Sage

Sage is a private journal for macOS with a built-in AI agent. Your entries, chats, and the agent’s memories stay on your Mac, and you can encrypt all of them.

What Sage does

• Journal: write entries in a Markdown editor. Import Markdown files from apps such as Notion, and export entries as Markdown.
• Chat: ask an agent about your journal and your earlier chats. It runs local models through Ollama, and you pick the model, context length, and Thinking for each conversation.
• Search: local embeddings let Chat find entries by meaning, not only by matching words.
• Dream: a job you start from the sidebar. It summarizes entries and chats, titles chats, and indexes chat messages, and in a memory build it also extracts memories.
• Memories: an opt-in build feature that keeps facts about you and dated events. It is off by default.
• Lock and encryption: lock Sage with a password or Touch ID, and encrypt entries, chats, and memories inside the database file.

Sage is a native desktop app built with the Vercel Native SDK and eve.

Requirements

• macOS
• Node.js 24, for make setup and make dev
• Native SDK CLI: npm install -g @native-sdk/cli@0.11.0
• Chat and Dream models: qwen3.5:9b and nomic-embed-text by default. Setup and Settings > Models download them. You don’t install Ollama, because Sage ships a pinned copy and runs it itself.

The pinned CLI version is in .native-sdk-version, and CI installs exactly that version. Keep the two in step when you upgrade. The first build downloads Zig 0.16 if it is not already installed.

Getting started

1. Install dependencies: make setup
2. Start the app and the Chat agent: make dev

make dev starts the Chat agent on port 2000, then opens the app window. It also fetches the pinned copy of Ollama into third_party/ollama/ and runs it on port 11436 for the dev app. If the agent is already running, Sage leaves it alone. Stopping make dev stops only the agent it started.

Memory is off by default. To build with memory on, add SAGE_MEMORY=true in front of make dev, make build, or make package. Memories explains the feature.

A packaged app starts the agent itself on port 2001, using the Node.js bundled inside the app. It also starts the bundled Ollama on port 11435. If the agent cannot start, the Chat screen says why, and the journal still works.

Commands

Command	What it does
make setup	Install frontend, agent, and eval-viewer dependencies
make dev	Start the Chat agent, the bundled Ollama, and the app window for development
make seed-dev	Reset the dev app and fill its journal with the entries in scripts/seed, dated up to this week
make build	Build the frontend and the release binary in zig-out/bin/
make check	Check the pinned CLI version, that every version matches app.json, app.json itself, and the lint and types for the frontend, agent, and eval viewer
make test	Run the Zig, frontend, agent, eval-viewer, and security tests
make precommit	Run make check, make test, and make package
make eval	Seed a throwaway journal, run Dream, and grade with eve eval
make eval ARGS='--tag example'	Run only the Sam format-template cases
make eval-upload	Upload saved eval reports to Vercel Blob (asks first)
make eval-viewer-dev	Start the eval report viewer in development
make eval-viewer-build	Build the eval report viewer for production
make eval-viewer-start	Serve the production eval report viewer (run make eval-viewer-build first)
make package	Build a Mac .app bundle with the Chat agent, Node.js, and Ollama inside. The first run downloads Node.js into third_party/node/ and Ollama into third_party/ollama/, and reuses both after that.
make package-archive	Package the .app, then build a .dmg and the update .zip from it

Project structure

The app is a Zig core built on the Vercel Native SDK, a React frontend, and a Node.js agent. Data lives in SQLite, and local models run through Ollama.

• src/: the Zig core that owns the window and the database. src/journal.zig has the SQLite commands.
• frontend/: the React app with the TipTap editor, built with Vite
• agent/: the eve agent behind Chat, a Node.js program that talks to Ollama
• eval-viewer/: the Next.js app that lists eval reports from Vercel Blob
• scripts/: shell scripts that package the app and run evals
• security-tests/: tests for the token, origin, and packaging rules, run by make test
• docs/: how Sage works, including how to release it
• app.json: app identity, window size, and bridge allowlists

Documentation

Start with Architecture. The docs index lists every page.