README.md

August 23, 2026 · View on GitHub

Logue

Privacy-first AI meeting notes & writing assistant for macOS.
On-device transcription, Smart Minutes, and 60+ writing modes — all powered by Apple Silicon. Nothing leaves your Mac.

_{First reached July 28, 2026}

Download for macOS • Build from Source • Contributing • Request a feature

Logue home dashboard on macOS

Introduction

Logue is a privacy-first AI meeting-notes and writing assistant for macOS. It combines real-time audio transcription, speaker diarization, and intelligent document editing — all powered entirely on-device via MLX on Apple Silicon. By default your meetings, notes, and conversations stay on your Mac — the only network calls are on-device model downloads, update checks, and the opt-in features you explicitly enable (web search and external AI providers).

This repository contains the full Logue app source: the LLM engine, recording pipeline, agentic AI chat, writing editor, and test suites. Bitwize builds Logue for professionals and students who want the full power of AI without sacrificing privacy.

New here? Read CONTRIBUTING.md to get set up and learn the standards every pull request follows.

✨ Features
🖼 Screenshots
📥 Install (users)
🚀 Build from Source
⚙️ Configuration
📚 Documentation
🗺 Project Layout
👏 Contributing
❓ FAQ
📄 License

✨ Features

🎙 Meeting Intelligence

Real-time transcription — powered by Apple's SpeechTranscriber (macOS 26+), streaming audio directly with no buffering delays
Speaker diarization — identifies and labels individual speakers using FluidAudio's Sortformer model, with a batch fallback for accuracy
System audio capture — records mic and system audio simultaneously via ScreenCaptureKit; no meeting bot joins your call
Smart Minutes — generates structured summaries with key decisions, themes, and follow-ups
Smart transcript highlights — AI automatically extracts the most important moments; manual bookmarks are always preserved
Action item extraction — pulls tasks, owners, and due dates from the transcript automatically
Meeting prep briefing — AI-generated context brief before a meeting starts, based on linked documents and prior notes
Meeting ↔ Document links — bidirectional links between meetings and related documents with back-link chips in the editor

✍️ AI Writing Editor

Block-based rich text editor — built on NSTextView with full SwiftUI integration
Grammar & clarity suggestions — inline AI-powered suggestions with one-tap acceptance
Tone analysis — detects and adjusts the tone of selected text
Vocabulary enhancement — suggests richer alternatives, validated against the actual document content
Rewrite & review panels — rewrite selected passages or get structural feedback on the whole document
Fact verification — flags claims that may need a source
Table support — full table creation, editing, and context menus inside the editor
Writing goals — set and track word count, reading level, and tone targets
Wiki links & backlinks — type [[Document Name]] to link documents; links show the target's real name, are clickable, survive renames, and every document lists what links back to it
Properties & relationships — give documents typed fields (text, number, date, select, checkbox) and named links to other documents, all editable in a side panel
Saved views & inbox — save any filter as a sidebar entry, and triage unfiled documents from an inbox with keyboard-driven bulk actions
Plain markdown storage (opt-in) — store documents as ordinary .md files in ~/Logue instead of encrypted storage, and edit them in any editor, track them in git, or point an agent at them. The folder is the storage, not a copy: edits, new files, renames, moves and deletions flow both ways, and folders map to spaces including nested ones. Off by default, and Logue explains the encryption you give up before turning it on

🤖 Ask Logue — Agentic AI Chat

Multi-step agentic reasoning — powered by LangGraph-Swift; plans, reasons, and executes multi-step workflows
Streaming responses — live token-by-token output with animated indicators
Inline diagrams & LaTeX — renders Mermaid diagrams and math expressions directly in the chat thread
Deep research mode — progress tracking for long-running multi-source research tasks
Tool execution UI — shows exactly what actions the agent is taking, with approval controls
Sources panel — lists every document and meeting the agent consulted
Persistent memory & history — the agent remembers context across conversations, all searchable

🔍 Search, Tasks & Discovery

Unified ⌘K search — searches across documents, meetings, and action items in one keystroke, with keyword-centered snippets
Full-text search (FTS5) — fast, ranked full-text search across all meeting transcripts and document bodies
Action item dashboard — every task from every meeting in one place, with smart filters (All / Pending / Overdue / Today / This Week / Completed), sort, search, and a live sidebar overdue badge

🔒 Privacy & Security

On-device AI by default — inference, transcription, and diarization run locally via MLX on Apple Silicon; no content is sent to any cloud service unless you opt into web search or an external AI provider
AES-256-GCM encryption at rest — meetings, transcripts, audio and documents are encrypted on disk. Documents are the one thing you can opt out of, by turning on plain markdown storage; meeting data is always encrypted
No account required — download and start working; there's no sign-up, no telemetry gate, no cloud dependency
Hardened Runtime, notarized, not sandboxed — Logue ships Developer ID signed and notarized with the Hardened Runtime on. It is deliberately not sandboxed: a sandboxed process cannot be an Accessibility API client, so macOS never lists it under Privacy & Security → Accessibility, which silently disables the inline assistant, global hotkeys and Command Center. Every entitlement exception is documented in the entitlements files

🖼 Screenshots

Meeting transcription + Ask Logue	Smart Minutes

AI writing editor with fact-checking	On-device PII detection

📥 Install (users)

Requirements: macOS 26.0 (Tahoe) or later on Apple Silicon (M1 or newer).

