private beta · 2026

Your idea. Our factory.
Discovery to production. One walk.
Software, hardware, or both.

Dark Product Factories is a guided, fail-closed walk from a customer problem to a product in production. Every stage draws on a library of helpers we research daily, and every ticket names the skills it turns on and the proof it must produce. Your own AI tools do the work, on your machine. The cloud is the control plane; we never clone your tree.

Request private beta access
cli on your machinewe never clone your treesoftware, hardware, or both
01 · the problem

Fast code is not a product.

Claude Code, Codex, Cursor and Grok Build will write you a function before you finish the sentence. What they will not do is ask whether anyone needs it, hold you to a requirement, insist on proof, or remember why a decision was made. Products stall in that gap, between I have an idea and the factory is walking it. dpf is that layer: from the first conversation about the customer to the day the product is in production, for software, hardware, or both.

02 · how it works

One walk. Seven stops.

Discovery to production, stage by stage. You stay in your own AI tool; Atlas shows which stage is open, what is blocked, and what needs a decision from you.

  1. ┌────────────────────┐
│  setup             │
│  tools · tracker   │
└────────────────────┘
    01setup

    You bring your tools.

    Pick the AI tool you already use, the models you can reach, your tracker, your channels, and how much you want to review. dpf only proposes helpers you can actually run, whether that is one provider, a locked-down cloud, or local models.

  2. ┌────────────────────┐
│  discover          │
│  customer · problem│
└────────────────────┘
    02discover

    We start with the customer, not the code.

    An adaptive conversation, twenty questions or fifty, until the customer, the problem, the outcome and the evidence are clear. It ends in a brief you accept or send back for more digging. Nothing presumes software.

  3. ┌────────────────────┐
│  canvas            │
│  the business case │
└────────────────────┘
    03canvas

    Make the case.

    A business model canvas filled in as you talk, with bounded background research behind each claim. Customer materials to test the idea. Investor materials and a data room only if you ask for them.

  4. ┌────────────────────┐
│  prd               │
│  spec · diagrams   │
└────────────────────┘
    04prd

    Specify the whole product.

    Requirements and architecture for the whole thing, with a master PRD and component PRDs when the product spans several repositories. Diagrams and traced requirements. The audit fails closed: a gap is named, never skipped in silence.

  5. ┌────────────────────┐
│  tickets           │
│  work · proof      │
└────────────────────┘
    05tickets

    Every ticket prescribes its own proof.

    Each ticket says what to build, which skills and audits switch on, what context an agent gets, and what evidence counts as done. You can hand-carve any of it before a single agent starts.

  6. ┌────────────────────┐
│  build             │
│  on your machine   │
└────────────────────┘
    06build

    Your agents build. Evidence graduates.

    Work runs on the machine that owns your repos, under the autonomy you chose: review every ticket, a few, or an overnight run. Tests, mutation testing, coverage and an independent clean-code review decide when a ticket is done, not a confident summary.

  7. ┌────────────────────┐
│  release           │
│  sign-off · operate│
└────────────────────┘
    07release

    Ship it, run it, keep improving it.

    Milestone validation, then a human sign-off before production. A rollback policy you set. Incidents open tickets. And when the library learns a better way, dpf shows what it would change in your repos and waits for your yes.

03 · what you get

A harness, not a wrapper.

Everything below is composed into the harness dpf writes on your machine, from the library, for your stack and your tools. Nothing is a hosted bot working on a copy of your code. The harness is yours.

[01]

A library, researched daily.

Skills, agents, audits, tools, models and connectors, each with its own page: what it works with, what it does not, what it is optimized for, and how far it has earned trust. Crawlers refresh it every day. A human merges.

[02]

Helpers written into your repos.

Hooks, sub-agents, commands, settings and CI gates composed for your stack, Rust or TypeScript or Python, and for the AI tool you use. Yours to keep, subscription or not.

[03]

Disciplines wired in.

Test-first work, mutation testing, coverage gates, dependency rules, independent clean-code review. Prescribed per ticket, enforced in CI, drawn from the manifesto.

[04]

Atlas.

Your whole product in one place, on any device: stages, milestones, repositories, requirements, tickets, blockers, decisions, diagrams, evidence and token spend. A control surface, not a status page.

[05]

Your tools, your tracker, your channels.

Claude Code, Codex, Cursor or Grok Build. Linear, Jira or your own tracker. Telegram, Slack or iMessage. One accountable person by default; add roles only when you need them.

