May 19, 2026: Antigravity Pro users report Gemini 3.1 Pro quota stuck on a 5-day refresh. No public Google release since April 16. There's another way.

Bodega One vs. Google Antigravity

Local-first. One-time. Your code stays yours.

Antigravity is cloud-only, your prompts train Gemini by default, and the quota math keeps changing on people who already paid. Bodega One runs on your machine, charges once, and never sees your code.

Join the Waitlist See Pricing

At a glance.

The architectural defaults that drive every other difference.

Feature	Google Antigravity	Bodega One
Pricing model	Free (capped) / Google AI Pro / Ultra ~$250/mo + credit packs	$79 one-time
Quotas	Weekly + multi-day refresh, changed twice in 2026	None (you bring the LLM)
Data training default	On for personal Gmail / Pro / Ultra accounts	Off (no upstream recipient)
Model location	Google-hosted cloud only	Local or BYOK cloud, your choice
Account required	Google account, mandatory	None for offline use
Supported models	Gemini 3 Pro, Gemini 3.1 Pro, Claude Sonnet 4.5, GPT-OSS	10+ provider presets (BYOLLM)
License to your code	Used for training unless Workspace / GCP	Never transmitted
Platforms	Windows / macOS / Linux	Windows / macOS / Linux

Your code is Google's training data by default.

The Antigravity Terms of Service say it plainly: “When you use the Service, we record and store your user data, interaction data pertaining to your usage of the Service.” There is one exception, called out in the same policy: “If you are accessing the Service via Google Workspace or the Google Cloud Platform, we will not collect your prompts, content, or model responses.” Source: Google AI Developers Forum.

Read that twice. The exception is for the people Google is invoicing through enterprise contracts. Everyone else, free, Pro, and Ultra users on personal Gmail accounts, has their prompts and source code sent to Google and used for training unless they navigate to settings and opt out. Multiple forum posts report that the opt-out UI is missing or unclear in shipping versions, including a thread titled “Urgent: data deletion / opt-out request ignored, missing privacy settings.”

If you are a contractor, an indie shipping client work, a founder with unreleased IP, or anyone with an NDA in the room, the default posture is “your customer's code is now part of a Google training set unless you do extra work.” That is a different conversation than “the LLM is helpful.”

Bodega One does not have this conversation. With a local model, there is no upstream recipient. Your code is processed on your GPU, full stop. With air-gap mode on, nine independent enforcement layers verify zero bytes leave the machine. How air-gap mode works →

The quota keeps moving.

Antigravity does not have its own subscription. Paid usage runs through Google AI Pro and Google AI Ultra, plus pay-as-you-go AI Credits. The meters on those tiers have moved twice in 2026, and not in the customer's direction.

Launch (Nov 2025)
Antigravity ships with free tier "at no cost for individuals." Ultra advertises "No weekly rate limit."
Dec 2025
Google moves free tier from per-day caps to a "larger, weekly based rate limit." Pro and Ultra get quotas that "refresh every five hours."
Mar 10, 2026
Google staff publicly acknowledges Pro quota issues in the dev forum. No fix shipped in-thread.
Mar 26, 2026
Ultra customers post: "$250/month Ultra + 40K Avail Credits = You need Credits. What am I paying for?" Google staff acknowledges. No price change.
Mar 28, 2026
Ultra copy is silently rewritten from "No weekly rate limit" to "Highest weekly rate limits."
May 19, 2026
Pro plan bug reports confirm Gemini 3.1 Pro quota stuck on a 5-day refresh instead of the advertised 5 hours.

Sources, in order of citation: Google's own announcement of the Pro / Ultra rate-limit policy, the Google-employee-acknowledged Pro quota issues thread, the Ultra customer escalation thread “$250/month Ultra + 40K Avail Credits”, the Ultra copy-change thread, and the live Antigravity bug-report board (5-day Pro refresh confirmed 2026-05-19).

On top of subscriptions, the in-product purchase flow surfaces AI Credit packs at 2,500 credits for $24.99, 5,000 for $49.98, and 20,000 for $199. Google has not published a credit-to-token conversion rate, so customers on the receiving end of the credit meter cannot reliably forecast what a refactor session will cost. Bodega One does not have a credit meter, because we are not the ones running inference. You bring the LLM. The price is $79 one time.