Download the latest Logue-<version>.dmg from the Releases page.
Open the DMG and drag Logue into your Applications folder.
Launch it. On first run, macOS Gatekeeper may prompt — right-click the app → Open.

Updates are delivered in-app via Sparkle: Logue checks the GitHub-hosted feed on launch and prompts you to install new versions. Updating only replaces the app — your meetings, documents, and downloaded models are stored separately and are never touched. To update manually, download the newest DMG and drag it over the old app.

🚀 Build from Source

Prerequisites: macOS 26.0+ (Tahoe) on Apple Silicon, Xcode 26+ (it ships the macOS 26 SDK required by SpeechTranscriber), and XcodeGen.

1. Clone with submodules (vendored dependencies live under Vendor/):

git clone --recurse-submodules https://github.com/bitwize-ai/Logue.git
cd Logue

2. Download the Metal toolchain (one-time; required by MLX):

xcodebuild -downloadComponent MetalToolchain

3. Install tooling and generate the Xcode project:

brew install xcodegen swiftformat swiftlint
xcodegen generate

4. Build and run — open in Xcode and hit Run, or from the CLI:

# The signing flags let you build without an Apple Developer account.
xcodebuild build -project Logue.xcodeproj -scheme Logue -destination 'platform=macOS' \
  CODE_SIGN_IDENTITY="-" CODE_SIGNING_REQUIRED=NO CODE_SIGNING_ALLOWED=NO
open Logue.xcodeproj

Already cloned without submodules? Run git submodule update --init --recursive. Re-run xcodegen generate whenever you add/remove .swift files or edit project.yml.

Dev builds use ad-hoc signing with the sandbox off — no Apple Developer account needed just to build and run. (In Xcode, if the build stops at signing, either pick your team under Signing & Capabilities or use the CLI flags above.) Full linting and git-hook setup are in docs/dev-setup.md.

⚙️ Configuration

Logue runs fully on-device out of the box — no keys or accounts are required to build, run, or use it. Everything below is optional and only relevant if you fork the project or cut your own signed releases. Nothing is hardcoded to secrets: all credentials are read from your own environment / GitHub Actions secrets.

Optional external AI providers. Alongside the on-device MLX models, you can add OpenAI-/Anthropic-compatible or OpenRouter/Ollama/LM Studio endpoints under Settings → Models. Keys are entered in-app and stored in your macOS Keychain — never in the repo.

Running your own signed builds / releases. To ship notarized builds under your own identity, set these as GitHub Actions repository secrets (the release workflow reads them — none are committed):

Secret	Purpose
`APPLE_TEAM_ID`	Apple Developer Team ID
`APPLE_CERTIFICATE_BASE64` / `APPLE_CERTIFICATE_PASSWORD`	"Developer ID Application" cert (base64 `.p12` + password)
`APPLE_ID` / `APPLE_APP_PASSWORD`	Apple ID + app-specific password for notarization
`SPARKLE_PRIVATE_KEY`	EdDSA key that signs auto-update ZIPs

For a signed fork you'll also want to change, in project.yml:

the bundle identifier (PRODUCT_BUNDLE_IDENTIFIER, default com.bitwize.logue);
the iCloud container identifiers (iCloud.com.bitwize.logue) in Logue.entitlements — these are tied to your Apple Developer account, so point them at a container you own if you want iCloud sync (or remove them). The shipped release entitlements don't include iCloud, so signed release builds aren't affected;
to run your own update channel, SUFeedURL + SUPublicEDKey.

Building and running unsigned dev builds needs none of the above. See docs/dev-setup.md and docs/SPARKLE_UPDATE_FLOW.md.

📚 Documentation

🗺 Project Layout

Logue/ — Main Swift source; SwiftUI + AppKit application target.
Logue/Engine/ — LLM inference actor, prompt builders, and retry helpers. All AI logic lives here.
Logue/Services/ — Recording pipeline, transcription, speaker diarization, encryption, document storage, and scheduling services.
Logue/Views/ — All SwiftUI views: Meeting workspace, Writing editor, Ask Logue chat, Action items, Settings, and more.
LogueTests/ — Swift Testing suites (@Suite, @Test), including real-inference LLM integration tests.
Vendor/ — Vendored git submodule dependencies — no remote access required at build time.
docs/ — Developer setup and release documentation.
scripts/ — Release build script, Sparkle appcast updater, and export options plist.