under a hard USD cap, with a kill-switch — without a human at the keyboard.
| Capability | Claude Code | Copilot / Codex | Cursor | Devin | KageOps |
|---|---|---|---|---|---|
| Primary mode | Interactive pair | Autocomplete + chat | AI-native IDE | Cloud autonomous agent | Unattended local pipeline |
| Specialist agent roster | — | — | — | single agent | ✓ 8 specialists + Sensei |
| Phase-gated lifecycle | — | — | — | checkpoints | ✓ 6 phases, hard gates |
| Hard per-run USD cap | — | — | — | credits, no ceiling | ✓ 3 s kill-switch |
| Multi-provider (BYO key) | Anthropic only | limited | ✓ | — | ✓ Claude · OpenAI · OR · Ollama · Gemini |
| Local-first (code stays put) | ✓ | ✓ | ✓ | cloud only | ✓ embedded Postgres |
| Auto build + acceptance gate | — | — | — | partial | ✓ npm build · ID acceptance · retry |
| Human approval at phase gates | manual | — | — | — | ✓ trust-level gates |
| Team chat + connectors | — | — | limited | — | ✓ Slack · Discord · Teams |
| Audit log + prompt versioning | — | — | — | partial | ✓ every prompt + cost recorded |
| Client cloud-burst billing (Azure) | — | — | — | — | ✓ agency-friendly |
| Pricing model | Anthropic API tokens | Per-seat sub | Per-seat sub | Opaque ACU credits | $0–$99/mo · BYO API keys |
Honest read: pair-programming tools beat KageOps for interactive work; KageOps beats them for unattended, budgeted, concept-to-product runs. Most teams want both.
From prompt to production artifact.
You
Sensei
Autonauts
Artifact
Hit the cap. Watch it stop.
Cap$0.50
Spent0.00 USD
Tokens0
Statusrunning
Every run carries a hard USD cap. Sensei polls cost every three seconds. Cross the line and the entire orchestration halts cleanly — no zombie agents, no surprise bills.
One conversation. Eight specialists working.
YOU
you · 14:23:08
Build me a landing page for a plant blog. Keep it under 50 cents.
S
sensei · 14:23:11
Decomposing into 6 phases. Routing to Scout for segments, Pixel for tokens, Forge to scaffold. Cap engaged at $0.50.
SC
scout · 14:23:42
Two segments: hobbyist growers + apartment beginners. Brief written. Handing to Blueprint.
BP
blueprint · 14:24:18
Static SSG. Three pages: home / posts / about. Markdown + tokens.css. Decision register updated.
PX
pixel · 14:24:55
Sage-green palette, Inter + Lora. Tokens shipped. Forge can pull anytime.
FG
forge · 14:25:31
Scaffold emitted. 14 files. Lighthouse 98/100/100/100. Opening PR.
VG
vigil · 14:25:48
Acceptance gate green. All required IDs present. Approving.
S
sensei · 14:25:51
Run complete. Spent $0.31. Under cap. Awaiting your review.
Safety isn't a feature flag.
Path-scoped I/O
Every agent file write is repo-scoped. Traversal blocked.
Parameterized SQL
No string concat. No injection surface.
Phase gates
Trust level controls when humans must approve.
Audit log
Every prompt, response, cost — recorded.
The technical fingerprint.
- Runtime
- Electron 34 · Node 20 · TS strict
- DB
- Embedded PGlite · external Postgres 16 + pgvector
- Bus
- Postgres LISTEN/NOTIFY
- Tests
- Vitest · 3712 tests · 158 files
- Providers
- Claude, OpenAI, OpenRouter, Ollama, Gemini
- Deploy
- Local-first · optional Azure burst
Pay for what you ship.
All paid plans include a 30-day free trial. BYO API keys throughout — KageOps is the orchestration layer; your AI provider bill stays on your own account.