Shipping cadence has slowed.

From early March through mid-April 2026, Antigravity shipped a steady drumbeat: v1.20.3 (Mar 5), v1.20.5 (Mar 9), v1.20.6 (Mar 17), v1.22.2 (around Apr 8), v1.23.x (by Apr 19), v1.23.2 (Apr 28). Then it stopped.

A May 9, 2026 thread on the Google AI Developers Forum, “No Updates Since April 16: Is Google Antigravity Slowing Down or Preparing for a Major Release?” drew 13 community replies and zero from Google. A May 17 follow-up asked whether the product is being abandoned, also without Google reply.

The regressions are not cosmetic. v1.22.2 introduced a new Agent permissions UI that fails to persist settings and can SIGABRT-crash and corrupt the database. v1.23 removed terminal interactivity in the AI Agent surface, tracked in “Regression: AI Agent terminal no longer accessible (v1.23).” The forum on May 19, 2026 is wall-to-wall bug reports: 5-day Pro quota refresh, “submitting” hang with Claude Opus, missing terminal, account-specific 503 MODEL_CAPACITY_EXHAUSTED, in-app bug reporter broken.

For contrast: Joe published Bodega One beta-19 on May 13, 2026. Beta cadence is roughly weekly. The product roadmap is set by one builder shipping fast, not by an internal-priorities reshuffle at a $2 trillion company. We are small. That is the point.

Cloud-only vs. local-first.

Antigravity's launch post is explicit that the product “combines a familiar, AI-powered coding experience with a new agent-first interface that allows developers to deploy agents that autonomously plan, execute, and verify complex tasks across their editor, terminal, and browser.” Source: developers.googleblog.com. The Antigravity download is a thin client. All model inference is server-side, on Google-hosted endpoints, gated by a Google account login. There is no local-model option and no way to run offline.

That is not an oversight. Google's incentive is to keep Gemini inference inside Google. Antigravity is a distribution surface for Google AI subscriptions. The architecture follows the business model.

Bodega One's incentive is to ship something you pay for once and never have to ask permission to use again. The architecture follows that business model in the other direction. You buy a license, you install on Windows, macOS, or Linux, you point at a local model via Ollama, LM Studio, vLLM, llama.cpp, or any of the other 10+ supported providers, and your prompts never touch our servers because we do not have any in the request path. With air-gap mode on, nine independent enforcement layers verify the absence of outbound traffic. Disable one and the other eight still hold.

That is the test Antigravity cannot pass. Not because Google shipped it badly. Because the business model requires the opposite.

When Antigravity might still be right.

We do not think every developer should switch. A few real cases where Antigravity is the better tool today:

You live inside Google Cloud. Antigravity ships first-party MCP integrations for AlloyDB, BigQuery, Cloud SQL, Spanner, and Looker that landed December 15, 2025. If 90% of your day is querying GCP data services, the integration depth and IAM-credential handling matter more than local-first.
Your org is already on Google Workspace or GCP enterprise. The same ToS that opts personal accounts into training explicitly exempts Workspace and GCP customers from prompt and response collection. If your billing already runs through Workspace, you get the carveout for free.
You want managed Gemini 3.1 Pro with zero setup and you do not mind the quota math. Antigravity hands you a frontier model in a polished agent UI. Cloud-managed inference removes the hardware question. If you would rather pay Google a subscription than configure Ollama, Antigravity is faster to first commit.
You specifically want the Manager Surface. The asynchronous multi-agent Kanban metaphor is genuinely well-done. We have a multi-agent system, but we do not currently match the Manager UI one-for-one.

If you read those four bullets and none of them describe you, keep reading.

When to switch to Bodega One.

The opposite list:

You work on client code, customer code, or anything under NDA, and you do not want to read a 4,000-word ToS to find out where the opt-out is.
You have been burned by a quota change on a tool you already paid for, and you want a meter that cannot move.
You want to run a real local model on your own hardware, not a Gemini wrapper. Bodega One ships with a hardware recommender and Ollama copy-paste commands for 25+ models.
You are in a regulated industry where the air-gap test is not optional. Defense, healthcare, finance, government.
You want the four things we will not change: Own It. Private. Flexible. Powerful.

