VocalFuse is a Fuse Intelligence product.

AGENT HARNESS · MIGRATION · 2026

Gemini CLI shutdown: why your terminal says 410 Gone, and what to do about it

If a script that ran fine for months started returning HTTP 410 Gone around June 18, 2026, nothing is broken on your machine. Google shut down Gemini CLI's consumer access on that date: free, Google AI Pro, and Google AI Ultra accounts stopped being served, the Code Assist IDE extensions went with it, and the replacement is the closed-source Antigravity CLI (agy). This page is the post-shutdown playbook: who is actually affected, the migration steps that break silently, the weekly-compute-cap reality, and the four alternatives worth benchmarking before you re-run your pipelines.

Who is affected Migration steps The 4 alternatives

Who got cut off on June 18, 2026 — and who didn't

Account type June 18 status What to do
Free tier (personal Google account) Cut off — every request returns 410 Gone Migrate to Antigravity CLI or an alternative below
Google AI Pro / Ultra subscriptions Cut off — same 410 Gone on every request Same; the quota moved inside Antigravity CLI
Gemini Code Assist for individuals + IDE extensions Cut off; GitHub app stopped accepting new org installs Existing org installs stopped serving in the weeks after
Gemini Code Assist Standard / Enterprise licenses Unaffected — access continues on the latest models Nothing required; ecosystem support and tutorials drift toward agy
Paid Gemini API key users (GOOGLE_API_KEY) Unaffected — the CLI keeps working on metered keys Keep going; budget by requests, not months

The binary itself still installs and launches — the shutdown retired the endpoint, not the code. That distinction matters if you are debugging at midnight: there is no fix on your side, no re-auth that helps, and no config change that revives the consumer quota. The request dies server-side with a 410.

The migration, step by step (and where it breaks silently)

  1. Install agy. One-line curl install; the Go binary has no Node.js dependency and shares its agent harness with the Antigravity 2.0 desktop app. First launch runs an onboarding conversion that detects your legacy Gemini CLI profiles and offers an import checklist.
  2. Auth. Browser-based Google OAuth replaces manual key management; tokens land in your OS keyring instead of a flat config file. For headless use the environment variable moves from GEMINI_API_KEY to AV_API_KEY.
  3. Import extensions. agy plugin import gemini converts Gemini CLI extensions into native plugins; manage them afterward with agy plugin list / enable / disable / uninstall.
  4. Relocate skills. Workspace skills must move from .gemini/skills/ to .agents/skills/ — the agent does not recognize them as active slash commands until you rename the folder. Global skills move to ~/.gemini/antigravity-cli/skills/.
  5. Rebuild MCP config. Server definitions move out of settings.json into a dedicated mcp_config.json (global: ~/.gemini/config/mcp_config.json, workspace: .agents/mcp_config.json). Remote servers must rename url to serverUrl — keep url and the server loads at startup, passes initial checks, then fails silently hours later when a tool is invoked. This is the single most-reported silent breakage; run agy doctor and agy inspect to verify before you trust it.
  6. Audit everything the import doesn't touch. CI configs, cron jobs, shell aliases, and scripts that call gemini directly are not migrated by any tool — find-and-replace to agy yourself. Exit codes and SSE-based streaming defaults changed too; scripts checking specific Gemini CLI exit statuses can misbehave without errors.
  7. Validate on a real task, then retire gemini. Run an actual ticket through agy before removing the old binary. Both can coexist temporarily — which is trivially true now, since only one of them can reach Google.

One honest caveat: Google states there is not 1:1 feature parity at launch. The ACP (Agent Client Protocol) mode has no drop-in equivalent — if your orchestrator wraps Gemini CLI over stdio, you are rewriting that bridge. Custom terminal themes and experimental overlays may not carry over. Workspace rules (GEMINI.md, AGENTS.md) parse identically, so those move free.

