You finish something good — and then you have to go log it. Update the status. Write the standup. Drag the card. Meridian makes that second job disappear.

It runs quietly on your Mac or PC, understands what you're working on, and keeps your tickets in Jira, GitHub Issues, and Linear current — so you never start a timer, fill out a form, or drag a card again.

Not a time tracker you fill out. It's a private timeline of your day - browse it anytime to see what you actually did, task by task - plus a background layer that keeps your project management honest while you stay in the work.

Demo

https://github.com/user-attachments/assets/501f41e6-aa89-404b-b430-a0b8b59c198e

Why

Every week, hours vanish into the work about the work — the status updates, the time logs, the standups, the cards. None of it is hard. It's just relentless, and it pulls you out of flow every single time.

But everything needed to do it already exists in what you just did — the code you wrote, the PRs you reviewed, the branch you're on. Meridian reads that context, works out which task it belongs to, and puts the update where it goes. The busywork doesn't get faster. It gets gone.

• Zero effort — no timers, no forms, no prompts. It just runs.
• Local-first — capture, OCR, and session structuring happen on your machine. The AI that summarises your work and matches it to tickets runs through the coding-agent CLI you already use (Claude, Codex, Cursor, Copilot) or a cloud LLM you choose; the ticket updates you approve go straight to your own trackers. (Privacy)
• Correct by design — a wrong task assignment is worse than no feature. Accuracy is the point.

How it works

   capture            classify             sync
 your activity  →   which task is   →   your tracker
  (on-device)        this?               (you approve)

1. Capture — Meridian bounds your activity into clean, app-based work sessions, accurate across sleep, idle, and restarts.
2. Classify — Meridian labels each session and links it to the specific ticket it belongs to, using what's on screen, the branch you're on, and the tools in play. The matching runs through the AI CLI you already use (or a cloud LLM you configure).
3. Sync — the matching ticket in Jira / GitHub Issues / Linear is updated for you. Nothing posts without your approval.

A dashboard inside the Meridian app (open it from the menu-bar tray icon) shows your day as a timeline and per-app breakdown. A built-in MCP server makes the same data available to AI tools like Claude and Cursor.

Install

Requirements: macOS (Apple Silicon or Intel), or Windows 10/11 (x64).

On macOS Meridian ships as a universal binary, so Apple Silicon and Intel are both native. The local embedder is CPU-only (not Metal-accelerated), so it behaves identically everywhere. Linux is not supported.

Windows support is newer than macOS and one capability differs: notifications are plain title-and-body toasts, without the action buttons macOS shows, so a notification asking you to confirm something can be read but not answered from the toast itself. Everything else - capture, the timeline, tracker sync, auto-update - works the same.

macOS

curl -fsSL https://raw.githubusercontent.com/Meridiona/meridian/main/scripts/bootstrap.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/Meridiona/meridian/main/scripts/bootstrap.ps1 | iex

This downloads the latest Meridian build, installs it, and launches it. The app stages its own background daemon on first run and opens the setup wizard — permissions (macOS only; Windows needs none), the AI that writes your summaries, and connecting your tracker. Updates install themselves from within the app.

Prefer to do it by hand? From the latest release, download Meridian.dmg (macOS) and drag Meridian to Applications, or Meridian-setup.exe (Windows) and run it.

Supported PM tools

Connect one or more trackers and Meridian maps captured work sessions to tasks, then posts time-logged worklogs as comments on the task.

Azure DevOps quick-start

Just two variables — paste your project URL from the browser and your PAT:

# Add to .env (or run meridian setup to be prompted interactively)
AZURE_DEVOPS_URL=https://dev.azure.com/your-org/your-project
AZURE_DEVOPS_PAT=your-pat-here

AZURE_DEVOPS_URL works for all three URL shapes: dev.azure.com/org/project, org.visualstudio.com/project, and on-premises servers — meridian auto-extracts the org and project.

To create a PAT: User settings → Personal access tokens → New token, scope: Work Items → Read & write.

# Verify tasks synced
meridian force-pm-sync
sqlite3 ~/.meridian/meridian.db "SELECT task_key, title FROM pm_tasks WHERE provider='azure_devops';"

Data location

👉 Full walkthrough — permissions, tracker setup, configuration, troubleshooting: SETUP.md.

Quickstart

meridian start     # bring everything up
meridian status    # check what's running
meridian logs -f   # watch the pipeline live
meridian doctor    # diagnose config / services / permissions

Open the dashboard from the Meridian tray icon in the menu bar. Stop everything with meridian stop.

Nothing posts to your tracker automatically. Meridian drafts worklogs and ticket updates; you review and approve each one in the dashboard. Approval is the only gate.

Privacy

Meridian is built to keep your data yours:

• Capture and session structuring run on-device — screen capture, OCR, and categorization happen locally; your screen content is not sent to Meridiana.
• Meridiana, the company, never receives your data. There are no analytics servers and no default telemetry.
• The only outbound traffic is yours, and you control it: approved ticket updates go directly from your machine to the trackers you connect (Jira, GitHub, Linear), and integration tokens are stored locally.
• AI summaries use the provider you choose: matching work to tickets and drafting worklogs run through the coding-agent CLI you already use (Claude, Codex, Cursor, Copilot) or a cloud LLM you configure, so session text is sent to that provider. There is no on-device generative model.

Full detail: docs/privacy.md.

Contributing

Meridian is open source and contributions are welcome — bug reports, fixes, docs, and features. See CONTRIBUTING.md to get a dev environment running, and CLAUDE.md for architecture and conventions. By participating you agree to our Code of Conduct.

Build from source

git clone https://github.com/Meridiona/meridian
cd meridian
./install.sh

Builds the daemon and dashboard from source and registers the same services. See SETUP.md for the development workflow, configuration reference, and MCP server setup.

Built on

Meridian stands on excellent open-source work:

• screenpipe — the capture crates Meridian's in-process capture is forked from (pinned at the last MIT release, 0.4.6).
• Tauri — the framework that wraps the dashboard and tray into a single native app.
• candle — the Rust ML framework running Meridian's local text-embedding model.

Thank you to these communities. 🙏

License

MIT — see LICENSE.