Gemini CLI memory your other machines can read

Gemini CLI remembers, just not the way older guides teach it. /memory add and the save_memory tool are gone, removed in a change that reached the stable release with v0.44.0 on 27 May 2026. Its memory is now Markdown the agent edits, and the docs say where each kind goes: "shared project instructions go in repository GEMINI.md files, private project notes go in the per-project private memory folder, and cross-project personal preferences go in the global ~/.gemini/GEMINI.md file."

Yes, in Markdown files its agent edits, and its private and global memory files stay on the machine they were written on. Connect it to Pando and have it write decisions into your outline, which every machine reaches, and so does every other agent you allow to read it.

First, the accounts Gemini CLI still serves

The announcement in Gemini CLI’s repository on 18 June 2026: "Starting today, Gemini CLI will stop serving requests for Google AI Pro, Google AI Ultra, and free tier individual accounts." And who is left: "Enterprise users with Gemini Code Assist licenses and API key authentication remain completely unaffected." On one of those personal accounts Gemini CLI serves no requests at all, and nothing Pando does changes that. Google sent those users to Antigravity CLI instead, which has not been tried with Pando yet. With a Gemini API key, or as an enterprise user with a Gemini Code Assist license, read on.

Set it up

With -s user the server goes into ~/.gemini/settings.json, so it is there in every project; without it, Gemini CLI writes it into the project’s own .gemini/settings.json. Keep the address exactly as it is, with no slash at the end, because Gemini CLI compares it with the address Pando publishes for itself. Starting gemini opens no browser, and it says nothing about Pando either, because Gemini CLI keeps MCP problems quiet at startup: /mcp list shows pando as not authenticated, and /mcp auth pando, typed in the session, is what opens the browser. gemini mcp list in the shell, or /mcp list in a session, shows whether it is connected. If the sign-in does not finish, connect with a key instead, as further down.

Then add one line to ~/.gemini/GEMINI.md, the file its docs keep for preferences across projects: recall from Pando before assuming, and keep decisions and their reasons there.

What you approve in Pando

The browser opens Pando, where you sign in first if you are not signed in already, and then its consent page, which offers two levels, “Read and write your outline” or “Read your outline only”. Either way Gemini CLI starts with your whole outline. A box underneath, ticked by default, gives it a bullet under Home to remember in, named after whatever Gemini CLI calls itself when it registers. Untick it and it remembers nowhere until you assign a bullet under Agents and API keys. The token it gets has no expiry and works until you revoke it there. A bullet you protect refuses every change it tries, and a lock stops changes, not reading.

Without a browser, or with less reach

The sign-in needs a browser on the machine Gemini CLI runs on. Its docs say it will not work in headless environments, over SSH without X11 forwarding, or in containers without a browser. There, connect with a key instead, with the steps below. A Gemini CLI that did sign in can be held to one branch without a key: open that branch, open Agents and API keys in Settings, tap its row under Your agents, and press Let it reach only “…”, the bullet you are in.

• Open Agents and API keys, give a new agent a handle, untick both “Let it read and write this outline” and “Give it a bullet of its own to remember in”, and choose Create the agent. Copy its key, which is shown once.
• Zoom into the branch it may work in, open Agents and API keys again, tap the agent’s row under Your agents, and choose Remember in “…”, the bullet you are in, where the quotes hold that branch’s words. It then reaches that branch and everything under it, and nothing else.
• Add Pando with the command below instead of the two steps under Set it up, and do not run /mcp auth pando: it signs Gemini CLI in over the consent page, which approves it onto your whole outline. If it already signed in that way, revoke that connector under Agents and API keys.

gemini mcp add --transport http -s user --header "Authorization: Bearer <key>" pando https://pando.ink/mcp

What each place is for

• A GEMINI.md in the repository, for instructions the project shares.
• The private MEMORY.md under ~/.gemini/tmp/<project-slug>/memory, for notes about one project that are yours alone, on this machine.
• ~/.gemini/GEMINI.md, for your preferences across projects, on this machine.
• A saved session, to pick up where you stopped: gemini --resume brings one back, each project has its own, and they are kept 30 days by default. To another machine it goes only as a file you export by hand with /export-session.
• A branch of your outline, for decisions and their reasons that have to reach a second machine, another agent, or you on your phone.

When the Markdown files are enough

If you work on one machine and Gemini CLI is your only agent, its own files are enough, and you should use them. Shared project instructions belong in the repository’s GEMINI.md either way, next to the code, not in your outline. The outline earns its place the week a second machine or a second agent is part of the work. Claude Code, Codex and ChatGPT connect to the same address, so what Gemini CLI decided on the work laptop is what Codex reads on the one at home.

What it costs

Nothing, for 1,000 bullets, with every feature included and no payment card.

Deeper

• Claude Code memory on every machine you use
• Keeping an agent out of the rest of your notes

Connect an agent, about two minutes · The twelve tools · Which note apps an agent can reach · Who runs this