Cipher is a sovereign AI workspace for Apple Silicon. It routes models locally, keeps your secrets in an encrypted vault, remembers your work across projects, reads your state, thinks in the background, and takes real initiative under your control — all on-device, nothing handed to the cloud.
Cipher isn’t a prompt box. It runs a brain, a layered memory, a read of your state, and background loops that think and learn while you work — all on-device.
Every person, project, decision and date from any studio flows into one typed knowledge graph that Cipher queries before it answers — a working model of your world, held on your Mac.
Semantic, episodic and hash-chained memory, each entry carrying a confidence and a decay curve. Cipher recalls across sessions and lets the trivial fade — remembering like a mind, not a log.
Cipher reads the emotional register of every message — calm, curious, focused, stressed, frustrated — and adjusts its tone and pace to match. On-device, instant, no tokens spent.
Even when you’re not chatting, Cipher reviews recent work on a timer — proposing goals and synthesising the patterns it notices — and surfaces them only when they’re ready for you.
A subconscious loop reads what every autonomy layer has gathered and writes back a reflection — recurring themes, blind spots, what to try next — a journal that compounds, plus on-device fine-tuning.
Cipher’s architecture grows like an organism. A new sense or studio is a single declarative file dropped into its spine; the registry discovers and wires it automatically — capabilities accrete over time without the system being rebuilt.
Cipher watches its own health — process liveness, configuration drift, foundation checks — and when something slips it proposes a remediation and applies it under your approval. A watchdog reconciles stuck work against what’s actually running, so the system rights itself.
Cipher is anchored to authoritative facts about its own origin — who built it, its mission, the systems it belongs to — so it never drifts into generic-AI confusion about what it is. And it’s aware of its own body: the machine it runs on, its battery, its resources and whether its model engine is live — it can report its own condition.
Built end-to-end — the thinking, the memory and the making all run on your own hardware.
App and website builders turn intent into real, runnable projects — this site’s sibling, a live personal portfolio, was generated by Cipher. Plus studios for legal, marketing, R&D and writing.
A multi-model council that debates and votes, ReAct and goal-oriented planners, goals and a self-improvement loop — all behind an operator on/off switch and approval gates.
On-device vision, video generation and image tools, plus fully offline text-to-speech and voice — no cloud media APIs, no audio leaving the machine.
A code brain, git, a persistent terminal, build pipelines and on-device model fine-tuning — Cipher works on your machine, as your machine.
An AES-256, machine-locked vault keeps your data private. Cipher also defends itself: a memory firewall stops its recall being poisoned, builds and tools run sandboxed, and grounding, hallucination and identity guards keep it from being talked out of what’s true or who it is — its own security skills, turned inward first.
Automatic local-and-hybrid routing picks a fast on-device model for most work and escalates only when a task genuinely needs it — sovereign by default, hybrid only when you choose.
One assistant across your Mac, iPhone, iPad, HomePod, Apple TV and Watch. The Mac is the brain; Apple’s on-device model is used purely as a multilingual translator in front of it.
When the Mac’s memory fills, Cipher ferries heavy persona work to sibling machines on your private tailnet — and gives you a live view of the k3s + Tailscale fleet it runs on, building and managing nodes from inside the workspace.
Cipher doesn’t answer as one generalist model. For each task it steps into the right skill — a curated expert briefing, many of them grounded in their own knowledge canon — so it becomes exactly the specialist the moment needs, and writes new ones as the work reveals what’s missing.
They span software engineering and testing, agent-building, deep research and analysis, writing and content, on-device media and design, security, business and strategy, and regulated work in law and healthcare — over a meta layer whose only job is to author and refine new skills from how you work. The library is a floor, not a ceiling.
Purpose-built studios turn a request into finished work, each in its own project space.