[06]

Add your own bag of tricks.

Contribute your own skills, specialist agents and reference material to the library, where every customer benefits and the source is always credited. Material you would rather keep private stays on your machine, and you hand it to your AI tool yourself.

04 · pricing

One price. Per person. Every project.

A flat monthly subscription with no per-project fee. The harness it produces lives in your repos and stays yours.

dpf
$35/ person / month
private beta
  • →The guided walk, discovery through production
  • →The library, researched daily, delivered to your machine
  • →Helpers written into your repos, yours to keep
  • →Bring your own AI tool and keys: Claude Code, Codex, Cursor, Grok Build
  • →Unlimited projects and repositories
  • →Atlas: your product, tickets, evidence and decisions in one place
Request private beta access

Your AI tool subscription and model keys are yours; model charges stay with your provider.

Cancel any time. Work already delivered into your repos stays usable.

Priced per person. People who only watch progress in Atlas are free.

05 · faq

Questions worth answering.

What am I paying for?
The research, kept current. Dark Product Factories is the most thoroughly researched turnkey harness companion there is. You start with an idea and finish with a product, and at every stop in between your AI tools are handed the best-known way to do that piece of work, along with the proof it must produce. The skills, agents, audits and tools in the library behind it are refreshed and curated every single day, including adaptation to the newest frontier models as they ship. We never stop making it better, so every step toward your product gets sharper the longer you are with us.
What does it cost?
$35 per person per month. Unlimited projects and repositories. Your AI tool and model keys are your own, so model charges stay with your provider. Teams pay for each person who runs the walk; people who only watch progress in Atlas are free.
Which AI tools does it work with?
The one you already use. Claude Code, Codex, Cursor and Grok Build are the tools we write helpers for today. During setup dpf looks at what you can actually run, one provider, a locked-down cloud, or local models, and only proposes helpers that fit. Where your language or tools are thin in the library, a background research job goes looking, at no charge to you.
Can I bring my own skills and standards?
Yes, two ways. Contribute them to the library and dpf treats them like any other helper: they show up in ticket prescriptions, audits and Atlas, for you and for other customers, delivered inside the service and never openly published. Or keep them private on your machine and hand them to your AI tool yourself, telling it where to use them; dpf will not do that part automatically, because it prescribes from the library only.
Do I have to run the whole line?
No. Setup is a conversation. You and dpf decide which stages run, what is mandatory for a brand-new product versus an existing system, and what to skip. A skip is recorded with its reason. Nothing is skipped in silence. You can also pin a tool, a model or a skill to any stage or ticket, and dpf shows you if the pin fights a rule.
Where does my code live?
On the machine that owns your repos. dpf writes the harness there and your agents work there. We never clone your tree, and no bot in our cloud opens pull requests on your behalf.
What runs in the cloud?
Your account and subscription, the online library, and the research that keeps it current. The factory itself runs on your side: the command line tool, your AI tools, and the shared project state your Atlas reads.
What about my data?
Your product record, meaning PRDs, tickets, evidence, decisions and Atlas state, stays in your custody on your machines. We hold your account and billing details. If you leave sharing on, we also receive usage signals about which helpers worked, without your content. The library holds only what customers choose to contribute to it, and nothing private: anything you want to keep to yourself stays on your machine and never reaches us. We never train models on your material and never hand it to anyone who would. The privacy policy has the details.
Can my team use it?
Yes. One shared project state for the team. It starts with one accountable person, and you can add a lead per repository and other roles when the work needs them; we will keep nudging you to stay minimal. Everyone works from their own AI tool and Atlas shows the same truth to all of them. You choose how much runs unattended. Each person who runs the walk needs a seat; read-only viewers do not.
What happens if I cancel?
Your codebase, the harness, delivered tickets and instructions, and the local tools stay usable. Online library access ends, including updates. Anything you chose to share with the library stays shared; ask and we handle removal case by case.
Can I get my tickets and projects out?
They were never locked in. Tickets live in your tracker, Linear, Jira or your own, and everything else lives in your repositories and your Atlas state.
When is general availability?
After the private beta. Join the waitlist; first in.
06 · beta access

Request
private beta
access.

We're taking names for the private beta: solo founders and teams, software and hardware. Tell us about your product and we'll be in touch when the first cohort opens.

@yourname
where are you in the build? // required
which AI tools do you use today? // required · multi-select

We'll never spam you. Replies go to a real human.