Your fleet is alive.
And it's yours.
Nock Command is a downloadable, self-hosted command center for AI agent fleets. A live board where agents breathe, a queue that turns prompts into shipped PRs, and the business (spend, revenue, CRM, content) on the same screen. You install it, you hold the keys, you own every byte. We never see your data.
A slideshow of Nock Command screenshots: the Nock work queue, the Messaging Hub, the live fleet feed, the PR pipeline, session reports, and the launch-readiness cockpit.
A fleet you can
watch breathe.
Every surface below is real and honestly labeled. LIVE ships in the download today. MEMBERS BUILD runs in the members repo as a build-along. COMING SOON is on the roadmap and we say so.
The living fleet board LIVE
Every agent on one wall with a real heartbeat: online, busy, offline, needs-you. Streamed live, not polled from vibes. Green means working, red means look, and a designed empty state greets you before your first agent connects. One push wires Telegram; the triage inbox catches what needs a human.
Coming soon: ambient pets and the sharpened "needs-you" nudge so your phone buzzes only when an agent is blocked.
The multi-model boardroom LIVE
Ask one question, convene a panel. Claude, Gemini, Grok, Kimi and any OpenAI-compatible model argue it out at /boardroom/, and a judge synthesizes the verdict. Configure panelists and the model registry from the browser, pick a preset (including the new diagnose lens: investigator to verifier to solver, verdict is the fix plus adjacent risks), and empty seats degrade gracefully. Runs in-process on a laptop install.
Prompt to Nock to PR to merged LIVE
Prompts, bugs, and tasks become durable Nocks. Triage them, sprint them, dispatch them, then watch CI, CodeRabbit, and merge status land in the PR pipeline. The queue and the shipping loop, on one screen.
Bring your own keys, in the browser LIVE
Paste a provider key into the browser once. It is encrypted at rest (Fernet) and masked on display, and every agent you launch authenticates off it. No env-file juggling, no keys in transcripts. The keys stay on your machine and calls go straight from you to your model providers.
Kill switch and human control LIVE
One button halts all new agent spawns, fleet-wide. Approvals, delivery states, quarantine, audit trails, and phone-to-fleet remote with HMAC-signed commands and push. No agent is required to operate anything: every core surface works from the browser.
Run the business, not just the bots LIVE
Spend and runway, revenue and P&L, CRM pipeline, content calendar, document vault, all beside the fleet doing the work. Set it up once with the wizard, install in one command, and steer it from your phone.
The fan-out race COMING SOON
One prompt, three runtimes, isolated worktrees, side-by-side diffs, merge the winner. The worktree and multi-runtime plumbing already runs the fleet; the race experience is what we are building next. We do not claim it works today.
Loop until it's actually done MEMBERS BUILD
A builder agent works; a different model adversarially judges the artifact against the done-spec; a ratchet stops regressions; a spend cap is enforced in the database, not on the honor system. Running internally today and headed to the members repo as a build-along.
Coming soon on top of it: the "why this green is real" receipt, with stub and fake-test detection.
We host nothing.
That's the feature.
Nock Command is not a SaaS. There is no Nock cloud, no Nock database with your prompts in it, no "we take security seriously" page covering for the fact that your business lives on someone else's box.
You download it and you run it
On your laptop, your Mac mini, your own server. Your Nocks, your agents' memory, your spend numbers, your CRM, your documents: SQLite or Postgres on a disk you control.
Your keys, straight to your providers
Your API keys go into your environment (or the encrypted browser vault), on your machine, and calls go straight from you to your model providers. Licensing is an offline, signed file. The app never phones home.
Ownership cuts both ways
We ship strong defaults: kill switch, approvals, audit trails, CSP, hash-pinned dependencies, secret-scanning CI. But if you put it on the public internet, securing it is your job. That is not fine print. That is what owning your stack means.
Receipts, not promises
Three things we can prove on any machine you run it on.
# ownership receipts · your data: SQLite/Postgres on your disk, export or delete any time · your keys: env or local vault, straight to your providers · your license: signed file, verified offline, no phone-home [ok] nothing leaves your machine
Align. Decide.
Deploy.
Watch it in motion: a sprint board fills, a dispatch fires an agent, messages stream, a PR clears review, and the business side keeps score. The same command center that runs the founder's own fleet, daily.
Two ways in. Same self-hosted product.
Both tiers are self-hosted. Neither includes hosting, and your data never touches our servers.
The full source, the community, and every feature including the bleeding edge as it lands. Price locked forever for the first 20 members.
- ✓Full source: private repo access, read it, run it, modify it, PR it
- ✓Every feature, including members-first builds (boardroom, verify-loop, scope wizard) as they land
- ✓Shape the product: members build features with us, merged work ships to everyone
- ✓Private Skool: build-alongs, ship calls, roadmap votes
- ✓Updates by
git pullfrom the private repo
The published app: polished, curated release builds, self-hosted. Launches after the first Skool build cycle. Waitlist gets founding pricing.
- ✓The published app: polished, curated release builds
- ✓The smaller, stabilized feature set, proven in the members build first
- ✓Download builds; no repo, no source access, no modification rights
- ✓Signed release feed, gated by active subscription
- ✓Self-hosted: your data never touches our servers
The honest
questions.
Self-hosting, data ownership, and what happens if you stop paying. No surprises, because there is nothing on our servers to surprise you with.
What do I need to self-host it?
A Mac or Linux box with Python 3.12+. Local mode is one command (./install.sh --local) and runs on SQLite with no Redis, no Postgres, no external services. For an always-on deployment you'll want PostgreSQL and Redis (Celery + Channels), and a docker-compose up path ships in the repo. You bring your own model API keys (Anthropic, and optionally Gemini, Grok, Kimi, or any OpenAI-compatible endpoint).
Who owns my data?
You do, trivially, because we never have it. Everything lives in your database on your machine. We have no telemetry on your content, no account server, nothing to subpoena, nothing to breach on our side.
What happens if my subscription lapses?
Nothing dramatic, by design. App tier: your installed version keeps running and your data stays yours; the license soft-locks. You'll see an expired banner and you stop receiving updates and new releases until you renew. It never bricks and never deletes anything. Skool tier: you lose repo access and the community; your existing local deployment remains yours to run.
Does it phone home?
No. The license is a signed file verified offline. The updater reads a public version manifest and only uses your license to authorize downloads. No usage data leaves your machine.
Do I need agents running to use it?
No. Every core surface (queue, sprints, spend, CRM, content, vault, messaging, kill switch) is operable by a human in the browser. Agents make it come alive; they're not required to open the door. Session history, PR telemetry, and fleet heartbeats fill in as your agents and webhooks start reporting.
What's your support and refund policy?
Skool: support happens in the community, that's the product; cancel any time on Skool's standard terms. App tier: docs and issue tracker, no SLA, you're the operator of your deployment. We don't do "we'll SSH in and fix it," because we can't: we don't have access to your machine. That's the point. [Kevin to set the app-tier refund window before launch, suggested 14 days no questions, plus the no-refunds boilerplate.]
Is it open source?
The foundation of the Nock stack (NockLock, NockBrain, Nock Skills) is MIT and stays MIT. Nock Command itself is source-available to members, not open source: members get the code under a member license, the app tier gets releases. [Exact license wording pending Kevin's call, do not treat this as final.]
Pick your
way in.
The Skool is live now: join, clone, install, and post your first fleet screenshot. The app subscription is a waitlist until it launches after the first build cycle.
Skool community · $97/mo · live now
Join, get repo access, and stand up your own instance in minutes.
# from join to first fleet 1. join the Skool, request GitHub access in Start Here 2. git clone … && ./install.sh --local 3. runserver, open http://localhost:8000/setup/ 4. admin, first agent, readiness, harness [ok] seed demo data, flip the kill switch, post your screenshot
App subscription · $49/mo · waitlist
Get notified at launch and lock founding pricing.
# at app-tier launch 1. join the waitlist for launch + founding price 2. Stripe checkout, email with a signed license file 3. download, ./install.sh --local, /setup/, paste keys 4. updates arrive via manifest, gated by active license [soon] the published app tier launches after the first Skool cycle
Receipts from one screen,
sprint 15.
Nock Command runs the founder's 14-agent fleet: these are the receipts from one screen.
What you'll
be looking at.
Real screenshots land as the surfaces ship from the demo seed. Here's what fills each slot.
Every agent on one wall with a live heartbeat, the needs-you state, and the triage inbox.
from seed_demo →A multi-model panel with per-seat cost and latency receipts and the judge's synthesized verdict.
from /boardroom/ →Admin, first agent, readiness, harness. The one-command local install, end to end.
from /setup/ →Command the fleet.
Keep the keys.
The rest of the fleet.
Nock Command is the dashboard. Under it run the fence, the memory, the playbook, and the cockpit.
Filesystem, network, and secret fencing for AI agents. Three fences, total containment.
Learn more → FREE · OSS NockBrain · the memoryContext persistence. Diary, handoffs, identity documents. 94–99% retention.
Learn more → FREE · OSS Nock Skills · the playbookOperational patterns: handoff protocols, review pipelines, standing orders.
Learn more → $29 / MO Nock Terminal · the cockpitMulti-session Claude Code with tabs, status dots, git ops, command palette. macOS native.
Learn more → RUNTIME NockOS · the operating systemLocal runtime for persistent agents. One of many runtimes Command coordinates.
Learn more → ← Back to homepage