v0.8.1 · released

Meru मेरु

In some Eastern traditions, Meru (मेरु) is the cosmic mountain that the sun, moon and stars turn around.

A personal AI assistant that runs on your own computer.

Free software. No account, no subscription and no cloud model.

What it does for you

Meru answers questions from your notes and files, and uses the tools you allow. The models run on your computer, so your questions reach no AI company unless you name one and say yes.

  • Your data stays home

    Your files, mail and questions stay on your computer. Ollama, a free program, runs the models there too.

  • No account and no bill

    Meru needs no sign-up, no subscription and no cloud model. A question costs the electricity your computer uses. Meru sends one anonymous count a day, which you can turn off (what it holds).

  • Answers from your folders

    You pick the folders Meru may read. It searches them by keyword and by meaning, and names the files behind each answer.

  • Tools you choose

    Mail, calendar, notes, web search and GitHub start switched off. Meru asks you before any call that sends, deletes or changes something.

  • It remembers what you tell it

    Meru keeps what it learns about you as small Markdown files. You can read, edit or delete each one.

  • Every chat is a file you own

    Each conversation lands in one plain text file on your disk, ready to back up, search or delete.

What runs where

Two programs do the work: merud runs in the background, and you ask it questions through meru in the terminal or the desktop app.

The parts of Meru A dashed line marks your machine. Inside it, you type into meru or meru-desktop, for a chat or one question, and it passes the question to merud, which runs in the background. merud holds three parts: the scheduler, which starts jobs such as morning briefs; the agent loop, which answers your question; and dispatch, which checks each tool call and asks before risky calls. The agent loop uses local models run by Ollama (fast, main and embed, in a lite or full setup), your data (the folders you index, memory files, conversation transcripts and a SQLite search index) and skills (writing, explainer, web-research, file-research, mail-research and any you add). Dispatch reaches other agents over A2A and tools over MCP, such as Gmail, Calendar, web search and your notes. The Usage page shows what Meru measures, and a Prometheus you run can read the same numbers on the same machine. Outside the dashed line sit internet services such as a search engine or Google; only the tools you allow connect to them. your machine you meru or meru-desktop chat, or one question merud runs in the background scheduler morning briefs agent loop answers your question dispatch checks each tool call, asks before risky calls metrics (optional) Prometheus, local only local models · Ollama lite or full fast main embed your data your folders indexed memory files Markdown transcripts one per chat search index SQLite skills + yours writing · explainer web · file · mail research other agents · A2A tools · MCP, built in + yours Gmail · Calendar · web search · notes only tools you allow internet services search engine, Google
Everything inside the dashed line runs on your machine: the models, your data and the assistant itself. Only the tools you allow reach the internet, and each call goes through dispatch first. Apart from them, merud sends one anonymous count a day, with no ID and no text, which you can turn off.

How it works

Three pages explain the design, each one deeper than the last. Start with the one that fits you.

What it does for a company

Staff want an assistant, and the data they would use it on can't leave the building. Meru fits that problem.

  • The data stays put. Meru runs on a laptop, a virtual desktop or a server in your own account. It reads email, documents, code and notes without sending any of it to a model provider, so it can work on material that policy or regulation keeps in-house. The anonymous daily usage report carries none of it, and one line of config turns it off.
  • Tools stay off until you allow them. Each server's tools start switched off, and risky ones ask before they run.
  • Every action lands in a log. Each tool call goes into an audit table and the session transcript. One function, dispatch, handles every call, so a reviewer checks one code path.
  • You can measure it. Meru sends OpenTelemetry metrics and traces (tokens, time, which tool ran and which failed) to a collector you run on the same machine.
  • It installs anywhere. Meru ships as native programs with no interpreter to install. It runs on laptops, virtual desktops, servers on your own network and machines with no internet connection.
  • No per-question bill. You pay for hardware you already own.
  • A license legal teams know. Meru comes under the Apache License 2.0, with an express patent grant. You may use, change and ship it, in your own products included, as long as you keep the notices.

Meru serves one person on one machine today. Central control over who may use which tools, across a company, doesn't exist yet.

Try it on your computer

On a Mac

Open Terminal and paste this line:

curl -fsSL https://github.com/aarora79/meru/releases/latest/download/install.sh | bash
  • On a Mac with Apple silicon it also installs Meru.app, the desktop app.
  • It lists what it will install, then asks once: everything, one step at a time, or quit.
  • It takes several minutes, most of it downloading the models, about 2 GB.
  • It checks every download against the release's checksums. Run it again to update; Meru never checks for updates by itself.

Or install in a window: download the disk image below, open it and open "Install Meru.app". The first time, macOS refuses because Apple hasn't checked Meru: click Done, then Open Anyway in System Settings, Privacy & Security. It walks through each step, says what it downloads first, and sets up web search, your folders and Gmail if you want them.

On Linux

Open a terminal and paste this line:

curl -fsSL https://github.com/aarora79/meru/releases/latest/download/install-linux.sh | bash
  • Tested on Ubuntu 24.04.
  • It installs meru and merud, then offers the desktop app, Ollama, setup and starting merud at boot, asking before each step.
  • With an NVIDIA card, setup offers the largest model that fits the card's memory.
  • Run it again to update; it keeps your settings.

On Windows

Open PowerShell and paste this line:

irm https://github.com/aarora79/meru/releases/latest/download/install.ps1 | iex
  • Tested on Windows Server 2025 with an NVIDIA card; Windows 10 and 11 should work too.
  • It installs meru, merud and the desktop app, then offers Ollama, setup and starting merud at each sign-in, asking before each step.
  • With an NVIDIA card, setup offers the largest model that fits the card's memory.
  • Web search needs Docker or Podman; the Windows guide covers a machine without them.

The project