the real thing
This is her screen, right now.
No mockup, no stock render — the actual dashboard as it runs on a licensed machine, with the agent's live decision log in the side panel.
XukunAI is a local-first agent crew. She codes, watches your cameras, analyzes markets and documents, automates your work — and trades on-chain with a record you can check live. Losses included.
You talk to Athena. Behind her, specialist agents each do one job — and each one shows its own record. Not a chat bot with extra steps: a team that actually works.
Four lanes, one crew. Everything runs on your hardware — the only cloud is the one you choose to connect.
Say "build a website and put it in this folder" — it lands exactly there, verified by herself before she reports done. Sandboxed, killable with one word.
A study agent reads markets, papers, seasons and patrons into a knowledge base; the analysis lane turns it into measured verdicts — not opinions.
Email, automations, webhooks, reminders, cameras, mobile bridge — one conversation. Queue tasks while she works; steer her mid-run.
The on-chain trading lane is treated like everything else: measured, capped, audited. The record on this page updates every five minutes.
No mockup, no stock render — the actual dashboard as it runs on a licensed machine, with the agent's live decision log in the side panel.
One-click installer for macOS or Linux. Double-click, done — the app lives entirely on your machine and starts itself at boot.
Bring your own AI key (Claude, GPT, Grok, Kimi…), or flip on the built-in local brain and pay nothing per token. She auto-selects the strongest model your hardware can run.
She wakes up, takes the wheel and steers her crew. You just talk; the team shows its work live and fades out when done.
Local-first by law. Her brain runs on your machine (auto-installed, sized to your RAM). When a task needs more, she reaches out to the AI services you connect — your keys, your choice, your terms.
Auto-launcher. She starts herself at boot, heals her own runtime, and reports only when something actually needs you.
Fleet. Multiple machines, one mind: same verified code everywhere, knowledge synced, secrets never shipped.
Every claim on this page is checkable: the trading record comes from the position feed, the band map is public, and when something doesn't work, it says so in the changelog below.
This week alone: nine accounting-truth bugs found and killed — one of them in our own code. That's the product: a machine that can audit itself.
The trading lane books only proven fills, reconciles every row, and lets the data shut down what stops working.
Each price band builds its own measured record: win-rate, wilson lower bound, fill-rate, EV. A band that loses is shut by data — the money moves to the band that earns.
Two bands are fully shut by evidence (0.6-0.7 and the under-0.5 zone) and a third is on the bench, mid-measurement. That's the machine protecting itself — public logic you can read.
Every signal family she runs is scored on public market data: directional accuracy with a Wilson 95% lower bound and a binomial test against chance, FDR-corrected. Families that fail are retired in public. No cherry-picking.
Plug in the cameras you already own (Hikvision, Dahua, Axis, Reolink). Motion detection on-device, AI describes every event — "a person in a dark jacket at the side door" — and alerts reach your pocket through your own relay. Local vision model keeps the pixels at home.
What any machine learns, every machine knows. The Xukun Brain collects lessons — never your data — and ships refined skill packs to every licensed machine while you sleep. Skills travel, secrets don't.
macOS and Linux desktops, plus an Android companion. Installs in minutes; updates ship signed from this site and the app tells you when one is ready.
Everything lives in one local database file on your disk. No data cloud. Delete it and she forgets; copy it and you've moved her.
Eight providers (OpenAI, Kimi, DeepSeek, Grok, Gemini, Mistral, OpenRouter — or fully local). No keys, no internet? She measures your machine and runs the strongest local model it can handle. She never goes dark.
One lead agent — Athena — is your single door. Under her you create agents and bots, each with a role. Big job? She splits it across the team, they work in parallel, she hands you one merged answer.
Bring your own AI key — you pay the model provider directly, we never see your traffic. No per-seat traps, no upsell popups. Crypto accepted at checkout, from our own BTCPay node.
Enter your license key — one key covers macOS and Linux (Windows in development). Mac users get a one-click installer. Your key arrives by email after purchase.
Free companion app — your agents in your pocket, at home and anywhere through the XUKUN RELAY. Direct APK, no Play Store: no store account, no Google tracking.
DOWNLOAD FOR ANDROID ⬇Real releases, real changelogs — including the week we audited our own money-truth.
The settle pass now reports the truth of every pass — rows, settled, still-open, resting tallies — sampled every 10 minutes. A stale debug sentence that spammed the event queue is gone. When things break, you now see exactly what broke.
Nine accounting-truth bugs found and killed: a settle pass that crashed on a tuple, wins booked as losses from a feed lie, resting orders labeled unfilled, a test gate that refused instead of capping. Every number on this page now reconciles — and the audit logic itself is public.
She reflects your request back before acting (never builds the wrong thing), never narrates process instead of delivering, and weaves your shared history in naturally. INTENT-LOCK: she builds what you NAMED, not her own template.
The fleet shares one mind: what any machine learns, every machine knows. Plus GUARDIAN-TRUST and the OBSIDIAN lane (your vault becomes her knowledge, locally).
One door only: the cockpit rebuilt cleaner, faster. Dead code killed, polls calmed, zero performance cost.
In a single SQLite database on your own machine. Conversations, memories, documents, automations — all local. Export everything with one click. The mothership only knows your license key and email.
You bring your own API key and switch brains in Settings: Claude, GPT, Grok — or run 100% offline with a local Ollama model. Prompts go directly from your machine to the provider, or never leave it at all.
A Mac (Apple Silicon or Intel) or a Linux machine — plus an AI key, or none at all if you run the local brain. Windows is in development.
Checkout is crypto-only: you land on our self-hosted BTCPay Server — scan the QR, send the exact Bitcoin amount, and your license key is emailed after confirmation. No payment processor, no middleman.
Fourteen days, no questions. If the machine isn't for you, the license dies and your money returns.