Command and config mapping: gemini → agy

You want to Gemini CLI Antigravity CLI
Switch models in the TUI /model /model — now also runs Claude and GPT-OSS models
Plan mode /plan /plan (v1.1.0 renamed /planning → /plan; /fast removed)
Cycle approval modes Shift+Tab Shift+Tab (default → accept-edits → plan, GA in v1.1.0)
Auto-approve everything --yolo (-y) --dangerously-skip-permissions, or /permissions → always-proceed
Sandboxed execution --sandbox Native OS sandboxing: nsjail (Linux), sandbox-exec (macOS), AppContainer (Windows)
Check remaining quota n/a — daily request count /credits in the TUI — weekly compute, not daily requests

The weekly compute cap: what the free tier became

Gemini CLI's famous 1,000 requests/day is gone; Antigravity CLI allocates a weekly compute cap instead, and Google has not published the numbers. Community reports describe throttling after roughly 2,000 lines of generated code with multi-day cooldowns, and shared token pools that burn an order of magnitude faster than the old daily quota — "what previously was 5-8 hours of work now burned in 10-20 minutes." For interactive use the free tier is genuinely generous; for heavy generation it is a planning constraint, and that is the practical difference that pushes serious usage to one of the four alternatives below.

Four alternatives worth benchmarking before you re-run your pipelines

Tool License Wins on Cost
Claude Code Closed source Big multi-file refactors; best-in-class agentic search across large repos $20/mo Pro entry; Max removes usage anxiety
Codex CLI CLI open source Interactive single-file pair coding; lightest cold start; ChatGPT-native $20/mo Plus entry
Aider Apache 2.0 Git-disciplined incremental edits; commits every change; refuses to advance without a passing build BYOM — any API key, including budget models
OpenCode / Crush Apache 2.0 Open-source Claude Code equivalent; BYOM; plugin model, no vendor lock-in BYOM — any API key
VibeFuse Free harness The first ever free widget-based AI harness: swappable agents on a desktop canvas, local/offline processing, works in any app Free — plus an open-source marketplace where creators sell widgets, skills, and styling packs with revenue share

The decision rule that held up through the shutdown: if you bill clients for code and "open source for audit" is a contract requirement, Aider or OpenCode are the only Apache 2.0 binaries you can ship in a deliverable. If your day is big-repo refactoring, Claude Code still finishes the loop best. If you live in ChatGPT, Codex is the path of least resistance. And if what you actually valued about Gemini CLI was owning your tooling — scripts, MCP configs, GEMINI.md, skills, forks — that ownership layer is exactly what a harness like VibeFuse is built around: agents are perishable, harnesses persist.

CI/CD and automation: the checklist nobody runs

  • Grep your repos for the binary. gemini calls hide in CI configs, cron jobs, shell aliases, Makefiles, and pre-commit hooks. The migration command does not touch any of them.
  • Re-check exit codes. Antigravity CLI defaults to SSE streaming and introduces new non-zero exit codes; CI steps that matched specific Gemini CLI exit statuses pass silently or fail weirdly.
  • Re-test MCP tool calls, not startup. The url vs serverUrl trap passes startup checks and fails at invocation time — hours later. Run a real tool call in staging before you trust the migration.
  • Budget by requests, not months. Whatever you land on — agy's weekly cap, a metered API key, or a $20/mo subscription — the free ride is gone. Measure a week of real usage before committing a team.

If you got here because a pipeline is down right now

Shortest path back to green: confirm the failure is a 410 (not a proxy or DNS problem), point the pipeline at a paid Gemini API key as a stopgap if you have one, and schedule the real migration for this week — either agy with the steps above, or one of the four alternatives benchmarked against your actual workload. The one thing you cannot migrate is the quota you were relying on; budget the replacement by requests until you know your real usage.

Related reading

Gemini CLI after the cutoff

What still works: API-key routes, the Apache 2.0 repo, enterprise paths — and the durable lesson.

Gemini CLI, discontinued

Cursor under SpaceX

The other 2026 consolidation shock — what the acquisition changed for daily users.

