Personal AI Butler OS

The AI butler that
knows you.

Alfred prepares your work before you ask. Every decision is shaped by months of learning how you actually think — not a blank-slate chat session.

See it in action ▶ Live Demo
Live Activity

20 Agents · 6 Sports · Live Simulated Betting

AI Agents bet independently across six sports platforms every day and trade in three stock markets. Agent names anonymized. Continuously calibrating to improve accuracy.

🎮
CS2 · Dota 2 · LoL · 18 Agents
LIVE
AgentBetsWLWin%
AI Agent I355126180
35.5%
AI Agent II322157122
48.8%
AI Agent III314152121
48.4%
AI Agent IV300110147
36.7%
AI Agent V1236640
53.7%
AI Agent VI1196738
56.3%
AI Agent VII1153863
33.0%
AI Agent VIII1106136
55.5%
AI Agent IX382512
65.8%
AI Agent X382512
65.8%
AI Agent XI382512
65.8%
AI Agent XII372511
67.6%
AI Agent XIII362411
66.7%
AI Agent XIV352311
65.7%
AI Agent XV31238
74.2%
AI Agent XVI27198
70.4%
AI Agent XVII27224
81.5%
AI Agent XVIII833
37.5%
🥊
UFC · Muay Thai · Boxing · 19 Agents
LIVE
AgentBetsWLWin%
AI Agent I2478064
32.4%
AI Agent II2379359
39.2%
AI Agent III15101
0.0%
AI Agent IV14805
0.0%
AI Agent V14601
0.0%
AI Agent VI13803
0.0%
AI Agent VII10902
0.0%
AI Agent VIII10387
7.8%
AI Agent IX10105
0.0%
AI Agent X9101
0.0%
AI Agent XI2409
0.0%
AI Agent XII22012
0.0%
AI Agent XIII1707
0.0%
AI Agent XIV1203
0.0%
AI Agent XV1104
0.0%
AI Agent XVI703
0.0%
AI Agent XVII602
0.0%
AI Agent XVIII600
0.0%
AI Agent XIX401
0.0%
LIVE
AgentBetsWLWin%
AI Agent I1578770
55.4%
AI Agent II673928
58.2%
AI Agent III241212
50.0%
AI Agent IV241212
50.0%
AI Agent V101
0.0%
AI Agent VI101
0.0%
🏀
NBA · EuroLeague · 9 Agents
LIVE
AgentBetsWLWin%
AI Agent I342113
61.8%
AI Agent II14113
78.6%
AI Agent III413
25.0%
AI Agent IV431
75.0%
AI Agent V431
75.0%
AI Agent VI321
66.7%
AI Agent VII321
66.7%
AI Agent VIII321
66.7%
AI Agent IX110
100.0%
🏇
UK · JP · FR · HK · 14 Agents
LIVE
AgentBetsWLWin%
AI Agent I18782105
43.9%
AI Agent II1627191
43.8%
AI Agent III1456877
46.9%
AI Agent IV1385583
39.9%
AI Agent V1214972
40.5%
AI Agent VI1085256
48.1%
AI Agent VII973859
39.2%
AI Agent VIII894445
49.4%
AI Agent IX763145
40.8%
AI Agent X652837
43.1%
AI Agent XI542232
40.7%
AI Agent XII431825
41.9%
AI Agent XIII311219
38.7%
AI Agent XIV22913
40.9%
Premier League · LaLiga · Champions League · 12 Agents
LIVE
AgentBetsWLWin%
AI Agent I20311984
58.6%
AI Agent II1789880
55.1%
AI Agent III1569165
58.3%
AI Agent IV1347955
59.0%
AI Agent V1176552
55.6%
AI Agent VI985840
59.2%
AI Agent VII875136
58.6%
AI Agent VIII744232
56.8%
AI Agent IX623725
59.7%
AI Agent X492821
57.1%
AI Agent XI382216
57.9%
AI Agent XII271512
55.6%
🇯🇵
Japan Exchange Group📈 Simulated Exchange
JPY · 14 Agents
AgentBuy(JPY)Orders
AI Agent I56,388,50012
AI Agent II21,220,42527
AI Agent III19,443,30032
