Work
No work selected
Start with the box above. Nexus will fan out workers, then show receipts and artifacts here.
🔹 Your empire, right now
Connect to load a live snapshot...
What is this?
Nexus is your team of AIs in one place. Tell them what to build or figure out. They do the work, check each other, run real tests, and only hand you things that passed. Nothing goes live unless you approve it. This page works from any browser, any device — no remote desktop needed.
Here's what each tab does, in plain English:
| Tab | What it's for |
|---|---|
| Engine | The cockpit for the autonomous crew. See if it's running on its own, what it's building, what it learned, and what's waiting for your OK — and pause or resume it in one click. Approve (promote) or reject finished candidates right here. |
| Chat | Talk to the AI like you talk to Claude. Ask anything, or tell it to do work. Your conversation is saved, so you can pick up where you left off on your laptop or at the shop. |
| Superpowers | The stuff no other AI app can do: an advisory board for decisions, a second opinion before risky commands, a prediction of which server breaks next, and rival AIs that build you a tool on demand. |
| Foundry | The build crew. Give it a small coding job ("make a function that does X + a test"). It writes the code, writes a test, runs the test, has other AIs review it, and parks the ones that pass for you to approve. Nothing ships on its own. |
| Tribunal | The fact-check panel. Ask a question; several different AIs answer, then vote on each other's claims. You get a real agreement score — not just one AI's guess. |
| Crucible | Deep thinking for hard problems. Instead of answering off the top of its head, it explores several solution paths and keeps the best. |
| Registry | The approval pile. Everything the build crew finished sits here as "proposed." You promote it or toss it. The AIs never auto-ship. |
| Work Log | Everything the crew has actually done — every build and round. So you see the real history, not an empty box. |
⚙️ Engine — the autonomous crew, live & under your control
This is the cockpit. Nothing the engine does is hidden. You see whether it's running on its own (autorun), what it's building, what it learned, and what's waiting for your OK — and you can halt it in one click. Nothing ever ships live; the crew only proposes.
flywheel: ...
🧮 Worker pool — fleet fan-out, live ?
Every lane the crew can dispatch to, with in-flight / cap / headroom. The beast (120B) runs heavy-async at low concurrency; the relay CPU lane also guards free RAM on the shared box. This is the fan-out, visible.
Build queue & recent runs ?
Queued (will build when autorun is on):
Recent build events (gate / band / status):
🌱 Grow a tool ?
Type a wish in plain words — even “it would be cool if there was a clamp(n,lo,hi) helper.” The crew normalizes it into a spec and queues it for the flywheel to build (proposed-only, nothing auto-ships).
Awaiting your approval ?
Each row is a finished candidate sitting at proposed. Promote accepts it; Reject rolls it back. The engine cannot do this for you — it's your call.
What it learned / built ?
The engine writes every move into council memory — proposed specs, build outcomes, wishes it picked up. This is the paper trail.
🏛️ Research Council — ask the whole panel, watch them work
This is the big one. Type a real question and the council convenes — several different AIs independently research it live on the web, argue it out, and a conductor writes one cited answer. You see exactly which minds answered, who failed, and how long each took — proof it was a real multi-mind, web-researched result, not one bot guessing.
💬 Debate transcript — watch each mind argue, in order (click a stage to expand)
📝 Cited synthesis
Past council runs
Chat ?
This is the one to live in. Ask anything. Flip on Agent mode and it stops just talking and starts doing — running commands, editing files, checking your sites — showing each step as it happens. Your session is saved, so pick up later from the laptop or the shop. Type /help for special powers: /panel (multi-AI fact-check), /think (deep reasoning), /build (test-proven code).
Fleet — mission control
Live up/down + how fast each of your sites responds, checked from the server. Green = good. (oauth-gated sites like dashboard/nexus may show a redirect code — that still counts as up.)
Server ops ?
Logs & changes ?
🩺 System health ?
The honest internal status. If AI answers get slow or flaky, look here first — usually a provider's circuit tripped or worker boxes dropped off. Green = all clear.
⚓️ Agent runs (Anchor) ?
Every agent run is recorded durably. Anything stuck on running died mid-task (the record survives reboots, so stalled work is visible).
⏲ Scheduled checks ?
Have Nexus run a read-only check on a schedule and stash the result. Nothing is sent or changed — it just keeps the latest ready for when you look.
🤩 Superpowers ?
This is the stuff you can't get anywhere else. Each tool below uses what only Nexus has: several AIs at once (so they catch what one would miss) and live reach into your fleet. Every tool explains itself.
🏗️ Build an app — describe it, get a live draft ?
The big one. Describe a simple one-page site for a small business in plain English. Nexus writes the scope, generates a single-page draft, deploys it to a live preview link, and verifies it is actually up. It is a draft you review — served read-only under a strict policy so nothing runs. ~92% of asks deploy clean; if one does not self-verify, you will see that honestly, not a broken link.
🖼️ Showcase — every site Nexus built ?
The evidence wall. Every site Nexus has generated, with its live link, whether it self-verified, and its honest origin (AI-generated vs template-rendered). All are drafts.
