Voice

Always listening, never intrusive

Fae is an always-present companion, not a summoned assistant. She listens continuously and makes intelligent decisions about when to speak — responding naturally to direct conversation, politely interjecting when she overhears something useful, and staying quiet when the TV is on or conversations don't involve her.

The voice pipeline runs entirely on-device: echo cancellation separates your speech from Fae's own voice and ambient noise, voice activity detection filters silence, and speech recognition converts your words to text with no cloud dependency.

  • Echo cancellation for clear speech separation
  • Voice activity detection filters background noise
  • Parakeet ONNX for local speech-to-text
  • Kokoro-82M for natural text-to-speech

Memory

Remembers everything that matters

Fae has a durable memory system that learns about you over time — your name, preferences, work context, relationships, and routines. All stored locally on your machine in inspectable files you can review at any time.

Before every response, Fae automatically recalls relevant memories to personalise her help. After every conversation, she records useful facts, events, and context updates. When information changes, she supersedes old facts with lineage tracking — never silently overwriting.

  • Automatic recall before every response
  • Automatic capture of useful facts and events
  • Conflict resolution with lineage tracking
  • Ask Fae to remember or forget anything

Intelligence

Thinks ahead so you don't have to

After each conversation, Fae runs a background extraction pass — pulling out dates, birthdays, people mentioned, interests, and commitments. She uses this to create reminders, schedule research tasks, and prepare warm morning briefings, all without being asked.

Say "good morning" and Fae delivers a summary of upcoming events, people to reconnect with, and research she's done overnight. Relationship tracking remembers who you mention and gently surfaces connections you might want to check in on.

  • Conversation mining for dates, people, commitments
  • Self-scheduling reminders and research tasks
  • Morning briefings with warm, natural delivery
  • Configurable proactivity: Off, Digest, Gentle, Active

Automation

Manages your computer for you

Fae can open and interact with desktop applications, read and write files, configure software, and execute shell commands — all through natural conversation. A safety-first approval model keeps you in control of exactly what Fae can do.

Tool modes let you choose how much access Fae has, from conversation-only to full access with or without approval prompts.

ModeWhat Fae Can Do
offConversation only — no computer access
read_onlyRead files and check system state
read_writeRead and write files
fullFull access including shell commands (with approval)
full_no_approvalFull access without approval prompts

Vision

Sees and understands your world

With camera permission, Fae can see and understand the world around you using Qwen3-VL, a powerful vision-language model running entirely on your hardware. She can read text from images, identify objects, and process visual information — all locally and privately.

The vision model is selected automatically based on your system RAM: Qwen3-VL-8B for 24GB+ systems, Qwen3-VL-4B for lighter hardware. Images never leave your device.

  • Qwen3-VL vision model runs on-device
  • Automatic model selection by system RAM
  • Camera images never leave your machine

Growth

An AI that gets better for you

Fae doesn't wait for a software update to learn new tricks. When she detects patterns in how you work — tasks you repeat, workflows you describe, problems you keep solving — she can generate entirely new skills herself. Python packages, markdown guides, tool integrations — built to fit your life, not a generic template.

Every self-authored skill goes through a rigorous promotion pipeline before it touches your live environment: staging first, then structural validation, then a bounded canary run, then promotion to active use. If a skill causes repeated failures, it's automatically quarantined and rolled back to the last known-good version. You stay in control — Fae proposes, you approve.

The built-in scheduler handles reminders, recurring check-ins, memory maintenance, and background research. Set tasks through natural conversation — "remind me every morning to review my tasks" — and manage them through voice or the settings menu.

  • Self-authored Python skills generated from your needs
  • Promotion pipeline: stage → validate → canary → live
  • Auto-quarantine and rollback on failure
  • Built-in scheduler with natural language triggers

Security

Self-improving, with a safety net

An AI that can modify its own behaviour needs an architecture that prevents self-destruction. Fae uses a four-layer security model that separates what she can change from what must stay protected — giving her the freedom to grow while guaranteeing she can always recover.

Layer 1
Protected Kernel — Immutable safety core. Permissions, credentials, memory integrity, boot sequence, and scheduler authority. Fae cannot modify this layer, even if she tries. Human-reviewed only.
Layer 2
Guarded Shared Layer — Extensible Rust mechanisms with strict policy gates. The LLM engine, skill runtime, and UI components. Fae can extend these through approved interfaces, but not rewrite the gates themselves.
Layer 3
Self-Authored Layer — Skills, personality prompts, themes, and integrations. This is where Fae's self-improvement lives. She can create, modify, and retire content here freely — it's designed for evolution.
Layer 4
Ephemeral Runtime — Logs, temporary artifacts, and queue state. Disposable by design — nothing here persists beyond its immediate purpose.

If self-authored skills cause problems, Fae's built-in Doctor can diagnose, quarantine, and roll back — automatically. Doctor runs in Rescue Mode: a minimal, always-available local runtime that uses a dedicated rescue model with constrained tools. It can't be disabled by mutable layers, and it doesn't need the cloud.

  • Immutable safety core that Fae cannot modify
  • Built-in Doctor with automatic diagnosis and rollback
  • Local-only rescue model — no cloud dependency for recovery
  • Open source — verify every layer yourself

Connection

Fae talks to other Fae

Every person can have their own Fae. And when they do, something remarkable happens — she can securely communicate with other Fae, weaving connections between people the way the ancient guid folk wove connections between the seen and unseen worlds.

This isn't a social network. It's a network of trust, mutual aid, and shared purpose. Your Fae can coordinate with your partner's Fae, your colleague's Fae, your family's Fae — sharing only what you approve, keeping everything else private.

  • Secure peer-to-peer communication between Fae instances
  • User-controlled sharing — nothing leaves without your approval
  • Coordinate schedules, share context, and help each other

Ready to meet Fae?

Join the waitlist and be among the first to experience a personal AI companion who truly respects your privacy.

Get Early Access Read Privacy Details