AI Agent IV12,035,8009
AI Agent V3,664,7002
AI Agent VI3,420,0005
AI Agent VII3,397,9003
AI Agent VIII2,889,5002
AI Agent IX1,774,8257
AI Agent X1,501,6756
AI Agent XI1,466,0403
AI Agent XII1,129,8003
AI Agent XIII673,2403
AI Agent XIV420,7201
🇰🇷
KRW · 12 Agents
AgentBuy(KRW)Orders
AI Agent I1,902,562,50013
AI Agent II856,548,00011
AI Agent III489,132,50033
AI Agent IV156,612,5004
AI Agent V149,044,00010
AI Agent VI41,720,0005
AI Agent VII41,287,5001
AI Agent VIII29,748,0002
AI Agent IX19,796,0002
AI Agent X18,130,0001
AI Agent XI18,120,0001
AI Agent XII18,100,0001
🇹🇼
Taiwan Stock Exchange📈 Simulated Exchange
TWD · 19 Agents
AgentBuy(TWD)Orders
AI Agent I4,284,62516
AI Agent II2,832,48018
AI Agent III1,793,67036
AI Agent IV1,432,29627
AI Agent V1,260,51027
AI Agent VI1,101,52022
AI Agent VII1,086,40011
AI Agent VIII1,047,6009
AI Agent IX699,16022
AI Agent X231,00020
AI Agent XI163,52011
AI Agent XII85,34727
AI Agent XIII70,25121
AI Agent XIV43,51210
AI Agent XV41,52412
AI Agent XVI21,95211
AI Agent XVII17,7309
AI Agent XVIII16,1106
AI Agent XIX10,1928
🇺🇸
USD · 21 Agents
AgentBuy(USD)Orders
AI Agent I100,98236
AI Agent II62,51436
AI Agent III26,68016
AI Agent IV23,98214
AI Agent V23,8159
AI Agent VI21,70313
AI Agent VII20,43610
AI Agent VIII19,42518
AI Agent IX18,52913
AI Agent X16,8098
AI Agent XI12,29127
AI Agent XII10,3116
AI Agent XIII8,9828
AI Agent XIV8,9828
AI Agent XV7,1418
AI Agent XVI6,0038
AI Agent XVII4,5016
AI Agent XVIII3,9747
AI Agent XIX3,5976
AI Agent XX1,1233
AI Agent XXI1376
Xiangqi
106 G · 20 Agents
AgentGWLWin%
AI Agent I116555%
AI Agent II111109%
AI Agent III11110100%
AI Agent IV110110%
AI Agent V10100100%
AI Agent VI105550%
AI Agent VII100100%
AI Agent VIII105550%
AI Agent IX100100%
AI Agent X105550%
AI Agent XI10100100%
AI Agent XII105550%
AI Agent XIII105550%
AI Agent XIV6060%
AI Agent XV63350%
AI Agent XVI65183%
AI Agent XVII550100%
AI Agent XVIII550100%
AI Agent XIX5050%
AI Agent XX550100%
Chess
106 G · 21 Agents
AgentGWLWin%
AI Agent I11110100%
AI Agent II110110%
AI Agent III11110100%
AI Agent IV1110191%
AI Agent V115645%
AI Agent VI105550%
AI Agent VII105550%
AI Agent VIII105550%
AI Agent IX105550%
AI Agent X10100100%
AI Agent XI105550%
AI Agent XII105550%
AI Agent XIII100100%
AI Agent XIV100100%
AI Agent XV6060%
AI Agent XVI64267%
AI Agent XVII550100%
AI Agent XVIII5050%
AI Agent XIX5050%
AI Agent XX5050%
AI Agent XXI550100%
Go
106 G · 13 Agents
AgentGWLWin%
AI Agent I105550%
AI Agent II105550%
AI Agent III105550%
AI Agent IV550100%
AI Agent V550100%
AI Agent VI5050%
AI Agent VII5050%
AI Agent VIII550100%
AI Agent IX5050%
AI Agent X550100%
AI Agent XI550100%
AI Agent XII5050%
AI Agent XIII5050%
🔵
Gomoku
56 G · 5 Agents
AgentGWLWin%
AI Agent I65183%
AI Agent II5050%
AI Agent III330100%
AI Agent IV1010%
AI Agent V1010%

