Forkbench for Gemini CLI
Gemini CLI on a Mac, and what replaced it
Gemini CLI was Google's open-source terminal agent. Google retired it on 18 June 2026 for its AI Pro, Ultra and free tiers, replacing it with Antigravity CLI; it remains reachable on paid Gemini API and Gemini Enterprise Agent Platform keys.
Updated
Google retired the standalone Gemini CLI on 18 June 2026 for its AI Pro, Ultra and free tiers and replaced it with Antigravity CLI, which you invoke as agy; Gemini CLI itself still runs on paid Gemini API and Gemini Enterprise keys. Forkbench runs either in a real login shell on your machine, with Antigravity set up automatically and Gemini CLI launching like any other command, both reaching the same Thread's notes and board.
What breaks in a plain terminal
A retirement with no automatic migration leaves you doing the move by hand, and the guides all cover the same half of it: rename the instruction file, re-declare the MCP servers, check what has no equivalent on the other side. The half nobody covers is that whatever your Gemini CLI sessions had worked out - the approaches already ruled out, the module everyone avoids, why a dependency is pinned - was never in a file, and no rename brings it across. It is also the half that decides whether the new agent feels useful in week two.
How Forkbench runs Gemini CLI
Both agents, one Thread
Run agy in one tab and gemini in another inside the same Thread, and they share its notes and its board. Moving across becomes a change of tab rather than a migration, and you can keep the old one until the new one has been through a genuinely awkward job.
Notes that belong to the job, not the vendor
What an agent works out goes into the Thread as a note, so it survives the agent that wrote it. That is the material a retirement would otherwise take with it. Notes are sealed on your machine before they are stored.
Keys it can use and never read
A key in Vault reaches the one command that needs it. The agent gets the result and never the value, which matters twice over while you are running two vendors' agents against one project. Vault is free.
Folders it cannot wander out of
Lock a Thread to the folders the job is about. Every shell in it starts under a macOS Seatbelt profile that the kernel enforces and every process inherits, including build scripts and a dependency's install hook, so a read or a delete outside them is refused rather than discouraged.
A real login shell
gemini launches in a genuine PTY with your shell config intact. Nothing is wrapped or proxied, which is why a retired binary you still have installed keeps working here exactly as it does anywhere else.
Runs with the lid closed
Sleep Control keeps the machine awake with the laptop shut, so a long run finishes overnight instead of stopping when the machine sleeps.
Getting Gemini CLI running
- 1Open a Thread on the repo and run whichever of the two you still have installed.
- 2Add a tab for agy in the same Thread, so both read the same notes and claim from the same board.
- 3Copy your instruction file across, or symlink one to the other, so you maintain a single set of rules.
- 4File what the old sessions knew as notes on the Thread. That is the part the migration guides cannot help with, and it has to be written down by someone.
- 5Keep the old agent installed until the new one has been through a real debugging session. The gaps never show up in the first hour.
What this does not do: Dates and tiers here describe Google's announcement as of 6 September 2026 and are its to change - check its own documentation before relying on them. Forkbench does not auto-configure Gemini CLI: its MCP registration covers Claude Code, Codex, Cursor and Antigravity CLI. Gemini CLI still runs in a pane like any command, and still reaches your notes and board through the fb command on its PATH.
Gemini CLI questions
Is Gemini CLI dead?
Not entirely. Google stopped serving it on 18 June 2026 for AI Pro, Ultra and the free tier, which is most people, and it remains reachable on paid Gemini API and Gemini Enterprise Agent Platform keys. The replacement is Antigravity CLI, invoked as agy.
Does Forkbench still support Gemini CLI?
It runs it, because every tab is a real login shell and any command runs in one. What it does not do is register its MCP servers automatically - that covers Claude Code, Codex, Cursor and Antigravity CLI. Gemini CLI still reaches your notes and board through the fb command, which is on every pane's PATH and needs no MCP support at all.
What do I lose moving from Gemini CLI to Antigravity?
The instruction file moves with a copy or a symlink. What does not move is everything the old sessions worked out and never wrote down, because it lived in one vendor's session history. Writing that into notes on the Thread before you switch is the only way to keep it.
Can I run Antigravity and Claude Code on the same project?
Yes. Put both in one Thread, give each its own git worktree so their edits cannot collide, and they claim from the same backlog. No task records which agent wrote it, so either can pick up the other's work with the notes and discussion still attached.
What does it cost to run these in Forkbench?
Nothing extra for the agent. You keep your own Google plan or API key, and Forkbench never meters requests. The terminal, your notebooks and a board on every Thread are free. Vault is free too; it needs a signed-in account and nothing more.
Running something else?
Forkbench is agent-agnostic, so any CLI agent runs in a tab. These have a page of their own: