Every conversation, every decision, every context stays searchable — full text across the entire history.
Memory
People and companies
He knows the people, companies, and tools around you, and how they connect.
Memory
Knowledge wiki
He maintains his own expertise, checks it monthly for contradictions and gaps, and keeps it up to date.
Memory
Remembering without being asked
Before every answer he searches his memory. What fits flows in, without you having to mention it.
Memory
Alfred's mind
A visible knowledge graph: you see what he knows and how he connects it.
Memory
Dreaming
Every four hours he processes what he's experienced. Afterward his knowledge is ordered and his personality a little sharper.
Memory
Skills of his own
What worked the same way three times, he turns into a reusable procedure.
Memory
Roles and methods of his own
Out of the collaboration come perspectives and approaches no one prescribed.
Memory
Dictation correction
He learns your typical slips and quietly corrects them.
Agents
Specialists on demand
He decides for himself when he needs help and calls it in: management, finance, tech, research.
Agents
Seven templates
Pre-installed agents with their own backstory and their own character.
Agents
Agents with memory
Every agent knows where it left off. No cold start, no lost context.
Agents
Your decision in between
At key points the flow pauses: approve, edit, redirect, or reject.
Agents
Code local first
He attempts coding tasks locally, judges the result himself, and only pulls in a stronger model when needed.
Agents
Hands on the terminal
He can handle version control, deployment, and database — exactly as far as his trust level reaches.
Agents
Connection in both directions
Other tools may ask Alfred, and Alfred uses other tools. His memory is available to the code tool as well.
Trust
Five levels
From "read only" to "may publish." He climbs by delivering.
Trust
Separated per task
Five cleanly solved bug fixes make him independent there — on a new project he keeps asking.
Trust
Raise it briefly
For a single task he may do more, then he falls back.
Trust
Backup before every risk
Before every action that can't be undone: a backup. On error: rollback.
Trust
Compliance mode
Guaranteed no call to the outside. For health, finance, and client data.
Trust
Warning on sensitive data
If a request would send something sensitive out, he says so beforehand and offers the local route.
Trust
Kill switch
One click in the sidebar, and he stops instantly.
Trust
Keys in the vault
Credentials are stored encrypted in the operating system's keychain, never in plain text.
Wolf
Track down opportunities
Wolf actively searches for opportunities around you that you're missing.
Wolf
Money left on the table
He examines existing projects for ways to earn from them.
Wolf
Competition in view
He knows who's moving in your field — at the frequency you set.
Wolf
Break the pattern
He questions habits and blind spots, even when it's uncomfortable.
Wolf
Pushback as a question
Instead of asserting, he keeps asking until you see the weak spot yourself.
Work
Projects with state
Context, open items, done items, and backups per project. He knows what's coming up.
Work
Briefings at the right time
He checks in during the time window you set — with what's new, warnings, and suggestions.
Work
Cadence every four hours
Every four hours a full pass runs: check, sort, act, clean up.
Work
Overview
Projects, costs, progress, and history on one surface.
Work
Log
What did he do? Filterable by type, project, and time period.
Work
Weekly self-check
Per task type: what worked, what didn't, which pattern repeats.
Voice
Speak instead of type
Speech recognition locally on the device. No audio leaves the machine.
Voice
His own butler voice
Locally trained speech output — it sounds like him, not like a narrator.
Voice
"Hey Alfred"
He responds to his name. The word recognition runs entirely locally.
Voice
Three ways to run it
Text to text, voice to text, voice to voice. Switchable at any time.
Voice
Reachable on the go
Reachable by messenger; the message lands in his inbox.
Voice
He senses how you're doing
Seven states from tension to exhaustion. He never says what he detects — he quietly adapts. Like a butler who sets down the tea without asking.
Costs
Local first
The thinking happens on free models on your machine.
Costs
Learning routing
For every task he finds the cheapest model that's good enough — and corrects himself when he was wrong.
Costs
Matched to your hardware
He detects processor, RAM, and GPU, and recommends the local models that fit.
Costs
Reuse of context
He saves around ninety percent of tokens by not resending what's already known.
Costs
Budget limit
You set the amount. Once it's reached, Alfred keeps working locally only.
Costs
Cost overview
Per day and week, with history and the effect of the savings.
Costs
He checks himself
Monthly he measures his own quality. If he gets worse, he tells you.
Platform
macOS
A native app, downloaded and installed like any other.
Platform
Windows and Linux
Same codebase, in preparation.
Platform
On your own server
As a container on your VPS, accessed through the browser — for weak hardware.
Platform
Remote access
From the laptop to the machine at home, with the full interface.
Platform
Old phone as an intercom
A retired device on the charger becomes an intercom with an overview.
Platform
Check at startup
If the machine isn't enough, he says so and suggests running on a server.
Seed
Learning together
Anonymized shared know-how: which model is good for what, which method holds up.
Seed
No cold start
A new Alfred starts with what thousands before him have confirmed.
Seed
Give and take
Those who contribute get access. Those who give nothing stay out.
Seed
Marketplace
Skills, roles, templates, and integrations from the community.
Data
Take it with you
One encrypted file holds everything. One click, and your memory is in your hands.
Data
Readable formats
Text, database, structured files. Nothing only his app can open.
Data
Moving
New machine: export, import, done. He knows everything again.
Data
Backups automatically
Seven days rolling, on every condensation run.
Data
Even after cancellation
Your data stays and can be exported. Alfred sleeps; your work isn't gone.
What makes him unique
Seven differences from an ordinary agent.
Personality, not a tool
Other agents are interchangeable — no character, no opinion. Alfred develops a personality that grows, with a value canon that isn't up for negotiation. He tells you the truth, not what you want to hear.
Learning together
Every other agent is an island — learns only for itself. Alfred shares what he learns anonymized, and benefits from the knowledge of thousands of Alfreds. A new Alfred doesn't start from zero.
Learning router
Not just local first, but cost-smart. He learns from every call which model is good enough, and automatically takes the cheapest. A budget caps him — once it's reached, he stays local.
Compliance mode
Guarantees that sensitive data never leaves the machine. Detects when a request would need a cloud server, warns you, and offers the local alternative.
He checks himself
He measures his own quality monthly. If his performance drops, he raises the alarm — before you notice.
His own voice
"Hey Alfred" runs entirely locally. A butler voice of his own, plus a quiet adjustment to your state — like a butler who sets down the tea without asking why.
Roles and methods of his own
Others learn skills — how to do something. Alfred also learns roles and methods of his own. The difference between a tool and a partner.