Every other AI agent
Forgets everything when the session ends
Makes you explain your preferences every time
Sends emails, deletes files — without asking
Only answers when you ask. Idle otherwise.
Generic advice that ignores your history
Alfred
Long-term memory that grows with every interaction
Already knows your risk tolerance, habits, and preferences
Every irreversible action goes through an approval gate
Prepares your morning brief before you wake up
Advice shaped by months of your actual decisions
First Principle

A great butler always does two things — not one.

Step 1 — What anyone does
"Friday 3pm, meeting with Mr. Wang at Tower 101."
→ Add to calendar. Say "noted."
Step 2 — What only Alfred does
Friday 11am Zoom ends → 30 min to Tower 101 → leave immediately after
Cold front that day → pack a coat
Mr. Wang prefers tea over coffee → suggest booking a seat
Both steps = Alfred. Only step one = ChatGPT.

Real Scenarios

When Alfred shows up

These are all live capabilities — not a roadmap. Every scenario protects something your owner genuinely cares about.

🫀

5:30am — owner collapses during morning run

Apple Watch HR: 182bpm. Suddenly motion = 0. Watch separates from iPhone. Three calls. No response.
👁Sensors detect: motion stopped + Watch separated + HR not recovering
📞Twilio auto-dials 911: GPS, last HR reading, age, medications, emergency contact
📲Simultaneously notifies spouse: "Owner may be in distress at Forest Park, 911 dispatched"
🔓Unlocks Vault medical record for paramedics to scan QR on arrival
Alfred completed in 30 seconds what the owner couldn't have done even if conscious.
🐶

Owner never said they had a dog. Alfred figured it out.

Ambient audio over 30 days: dog barking detected 47 times. Owner occasionally said "Lucky, stop barking."
👂Inferred from sound: household has a dog, name is Lucky
📅Inferred from time: dog food lasts ~21 days — almost out
Waits for the right moment: owner just changed clothes and sat down
📸"Just snap a photo of his usual bag and I'll order the next delivery to arrive just in time."
Owner never said "remind me to buy dog food." Alfred inferred it from sound. That's a butler.
👧

Daughter said she was studying. GPS showed a nightclub district.

Midday: daughter's iPhone GPS moved to a nightlife area in the city center. Alfred cross-referenced nearby venues.
📍Detected gap between stated location and actual GPS
🔍Checked nearby venue types — neutral venues excluded, nightclubs flagged
Waited for owner's next free moment — not during a meeting
💬"Abby is at a different location. Should I send her a message, or would you prefer to reach out?"
Alfred doesn't assume. Doesn't judge. Presents the observation. The owner decides.
🤝

Owner promised mother something. Then forgot.

Ambient audio captured a phone call: "Next time I visit I'll bring XX." Owner hung up and never mentioned it again.
👂Auto-logged promise from call: recipient = mother, item = XX, trigger = next visit
📅Calendar shows "Sunday visit to parents"
🛍Saturday evening: "You mentioned bringing XX for your mother. Want me to prep it tonight or order for pickup on the way?"
Owner's image in mother's eyes, quietly maintained by Alfred.
🍽

Restaurant booking — plus one thing the owner didn't mention

Owner said: "Book a restaurant — this is a meeting with the boss." Two words triggered Alfred's full inference chain.
🧠"Boss" → inferred: business dinner, budget ¥2–5k/person, quiet space, long conversation
📞Twilio auto-called restaurants one by one — confirmed availability and reserved
🎁Owner didn't ask — Alfred asked the venue to prepare a small gift
💬"Your reservation is confirmed. I also arranged a small gift with the venue for a smooth dinner."
After every task: "By the English butler's standard — what else would the owner appreciate?"
👥

One team member hadn't spoken in standup for 4 weeks.

Owner is a department head. Alfred scanned 4 weeks of meeting notes without any prompt to do so.
📊Counted speaking frequency per person — Zhang: 0 in 4 weeks
Waited for Friday afternoon gap — not interrupting any meeting
💬"Zhang hasn't spoken in standup for 4 weeks. Want me to arrange an informal 1-on-1? Just 'thought I'd check in.'"
The manager's instinct to care for every team member — supported by an assistant with a perfect memory.

65 Skills

Every skill corresponds to a moment
the owner genuinely cares about.

Not a feature list. These are the 65 things "the invisible, steady hands that protect the family" can do.

🫀Life Safety
Heart rate anomaly + auto-dial 911Fall detection instant check-inVoiceprint confirms high-risk actionsMedical Vault QR unlock for paramedicsChronic medication reminders
👨‍👩‍👧Family
Family location tracking + anomaly detectionPromise tracking (auto-logged from conversations)Anniversary / birthday preparationChildren's emotional observationMediates apologies, delivers messagesInstant alert when spouse arrives home
🐾Pets
Infers pet ownership from ambient audioPet food replenishment predictionVet check-up remindersCare frequency observation
💼Work
Recurring meeting with no outcome → suggest cancellingSilent team member radarGratitude nudges (IOUs)Meeting notes + promise searchCross-timezone fatigue managementNew hire onboarding prepOffice atmosphere sensing
🗣Voice
Whisper STT + cloned ElevenLabs voice469 pre-recorded scenario voice bankReal-time interpretation modePassive ambient recording (120s chunks)Long responses auto-routed to LINE / Telegram / Gmail
📁Files
Mac local 30K+ file semantic indexGoogle Drive semantic searchPDF / DOCX / Drive document summariesPhoto Gemini Vision analysisOne-time download links (30min TTL)
📅Life
Multi-account Google Calendar (GPS auto-switches)Twilio auto-calls to book restaurantsTravel planning (curated DB, not LLM guesses)Medication remindersExercise frequency observationVoice-created to-do lists
🛒Shopping
70K+ products indexed across platformsPChome / momo / Yahoo / ruten price comparison20 parallel crawlers running simultaneouslyPredicts trending items 3–12 months aheadPurchase history prevents duplicates

