VictorOS

01 Private build · version 1.5 · macOS, Apple Silicon

AI agents test your business ideas.
You hold the gate.

VictorOS is a local-first Mac app that runs a small team of agents to find, vet and test ideas. Agents do the slow work. Nothing leaves the app until you click approve.

Move through the V. Scroll, swipe, or use the arrow keys.

02 What it is

Agents do the slow work.
You open the gate.

Scouts find ideas. An evidence check opens the cited sources. A filter ranks what is left and an analyst scores the best. Then it stops, and waits for you.

  • Evidence before claims. Every number an idea states is checked against a page that was really fetched. The rest is labelled unverified.
  • A person approves. Rejecting with a reason teaches the agents your taste.
PIPELINEAgents searching
Found 0Approved 0Rejected 0Sent without a click 0

    Simulation. Placeholder ideas, not real results. Keys: A approve, R reject.

    03 What it does

    Eight things,
    none of them hidden.

      04 Engines and cost

      Choose the engine.
      Watch the tokens.

      A switch in the header picks Claude, ChatGPT, Local or Hybrid. A bar shows what each has left, before a job starts. Hybrid sends small jobs to the local model for free and big ones to whichever cloud engine has more room.

      Run an engine empty below. It is not asked again until it should be back, and the job moves on. If none has room, the job stops with a clear message instead of failing partway.

      Simulation. Bars are illustrative, not real usage. Local mode has no web search.

      Try it. It halts every animation on this page, the way the header button halts every agent.

      05 Safety

      One button stops everything.

      • Kill switch. Cancels running agents, blocks new ones, and survives a restart.
      • Approval inbox. Outreach, landing copy, customer deliverables and alerts wait for a click.
      • Spending limits. Per search, per idea, per day, and for what agents do on their own.
      • Keys stay on your Mac. Readable only by you, changeable only from that Mac, and never in backups.
      • Web content is untrusted. Agents are told never to follow instructions found in pages or replies.
      • Email is on you. The app helps with opt-outs, a daily cap and a do-not-email list. It does not make cold outreach legal or safe by itself, and this is not legal advice.

      06 Under the hood

      Plain parts.
      Stated plainly.

      $ electron, macOS arm64, unsigned build

      $ node backend, zero dependencies

      $ plain JavaScript interface, no frameworks

      $ data: local folder, plain JSON files

      $ works with: Claude (Claude Code or API), ChatGPT (Codex CLI or API), Ollama, AgentMail, Reddit, Brave Search, ntfy

      $ tests: ~240 unit, 16 browser files, CI on every push, all against mock services

      $ updates: pulls from its git repo and rebuilds on the owner's Mac. Not a public updater

      Status: private build, tested against mock services. Not yet run with real accounts at scale.

      NOT BUILT

      • No payments, hosted landing pages or automatic delivery. Taking money is manual.
      • No cloud service, accounts or team features. One person, one Mac.
      • No keyword-volume lookup.
      • No Apple on-device model support.
      • No App Store or signed distribution.

      ON THE ROADMAP

      Intentions, not promises or dates: on-device models, keyword volume, signed distribution.

      07 Private build

      Be told when it drops.

      VictorOS is not public yet. Leave your address and you will get an email when access opens.

      Your address is used only to tell you when VictorOS is available. Ask and it is removed. No tracking scripts, no cookies on this page.

      Private build v1.5 · Claude, ChatGPT, Ollama, Reddit, Brave Search, AgentMail and ntfy are named only to say what it works with; their owners do not endorse it. The business-book summaries are not the books.

      Questions

      Is it available?

      It is a private build. Leave your address to be notified when that changes.

      Does it make money for me?

      It helps you find and test ideas with evidence. It does not promise income, and it does not take payments or deliver to customers.

      Will it email people on its own?

      No. Every outreach message waits for your approval. With email connected, it can send approved drafts, with opt-out handling, a daily cap and a do-not-email list. You are responsible for compliance.

      Where does my data go?

      Data and keys stay on your Mac. Agents send prompts to whichever engine you chose (Claude, ChatGPT), so that provider sees what the agents send. Local mode sends nothing off your Mac.

      Can I run it fully offline?

      Local mode runs on your own model with no cloud AI. It cannot search the web, and the evidence check holds back more ideas.

      Which AI do I need?

      Any of: a Claude plan with Claude Code, a ChatGPT plan with Codex, API keys, or a local Ollama model. Hybrid uses whatever you connect.

      Windows or Intel Macs?

      Currently built for Apple Silicon Macs only.

      Is it signed by Apple?

      No. It is an unsigned build, so macOS will ask you to confirm opening it.

      How accurate are the AI's claims?

      It checks stated numbers against pages it really fetched and labels anything it could not check. That reduces unsupported claims. It does not make the AI infallible.