A companion that sees what you see.
Lore lives in your menu bar. Push a key, ask a question, and it answers — out loud, with a luminous orb pointing at exactly what it’s explaining. Three modes for three ways of working: a patient tutor, a hands-off agent, and an ambient companion that maintains your knowledge graph in the background.
One overlay. Three voices.
The same Lore — push-to-talk, screen-aware, on your machine — shows up three ways depending on what you’re doing.
A patient tutor looking over your shoulder.
Ask about anything on your screen. Lore sees what you see, explains it in plain language, and points at exactly what it's talking about — no copy-paste, no describing what you mean.
- Push-to-talk, hands freeHold ⌃⌥, ask, and Lore answers — out loud, with a cursor pointing at the thing it's explaining.
- Multi-monitor pointingLore flies a luminous orb across any of your displays and lands on the exact pixel it's referencing.
- Saves what you learnEvery explanation goes into your personal wiki so the same answer never has to be asked twice.
Tell Lore what you want done. Watch it happen.
Lore reads your desktop, plans the steps, and executes them — clicking, typing, switching apps. Background-routed, so it never steals your foreground or warps your cursor.
- Accessibility-first controlNative AX actions when possible, falling back to coordinate clicks only when structure is weak. Cleaner, more reliable than pure pixel agents.
- Trust tiers you controlMad-Max for trusted flows, Guided for sensitive apps. Approve specific actions or pre-approve the whole task.
- Recursive learningAfter each task, Lore updates a per-app playbook in your local wiki. The agent gets sharper every time you use it.
Always there. Never in the way.
An ambient voice that watches the screen with you — talks when there's something worth saying, stays quiet when you're in flow, and quietly maintains your knowledge graph in the background.
- Lives in the menu barNo dock icon, no main window. Click once for a floating panel; double-tap ⌃⌥ to type instead of speak.
- Continuous synthesisWhile you read or watch, Lore writes structured notes, finds connections to past sessions, and surfaces them when relevant.
- Six voices, one identityPer-mode voice mapping, on-device Kokoro TTS by default, streaming xAI for anyone who prefers cloud — all behind one Companion.
“That's the eigenvalue equation — A scales v by λ without rotating it.”
A place where everything you learn lives.
Every conversation, clip, and answer is a markdown file in your vault — yours, on disk, in plain text. The Workspace is the window into that vault; clipping is one of the simplest ways to feed it.
A real workspace for everything you've learned.
Every clip, note, thread, and synthesis becomes a node in your vault graph — a force-directed map of how the things you know connect. Hover any node to see its neighborhood; click to open the underlying file in the same Workspace tab.
Eigenvalues
An eigenvector v of a square matrix A satisfies Av = λv, where λ is the corresponding eigenvalue1.
Geometrically, applying A to v stretches it along its own axis without rotating it2. The scale factor is λ.
See also [[singular-value-decomposition]] and [[principal-component-analysis]]. The connection to PCA falls out of the spectral theorem3.
Why it matters
Eigendecomposition is the engine behind PageRank, principal component analysis, quantum mechanics, and almost every method in linear algebra that “diagonalizes” a transformation.
An eigenvalue is the scale factor that survives a linear transformation1. If you apply a matrix A to a vector v, most of the time v tilts in some new direction. But certain special vectors — the eigenvectors — point in a direction that doesn’t move; the matrix only stretches them.
The amount of that stretch is the eigenvalue2. Negative eigenvalues flip the vector; zero eigenvalues collapse it3. The set of eigenvalues describes the “shape” of the transformation — what it stretches, what it leaves alone.
- Live-preview markdown editorCodeMirror-powered, with caret-aware decorations. Frontmatter chips, [[backlinks]], and inline citations all render in place — never a separate preview pane.
- ⌘K semantic paletteSearch the vault by meaning, not just keywords. BM25 + embedding cosine + recency, blended into one ranked list. Save the queries you reuse as Lenses.
- Vault graphA force-directed canvas of every link in your knowledge — clips, threads, notes, synthesis. Drag to pin, scroll to zoom, click to open the underlying file.
Talk to your second brain.
Every chat happens against your vault. Ask a question and Lore retrieves clips, notes, and prior threads — answers stream back with numbered citation pills you can hover, tap, or drag straight into the composer.
- Citations are first-classInline [1] [2] pills wired to the right-rail Sources cards. Hover either side to highlight its counterpart; tap to slide up a source preview sheet with the cited passage.
- Branch and quote anywhereRight-click any assistant turn to branch a new thread, quote into a note, or ask a follow-up with the turn pre-quoted. Every thread is a markdown file at ~/.lore/wiki/threads/.
- Push-to-talk in the composerThe same dictation provider that powers menu-bar voice — AssemblyAI streaming when available, local fallback when not. Speak the question, send it, walk away.
An eigenvalue is the scale factor that survives a linear transformation1. If you apply a matrix A to a vector v, most of the time v tilts in some new direction. But certain special vectors — the eigenvectors — point in a direction that doesn’t move; the matrix only stretches them.
The amount of that stretch is the eigenvalue2. Negative eigenvalues flip the vector; zero eigenvalues collapse it3. The set of eigenvalues describes the “shape” of the transformation — what it stretches, what it leaves alone.
Save anything. Find it later.
Drag a URL onto Lore, or type /clip in Quick Chat. Lore fetches the page, summarizes it, tags it, and drops a markdown clip into your vault — automatically backlinked to whatever you were working on.
- Drag-or-type captureDrop a URL into the menu bar, or use /clip <url> from Quick Chat. Both routes produce the same clean markdown file at ~/.lore/wiki/clips/.
- Auto-synthesisEach new clip runs through the per-capture enricher: title, summary, tags, suggested backlinks. Surfaces in the right places without any manual filing.
- Citation-readyClips become first-class sources — they show up as numbered citations in chat answers and as nodes in the vault graph. Your reading turns into a research library.
A practical guide on why fonts and spacing make professional documents read as legitimate — by the author of Practical Typography.
Built into the way Mac already works.
Personal wiki
Every conversation, every research finding, every dictionary term you teach Lore lives at ~/.lore/wiki/. Yours, on disk, in plain markdown.
Voice transcription
Dictate and Lore cleans it up — proper grammar, no filler words, smart formatting. Lists, paragraphs, and punctuation handled automatically.
Local-first synthesis
A nightly maintenance pass runs the synthesis, dictionary, and connection-finding work on your machine. Nothing about how you think leaves.
The fastest way to feel it is to use it.