Google ended the free individual tier and pointed users at a binary-only replacement. Here is what actually changed, and where to go instead.
If you have been running gemini from your terminal for free, that stopped working on June 18, 2026. Google announced the change a month earlier, on May 19, and steered individual users toward a new Go-based tool called Antigravity CLI, invoked as agy.
gemini-cli project is not gone. Gemini CLI's lead PM wrote in the May 19 announcement that the team will "continue to see us work on GitHub as we keep Gemini CLI updated with latest model releases, bugs and security fixes for our enterprise customers," and that "the project remains available to the community as an Apache 2.0 licensed repository with no changes." Both halves matter: the license is unchanged, and the user it is now maintained for is not you.custom:proprietary because there is nothing to build. The same Google org publishes its Antigravity Python SDK under Apache-2.0, so withholding the CLI's source was a decision, not org policy.Two groups still have one. Enterprise users with a Gemini Code Assist Standard or Enterprise license, or access through Google Cloud, keep full Gemini CLI support. And anyone willing to plug in a paid API key can keep running the open-source gemini-cli almost as before: either a paid Gemini API key from AI Studio, or a paid key for the Gemini Enterprise Agent Platform, which is the platform formerly branded Vertex AI. The word "paid" is load-bearing there, because a free-tier Gemini API key is not among the paths Google says will keep working.
This is the detail that decides the question for many people, and it is not in Google's messaging. agy's own documentation describes exactly one way to authenticate: interactively. It uses the system keyring, falls back to Google Sign-In in a browser, prints an authorization URL for you to complete locally when you are on an SSH session, and clears credentials on /logout. There is no documented environment-variable or service-account path at all. Users are currently petitioning for an agy --api-key flag or plain GEMINI_API_KEY support.
| Auth method | gemini-cli | agy |
|---|---|---|
| Env-var API key | Yes | Not documented |
| Service account JSON | Yes | Not documented |
| Application Default Credentials | Yes | Not documented |
| Interactive browser sign-in | Yes, withdrawn for individuals | Yes, and it is the primary path |
| Headless / SSH | Env vars | Prints a URL to finish locally |
So if your use is unattended, in CI, in a container, or on a box you only reach over SSH, the migration Google is pointing at does not cover you. The path that does is staying on the open-source CLI with a paid key.
If you do move to agy, more comes across than you might expect. Extension-bundled skills, agents and MCP servers all convert automatically via agy plugin import gemini, along with your GEMINI.md. The importer relocates server definitions out of inline settings and reports what it did, in the form mcpServers : 1 server definition migrated to mcp_config.json, landing them at ~/.gemini/config/mcp_config.json globally and .agents/mcp_config.json per workspace. Worth noting in passing that the replacement for Gemini CLI still keeps its config under ~/.gemini/.
Two things do need hands. Under a heading Google titles "Required schema updates," the guide says that "when manually migrating remote websocket or SSE server definitions, update the URI key parameter to match the current standard" -- that is the url and httpUrl keys becoming serverUrl, and only if you are moving a remote definition by hand rather than letting the importer do it. The other is custom workspace skills, which carry an explicit "Action Required" warning: a folder at .gemini/skills/ must be manually renamed or relocated to .agents/skills/ before the agent will recognise them as slash commands.
Install is unchanged:
npm install -g @google/gemini-cliThen pick an auth path. For an AI Studio key, one variable is enough. For the Enterprise Agent Platform, you point the CLI at the platform and give it a project:
# AI Studio key
export GEMINI_API_KEY="..."
# Or the Gemini Enterprise Agent Platform (formerly Vertex AI)
export GOOGLE_API_KEY="..."
export GOOGLE_GENAI_USE_VERTEXAI=true
export GOOGLE_CLOUD_PROJECT="my-project"
export GOOGLE_CLOUD_LOCATION="us-central1"
# Or a service account / ADC instead of a key
gcloud auth application-default login
# ...or: export GOOGLE_APPLICATION_CREDENTIALS=/path/to/keyfile.jsonOne failure mode is worth its own warning, because it fails silently. If you are moving to Application Default Credentials or a service account, unset the key variables first: Google's docs specifically instruct you to unset GOOGLE_API_KEY GEMINI_API_KEY, because either one wins over ADC. Export a test key while debugging and your carefully built service-account setup simply never takes effect, with no error to tell you.
What died is easy to state, because Google published it. The Google-login tier gave 1,000 model requests per user per day and 60 per minute. An unpaid API key gave 250 a day, 10 a minute, on Flash only. What survives is also published: Code Assist Standard allows 1,500 requests per user per day and Enterprise 2,000, shared with agent mode and aggregated across model families.
What replaced it is not published at all, and that absence is the finding. Antigravity gives no absolute request counts anywhere, only multipliers between paid tiers, and it meters the amount of work an agent does rather than counting requests, on a baseline quota refreshed weekly for unpaid users, with AI credits acting only as a paid overage mechanism once that baseline is gone. You cannot compare the old allowance to the new one, because Google stopped expressing the new one in a comparable unit. Treat the widely circulated "1,000 a day becomes 20 a day" line with suspicion: that 20 describes the Antigravity IDE free tier after a cut in December 2025, six months before the CLI existed, and Google has published no per-day figure for Antigravity CLI at all.
What you can point to is user reports, and one is unusually specific: an open, unanswered issue on the Antigravity CLI tracker titled "UX/Quota Regression: Pro Plan hits daily request limit after only 2 prompts in agy vs 1000+ changes in gemini-cli," quoting the error verbatim:
Individual quota reached. Contact your administrator to
enable overages. Resets in 2h40m11s.That error is itself a data point. Google documents the unpaid baseline as refreshing weekly, but a reset counted down in hours implies a per-day bucket underneath that nobody has published.
Five weeks after the cutoff, gemini-cli's readme still advertises "Free tier: 60 requests/min and 1,000 requests/day with personal Google account" as its first selling point, and still presents signing in with Google as the option "Best for: Individual developers as well as anyone who has a Gemini Code Assist License." There is no deprecation banner. Google did announce the cutoff in the repo itself, twice, in the Announcements discussions (the May 19 transition notice and the June 18 shutdown notice), but never touched the docs: the only acknowledgement in the documentation is an issue a community member filed the day before the cutoff, listing ten stale files, still open and unassigned. If you are checking any of this for yourself, read Google's deprecation page for consumer accounts rather than the repo readme, because the readme is describing a product that no longer exists for you.
The announcement did not land well. Google's May 19 announcement thread carries 295 thumbs-down reactions against 6 thumbs-up, plus 35 "confused," across 85 comments and 146 replies. The separate June 18 shutdown notice drew another 64 thumbs-down. The recurring, documented complaints are the quota opacity above, the missing non-interactive auth, the workspace-skill relocation the importer will not do for you, and trading an Apache-2.0 tool for a binary with no source. Those are the ones with receipts; the vaguer claims circulating about Antigravity being slower or burning tokens faster are harder to pin to a primary source, so weigh them accordingly.
You have three reasonable paths, and none of them forces you onto Antigravity.
The open-source gemini-cli is intact, and if you are willing to pay for a key you can carry on almost as before. But "nothing was taken from anyone" would be too generous: paying Google AI Pro and Ultra subscribers lost CLI access too, and the replacement they were pointed at cannot authenticate without a human at a browser. What Google removed is the free lane; what it is steering you toward is a binary with no source, no published quota, and no unattended auth. Treat the cutoff as a prompt to pick the agent and backend that fit your budget and your workflow, not as a forced march to the default replacement.
Published by Yaw Labs.