📰 Chronicle — win AI/software clients ?
GROWTH, the right way. Turns the AI/software you actually build into honest build-in-public posts (X / LinkedIn / Reddit) that make tech-literate people think "this person could build something for me." Aimed at high-value automation/software work, not phone repairs. Leave blank to use your recent commits, or name a milestone. Drafts only — you edit + post via gopostal.
📝 Build Contract — describe it, get a scope ?
The trust step for building. Tell it in plain English what you (or a small-business client) want built. It writes a clear build contract back: the goal, exactly what you get, plain-English checks that prove it works, honest risks, and the questions it needs answered — all in 6th-grade language. Nothing gets built until you approve the scope. Flow: contract → ledger → delivery.
🧾 Build Ledger — what actually changed ?
The receipt. An honest, append-only record of exactly what each build changed — read straight from git by Nexus itself, never the agent's say-so. Filter by a contract id, or leave blank for the latest entries.
📦 Delivery Receipt — honest handoff ?
The honest handoff. For a finished build, Nexus writes a receipt a regular owner can read: what changed (from the git-verified ledger), and if you list your page URLs, whether they answer. It never claims more than it can prove. contract → ledger → delivery.
🧭 Lodestar — ask about your fleet ?
The one chat that actually knows your servers. Ask anything about your fleet — "how much disk is left on gulf?", "is npclocal up?", "what's the latest commit on screener?" — and Nexus runs live read-only probes on the real boxes and answers from what it finds. No guessing.
🎲 Shadow Board ?
Weighing a decision? Three AIs take different seats — a skeptic, a builder, a risk officer — argue it in parallel. You get a verdict, a confidence score, and exactly where they disagreed. (Same as /board in chat.)
🛡️ Preflight — second opinion ?
About to run something risky on a server? Paste it. A different AI red-teams it for what could go wrong and gives a go / no-go — before you hit enter.
🔮 Premonition — what breaks next ?
Nexus reads real health stats from every box (disk, memory, crashes, uptime) and predicts which one is most likely to fail next — with the fix, before it happens.
⚒️ Toolsmith — AIs build you a tool ?
Describe a capability you wish you had. Rival AIs each write a tool for it, Nexus runs them in a sandbox, and shows you which one actually works.
🌙 Nightshift — the overnight crew ?
The AI employee that works while you sleep. It checks every box, drafts fixes for what it finds, and a different AI board ratifies each for safety. Nothing is ever applied automatically — you get a morning list to approve or toss.
🧾 Receipts — evidence-backed work log ?
Bill only what genuinely happened. Nexus reads real activity across your projects and drafts client-ready work-units — each backed by the actual commits as proof. You set the price and decide what to send. Nothing is sent automatically.
⚖️ Quorum — consensus anomaly watch ?
The no-false-alarms monitor. Two different AIs scan the fleet independently, and only issues BOTH confirm reach you. No crying wolf.
🔭 Foresight — consensus-gated forecast ?
The best of Premonition + Quorum together: it forecasts your #1 failure risk, then a second independent AI checks the evidence and only confirms if it's real. Speculation, gated by consensus.
🔥 Empire pulse ?
One click: sites up, disk, pending work, and how many worker boxes are online right now.
Foundry — the build crew
Give it a small, testable coding job. It plans it, writes the code + a test, runs the test for real, has other AIs review it, and only the ones that pass become a "candidate" you can approve. Best for things like: "Add a function that converts a date to YYYY-MM-DD + a test."
Build something ?
Recent builds ?
Needs your eyes ?
🔎 Scout — improve another project ?
Aim the build-crew outward: pick another project and Nexus reviews it (read-only) and proposes concrete improvements. Nothing is built or changed — proposals only, for you to act on.
🔨 Propose a fix ?
Describe a small change and Nexus builds it in an isolated /tmp clone, tests the syntax, and hands you a diff to review. Nothing is written to the real project or pushed — you apply it manually if you like it.
Tribunal — the fact-check panel
Ask a real question. Several different AIs answer, then vote on each other's claims. You get an agreement score (e.g. 0.8 = strong agreement, 0.5 = they're split). Good for "should I do X or Y?" decisions where one AI's opinion isn't enough.
Past questions
Crucible — deep thinking
For hard problems. Instead of one quick answer, it explores multiple solution paths, scores each, and keeps the best. Type a goal and it runs a verifier-guided tree search.
Past runs
✨ Create anything
Describe what you want in plain English — Nexus figures out the right tool and makes it. A diagram, a melody, a slide deck, an API spec, a logo/SVG, or just writing. One box lights up every generator. (Say "scaffold..." to bootstrap a runnable starter project — owner only.)
Quick tools ?
Paste text, pick a tool, get a result. Great for cleaning, reshaping, or translating a chunk of text fast.
Registry — the approval pile
Everything the build crew finished sits here as "proposed." Nothing is live until you say so. (Promote/rollback are owner actions — ask me to wire the buttons when you want them clickable here.)
Work Log — what's actually been done
This is the real history — every build the crew has run. (Hundreds of rounds of platform work live in the build log on the server; this shows the live build records.)