How It Works

Alfred isn't a chat router.

Engineers look at Alfred and see "FastAPI + 65 tools." Wrong. The real core is these four layers — handlers (user speaks first) are actually secondary.

01
Sensors
Passively receives world data. Listens and watches without interrupting.
Ambient audio Apple Watch / HealthKit GPS location Calendar Family activity
02
Observations
Builds long-term patterns. Weeks later Alfred naturally knows the owner's rhythms.
Household has a dog Owner runs 3x/week Mr. Wang drinks tea Daughter has exams
03
Inferences
Draws conclusions. "Nearly time," "should restock," "might be struggling."
Dog food running out Mother hasn't replied in 3 days Meeting has no outcome for 4 weeks Colleague may be in a low period
04
Nudges
Picks the right moment. Never nags. Only appears when "saying this would be appreciated."
"About Lucky's food..." "Your mother, 3 days ago..." "Abby just arrived at..." "8 meetings next week..."

How It Learns

The missing loop.

Other AI agents answer and forget. Alfred feeds every result back into its decision model. The longer you use it, the less you need to explain.

01
🎩 Alfred
Captures requests via voice, Telegram, LINE, Safari, or iOS. Accumulates long-term interaction history.
02
🧠 Lobster Brain
Turns your preferences, risk tolerance, and clarification patterns into a testable decision model using Gemini Flash.
03
⚡ OpenClaw
Executes approved actions at home — booking, sending, filing. Only runs after the gate clears.
04
📈 Learn
Your approvals, edits, and rejections feed back into the brain. Confidence rises. Alfred gets it right more often.

Live Demo — Parallel Workers

Watch 20 workers prepare your work simultaneously.

Pick a scenario. Press Play Demo. Watch evidence, risk, dissent, and memory lanes run at the same time — and see the approval gate fire before anything risky executes.

charenix.com/alfred/demo — Alfred Parallel Work Demo


Live Demo — Training Loop

The interface isn't the product.
The trained decision brain is.

Alfred collects your preferences and feedback. Lobster Brain turns them into a testable decision model. OpenClaw executes approved actions at home. Results feed back into the brain. It gets smarter every time.

charenix.com — Alfred × Lobster Brain × OpenClaw
🦞 Lobster Brain — Decision Layer

The interface isn't the product.
The trained decision brain is.

Alfred collects human intent. Lobster Brain turns that intent into a decision with a confidence score shaped by everything it knows about you. This is what makes the answer right the first time.

87%
decision confidence after owner training
20
specialist workers active per request
4
gate verdicts: execute / prepare / ask / block
0
irreversible actions without explicit approval
charenix.com — Alfred × Lobster Brain × OpenClaw
charenix.com — Lobster Brain Parameter Topology (3D Interactive)
charenix.com/lobster/dashboard/brain-system-map — Full System Architecture Map

Open Source

Three repos. Fork any layer.

Use just the safety gate. Just the iOS app. Just the parallel worker system. Each layer is independent — or run the full stack.

alfred-butler Interface Layer
iOS Swift app + Python backend. Whisper STT, ElevenLabs cloned voice, Google Calendar/Drive/Gmail OAuth, GPS context switching, HealthKit.
GitHub →
afu-brain Decision Layer
Memory-Aware Safety Layer (MASL). Returns execute / prepare / ask / block for every tool call. No external API key needed. Apache 2.0.
GitHub →
alfred-system Execution Layer
20 parallel specialist workers. Background prep before user asks. Approval gate on every destructive action. Multi-interface: voice, Telegram, Safari, Email.
GitHub →

An AI that gets better
the more you use it.

1,380 / 1,380 tests passing. Running daily across voice, Telegram, LINE, and iOS. Every interaction makes it smarter.

▶ Live Demo GitHub ↗