Free
$0
USD/mo
No card required, forever
- 1 active project
- All 8 Autonauts + Sensei
- Embedded Postgres (no Docker)
- Manual phase gates
- BYO API keys
Starter
$12
USD/mo
$120/year — save 17% (2 months free)
- Unlimited active projects
- 2 seats (you + 1 collaborator)
- APO scheduler (opt-in)
- Cost intelligence dashboards
- Email support
Team
$39
USD/mo · 10 seats
$390/year — save 17%, $3.25/seat/mo annual
- Everything in Starter
- Shared Sensei chat per project
- Task comments + claims
- Slack / Discord / Teams connectors
- Activity feed + presence
Enterprise
$99
USD/mo · unlimited seats
$990/year — save 17%. Custom SSO, Azure burst, audit log retention
- Everything in Team
- Unlimited seats
- SAML / SSO (Entra ID, Okta, Google WS)
- Azure cloud burst
- Dedicated Slack channel
- Priority support + SLA
A different kind of agentic dev tool.
"Agents that can spend money should answer to a kill-switch."
"Specialists beat generalists. Eight sharp tools beat one fuzzy one."
"Local-first means your code, your keys, your machine. No exceptions."
Common questions.
How is this different from Claude Code, Copilot, or Cursor?
Those are collaborators — you stay at the keyboard and the model assists. KageOps is a factory — you brief Sensei once and eight specialists run the six-phase lifecycle unattended, on a hard USD cap, with build + acceptance gates. They're complementary tools, not competitors: Forge inside KageOps can be powered by Claude under the hood. See the comparison matrix.
How is this different from Devin or Replit Agent?
Devin and Replit Agent are cloud-hosted autonomous agents — single agent identity, opaque credit billing, your code on their machines. KageOps is local-first, multi-specialist (eight named agents, not one), bills against your own provider key, and the budget cap is a hard USD ceiling polled every 3 seconds. For agencies, KageOps can also burst to a client's own Azure tenant so they pay for compute directly.
How is this different from Claude Code, Copilot, Cursor or Devin?
Different shape of tool. Claude Code, Copilot and Cursor are collaborators — you drive, they help you write. Devin is a cloud autonomous agent on opaque credits. KageOps is an unattended factory: eight specialist agents coordinated by Sensei, taking a spec from Discovery to Launch on your laptop, under a hard USD cap, with build + acceptance gates and a 3 s kill-switch. See the full comparison matrix — most teams use one of each.
Is my code sent to the cloud?
No. KageOps runs locally. Your repo never leaves your machine unless you opt into cloud burst.
Which providers are supported?
Claude, OpenAI, OpenRouter, Ollama (local), and Gemini. Mix per agent.
How does the cost cap work?
Sensei polls aggregate spend every 3 seconds and aborts the run cleanly when the cap is reached.
Can I override agent prompts?
Yes. All prompts live in versioned files. Fork, tweak, ship.
Is there a free tier?
Yes — the Free tier is free forever with your own API keys, capped at 1 active project. Starter ($12/mo) unlocks unlimited projects and a co-founder seat; Team ($39/mo, 10 seats) adds shared Sensei chat and connector integrations; Enterprise ($99/mo, unlimited seats) adds SSO and Azure cloud burst. All paid tiers come with a 30-day free trial.
Is there a CLI or keyboard shortcuts?
Yes — two surfaces. Press Ctrl+K (Windows / Linux) or ⌘+K (macOS) inside Command Center for the command palette: fuzzy-search to create projects, toggle theme, switch active preset, jump to project boards, run the setup wizard, and more. For agent-level control, type / in the Mission Control shell to use the KageOps CLI: /help, /models (truth-source list of every agent's active model), /models <agent> <model> to override one live, /reopen-project <id> to bring a completed project back into Sensei's queue.
Ship more with the team you don't have to hire.
Install KageOps in 60 seconds. Free tier needs nothing but your own API keys.