Cursor + SpaceX

Terminal agents compared

Claude Code vs Cursor vs Codex after the June 2026 shutdown reshuffled the field.

Claude Code vs Cursor

Free and open routes

What "free" means for AI coding agents in 2026 — and the open-source replacements.

Free Claude Code alternative

Explore related AI note taking guides

Gemini CLI shutdown — FAQ

Is Gemini CLI discontinued or deprecated?

Consumer access is, the code is not. Google announced the transition on May 19, 2026, and on June 18, 2026 Gemini CLI stopped serving requests for free, Google AI Pro, and Google AI Ultra accounts — every request returns HTTP 410 Gone. Enterprise access via Gemini Code Assist Standard/Enterprise licenses is unchanged, the CLI keeps working with paid Gemini API keys, and the Apache 2.0 repository remains open. The successor for individual users is the closed-source Antigravity CLI (agy).

Why does Gemini CLI return HTTP 410 Gone?

Because the hosted consumer endpoint was retired, not your install. The gemini binary still exists on disk and still launches — it simply cannot reach Google's servers for free, Pro, or Ultra accounts anymore, so every request dies server-side with 410 Gone. There is no client-side fix, no re-auth that helps, and no config change that revives the old quota. Paid API-key routes and enterprise licenses keep working.

Who was affected by the June 18, 2026 shutdown?

Free-tier users, Google AI Pro and Ultra subscribers, Gemini Code Assist for individuals and its IDE extensions, and new Gemini Code Assist for GitHub org installs (existing installs stopped serving in the following weeks). Explicitly not affected: Gemini Code Assist Standard/Enterprise license holders, GitHub-integration users through Google Cloud, and anyone on a paid Gemini API key.

How do I migrate from Gemini CLI to Antigravity CLI?

Install agy (one-line curl, Go binary, no Node dependency), complete the browser OAuth flow, then run agy plugin import gemini for extensions. Move workspace skills from .gemini/skills/ to .agents/skills/, and rebuild MCP config as a dedicated mcp_config.json — renaming url to serverUrl for remote servers, the silent trap that passes startup checks and fails at tool-call time. Verify with agy doctor and agy inspect, then find-and-replace gemini in CI configs, cron jobs, and aliases yourself — no migration tool touches those.

What happened to the free 1,000 requests per day?

It did not move — it shrank. Antigravity CLI allocates a weekly compute cap instead of a daily request count, Google has not published the numbers, and community reports describe throttling after roughly 2,000 lines of generated code with multi-day cooldowns. The old 1,000/day tier was real from June 2025 until the cutoff; articles still advertising it predate June 2026.

What is the best alternative after the Gemini CLI shutdown?

Depends on the workload. Claude Code wins big multi-file refactors ($20/mo Pro entry). Codex CLI is the lightest interactive pair-coder and the natural pick if you live in ChatGPT ($20/mo Plus entry). Aider and OpenCode are the Apache 2.0 options you can ship in client deliverables — the "open source for audit" contract requirement — both BYOM. VibeFuse is the free-harness angle: the FIRST EVER free widget-based AI harness with swappable agents on a desktop canvas, local/offline processing on Windows, and an open-source marketplace paying revenue share for widgets, skills, and styling packs. Agents are perishable; harnesses persist.

Do enterprise Gemini Code Assist users need to migrate?

Not on any deadline. Standard and Enterprise licenses, plus Google Cloud GitHub-integration access, were explicitly exempted from the June 18 cutoff — access continues with the latest Gemini models. The practical pressure is ecosystem drift: community extensions, tutorials, and tooling increasingly target agy, so plan an evaluation even though nothing breaks.

Can I keep using the old gemini binary at all?

Yes, on two routes: with a paid Gemini API key (set GOOGLE_API_KEY and the open-source binary keeps running — Google confirmed the repo stays up), or as a from-source build pointed at any Gemini-capable endpoint. What you cannot get back is the consumer quota. If your pipeline broke on June 18, a paid key is the shortest stopgap while you plan the real migration.