I wanted an AI team
I could actually run.

So I built Veyandor. You hand it a goal, an Overseer breaks it down, and a crew of agents does the real work in parallel — writing code, digging through research, drafting, building — while you stay in the director's chair. It runs on your own machine, on your own Claude. Nothing leaves your disk, and nothing ships until you say so.

Works on Windows today. macOS and Linux are close behind. Bring your own Claude — you pay Anthropic directly.

You don't start from a blank screen.

Twelve agents are there the moment you open the app — an Overseer to run the show, plus specialists for research, design, docs, automation, and the rest. Tell the Overseer what you're building and it shapes more around your goal. But you're never staring at nothing.

Overseer
Managing Partner
Managing Partner
Guide
Onboarding Agent
Onboarding Agent
Scout
Research Agent
Research Agent
Muse
Design & Mockups
Design & Mockups
Flow
Workflows & Automation
Workflows & Automation
Forge
Agent Creator
Agent Creator
Tinker
Skills · Plugins · MCP
Skills · Plugins · MCP
Archivist
Memory Keeper
Memory Keeper
Warden
Budgets & Guardrails
Budgets & Guardrails
Critic
Quality Review
Quality Review
Scribe
Docs & Summaries
Docs & Summaries
Janitor
Maintenance & Cleanup
Maintenance & Cleanup

Hand off a goal. Here's what actually happens.

1

Brief the Overseer

Tell it what you want in plain words — a feature, a bit of research, a whole project. The Overseer reads it, splits the goal into pieces, and hands each piece to the agent that fits.

2

The agents go to work — at the same time

Each one runs in its own isolated git worktree, so they don't step on each other. They write code, dig through sources, draft, and build — all on your machine, on your Claude. None of it touches a server I run.

3

It comes back for your sign-off

Anything that matters lands in the Approval Inbox first — the finished work, plus any change to a guardrail. You read it, then approve, send it back, or point it somewhere new. Nothing ships behind your back.

It all lives in one window.

The Overseer

Your right hand. It takes a goal, splits it into the right pieces, hands them off, keeps track of who's doing what, and tells you when something's stuck.

Real terminals

Every agent gets its own shell, and you can watch it work — code running, tests passing, builds finishing — right there in the window.

An editor that's already open

Read and change files without leaving the app. It's Monaco — the same editor as VS Code — so syntax highlighting and all of that just works.

A browser built in

Agents can open a real page, click around, and show you the result. Handy when they're building a UI and you want to see it, not just read about it.

A memory that sticks

Agents share one growing knowledge base, so what one figures out, the next one already knows. It tidies itself up over time instead of piling up junk.

Skills and plugins

Teach any agent something new with a skill — a reusable set of instructions — or wire in an MCP plugin. There's a marketplace; installing one takes a click.

A dial for how much leash

Four settings: Plan, Supervised, Auto, Unleashed. Supervised shows you every move before it happens. Unleashed lets the team run and report back. You decide.

Connect your tools

GitHub, Notion, Linear, Stripe, Figma, and 50-plus more, connected through OAuth — so there's no API key to dig up. Agents reach for them when a task needs them.

It runs on your machine, and it stays there.

  • It uses your own Claude. Veyandor drives the Claude subscription you already pay for. There's no markup and no middleman — every token is billed straight to Anthropic at your rate, not mine.
  • Your code never leaves this machine. I don't even have servers to send it to. The only things that ever go out are a check for updates and — once you put in a license — a tiny ping with just your license id. No telemetry, not now, not later.
  • Nothing gets uploaded. There's no cloud piece quietly watching your work. The whole thing lives on your disk, full stop.
  • You hold the guardrails. Set a budget per project. A permission broker asks first before anything touches a sensitive tool. And that four-way dial — Plan, Supervised, Auto, Unleashed — runs from "show me everything" to "just go."

Your work is safe, even when an update isn't.

  • A backup before every update. Before a new version touches anything, Veyandor quietly saves a complete, integrity-checked copy of your data — and keeps the last few. If an update ever misbehaves, the version from right before it is sitting there waiting.
  • One click back to your data. If something feels off after an update, restore the pre-update backup from Settings and the app restarts on your data exactly as it was. Nothing gets deleted to get there.
  • Updates come straight from me, signed. New versions arrive over a secure channel, only for people with a license — no third party in the middle — and the app checks the signature before it installs a thing.
  • Roll the whole version back, not just the data. If a release really doesn't sit right, drop back to the previous version entirely, with its matching backup — no work lost in between.

Made by one person

Veyandor isn't a company — it's me. I built it because directing a team of agents should feel like running a small shop, not juggling a dozen chat tabs. It's local-first because your code and your keys are yours, full stop. It's honest about what it does and what it doesn't. It's early, and I'm still shaping it — if something feels off, that's on me, and I'd genuinely like to hear about it.