Gemini CLI Shuts Down June 18: The Coding-Agent CLI Routing Decision Your Team Has 3 Days to Make

The deadline is in three days. On June 18, 2026, Google will stop serving requests from Gemini CLI and Gemini Code Assist IDE extensions for all consumer-tier accounts — free, Google AI Pro, and Google AI Ultra. The official successor is Antigravity CLI, now available to download, and the migration path is clear enough on the surface. But for engineering teams who embedded Gemini CLI into their coding-agent workflows, the real question is not just "how do I migrate" but "should I migrate to Antigravity at all, or is this the forcing function to move to a provider-independent routing architecture?"

What happened

On May 19, 2026, Google announced the Gemini CLI consumer shutdown as part of Google I/O 2026. The decision is architectural: Gemini CLI proved the terminal as an agentic interface, but Google concluded that multi-agent workflows need a unified backend — and Gemini CLI's standalone model couldn't deliver that. Antigravity CLI shares its agent harness with Antigravity 2.0 (the desktop app), is built in Go for lower latency, and supports asynchronous agent orchestration: you can run large-scale refactors across multiple agents in the background without blocking your terminal session.

The scope of the shutdown affects:

• Free tier users — Gemini CLI goes dark June 18
• Google AI Pro and Ultra subscribers — same date, same cutoff
• Gemini Code Assist for individuals — IDE extensions also stop on June 18
• Gemini Code Assist on GitHub (consumer) — no new installations from June 18; existing requests stop being served July 17

Who is NOT affected: Gemini Code Assist Standard or Enterprise license holders. If your organization pays for enterprise access, Gemini CLI and IDE extensions continue working with full model access. Gemini CLI will also remain accessible via paid Gemini Enterprise Agent Platform API keys.

Why it matters for AI engineering teams

Three engineering realities collide on June 18.

First, the urgency is real. Teams that adopted Gemini CLI for scripted coding tasks, CI pipeline assistants, or ad hoc terminal agents need a decision this week — not next sprint. Workloads that silently hit Gemini CLI endpoints will begin returning errors Wednesday.

Second, Antigravity CLI is opinionated. It is architecturally designed to route everything through Google's Antigravity agent harness. That is its strength for Google-native workloads — seamless future upgrades, shared state with desktop agents, Google Cloud project integration. But it is also a deliberate centralization. There is no provider fallback built in, no multi-model routing, and no OpenAI-compatible API surface. If your team builds on Antigravity CLI, you are building on Google's inference infrastructure.

Third, the competitive field is stronger than it was a year ago. When Gemini CLI shipped in mid-2025, Claude Code was early-access and Codex CLI was experimental. Today, Claude Code has nested subagents, enterprise-grade session persistence, and a managed-settings governance layer. Codex CLI has persistent cloud execution. The migration event gives teams a legitimate moment to ask: does Google's unified-backend model serve us, or does provider-independent routing?

The router/operator angle

For teams running routing-aware agent architectures, this shutdown surfaces a genuine architectural fork.

Path A — Antigravity CLI. The smoothest migration for teams already invested in the Google ecosystem. Antigravity CLI preserves Agent Skills, Hooks, Subagents, and Extensions (now renamed Antigravity plugins). For teams using Google Cloud credentials and Vertex AI integrations, the single-backend model reduces operational surface area. The trade-off: Antigravity is not OpenAI-compatible. You cannot route it through a standard /v1/chat/completions gateway. Billing, observability, and fallback policy live inside Google's infrastructure, not your own.

Path B — Multi-provider routing via Claude Code or Codex CLI. Both are OpenAI-SDK-compatible at their API layer, which means you can front them through a routing gateway, attach your own observability, and define fallback chains (e.g., route to Claude Code on Fable 5, fallback to Codex on gpt-5 on latency spike). This path requires more configuration upfront but gives you model-agnostic cost controls and provider independence. Claude Code in particular now supports availableModels governance controls and fallback model configuration at the enterprise level.

The key operator question: Do your coding-agent workflows need to be observable and fallback-aware at the routing layer, or are they isolated terminal sessions that can tolerate single-provider lock-in? If the answer is the former — especially for CI-integrated agents, billing-reconciled team workflows, or regulated-environment coding assistants — the June 18 event is an opportunity to audit rather than just migrate.

What to check now:

• Identify all workflows, scripts, or CI jobs invoking gemini CLI binary or Gemini Code Assist IDE plugins under consumer-tier credentials
• For enterprise users: confirm your org's Code Assist Standard/Enterprise license is active before June 18 — no action needed if licensed
• For consumer users migrating to Antigravity: test your Agent Skills and plugin configurations; there is no 1:1 feature parity at launch
• For teams evaluating Claude Code or Codex CLI: verify OpenAI-compatible SDK configuration and check whether your existing routing gateway supports CLI-invoked agent sessions

What TheRouter users should watch or try

The Gemini CLI consumer shutdown is a useful audit trigger even if you do not use Gemini CLI directly. Check whether any of your routing policies reference Gemini provider models that were accessed via consumer-tier credentials — those paths will fail after June 18. For enterprise-licensed Gemini routes, no immediate action is needed, but this is a good moment to confirm your fallback model chain is correctly ordered if Gemini capacity is constrained during the Antigravity migration wave.

If you are evaluating moving coding-agent CLI workloads from Gemini to Claude Code routed through TheRouter, the /v1/chat/completions passthrough and model-specific routing rules let you assign specific agent sessions to specific providers without changing the calling CLI's configuration. See the provider routing documentation for setup details and the model catalog for currently supported Gemini and Anthropic model endpoints.

Source: Google Developers Blog — An important update: Transitioning Gemini CLI to Antigravity CLI (May 19, 2026) and Google Gemini Code Assist deprecations page (updated June 11, 2026).