Pricing comparison.

Two business models. The shape of the table tells the story.

Google Antigravity

Free: weekly rate-limited (numerical cap not Google-published).
Google AI Pro: monthly subscription on Google One billing. Quotas now on multi-day refresh windows for Gemini 3.1 Pro.
Google AI Ultra: ~$250/mo (per customer-quoted Ultra forum thread). Was “no weekly cap.” Now “highest weekly cap.” Still requires AI Credit top-ups for heavy users.
AI Credits: 2,500 / $24.99, 5,000 / $49.98, 20,000 / $199. Credit-to-token conversion not published.
Cancellation: access stops when subscription stops. No local artifact.

Bodega One

Personal: $79 one-time. 2 machines.
Pro: $149 one-time. 5 machines.
Enterprise: Custom. 10+ machines, volume terms.
Inference cost: whatever you choose to pay your LLM provider, or $0 with a local model.
Cancellation: not a concept. You own the license.

See full pricing →

Common
questions.

Does Google Antigravity really train Gemini on my source code?+
By default, yes, if you sign in with a personal Google account. The Antigravity Terms of Service state that user data and interaction data are recorded and stored. The only exception called out in the same policy is access via Google Workspace or Google Cloud Platform, where prompts, content, and model responses are not collected. Free, Pro, and Ultra users on personal Gmail accounts are inside the training pool unless they hunt down the opt-out, and forum users report the opt-out UI is missing or unclear in shipping versions. With Bodega One running a local model, there is no upstream recipient at all.
What changed with the Antigravity Pro and Ultra quotas?+
Two things, neither of them small. Pro tier was originally advertised with a 5-hour refresh window. Forum bug reports on May 19, 2026 confirm Pro is now stuck on a 5-day refresh, which is roughly 24x longer between resets. Ultra was advertised at launch as "No weekly rate limit." By March 28, 2026 the copy was rewritten to "Highest weekly rate limits." Same plan, different ceiling. A Google employee acknowledged the Pro quota issue on March 10, 2026 and another acknowledged the Ultra credits issue on March 26, 2026, but the underlying pricing model has not been reverted.
Is Antigravity still being actively developed?+
The community is asking the same question. A May 9, 2026 forum thread titled "No Updates Since April 16: Is Google Antigravity Slowing Down or Preparing for a Major Release?" drew 13 community replies and zero Google staff responses. A follow-up on May 17, 2026 asks bluntly whether the product is being abandoned, also with no Google reply. The last confirmed release we could pin to a primary source is v1.23.2 from late April 2026. Meanwhile Bodega One shipped beta-19 on May 13, 2026.
Can Bodega One use Gemini 3.1 Pro?+
Yes, through any Gemini-compatible endpoint you bring. Bodega One is BYOLLM, so if you have access to Gemini via Google's API, an OpenAI-compatible gateway, or OpenRouter, you can plug it into Bodega One. The difference is you're paying Google directly for inference instead of paying Google a subscription that then meters access to the same model. And you can swap to Claude, GPT, Qwen, DeepSeek, or anything else without changing tools.
I use Antigravity to talk to BigQuery and AlloyDB. Can Bodega One do that?+
Bodega One supports Model Context Protocol (MCP), the same protocol Antigravity uses for its Google Cloud Data Services integration. If a Google Cloud MCP server exists for the surface you care about, it works in Bodega One too. The integration depth is different. Antigravity has tighter first-party UX for the Google Data Cloud surfaces. If 90% of your day is BigQuery and AlloyDB inside Google Cloud, that integration depth may matter more than local-first.
Does Bodega One have an equivalent to the Manager Surface?+
Bodega One has a multi-agent system where multiple agents can work on a task in parallel under the QEL gate. The Manager Surface metaphor of spawning many async agents from a Kanban-style board is something we are tracking but have not shipped a direct visual equivalent of. If a Kanban agent board is the core thing you need today, Antigravity is further along there.
When is Bodega One available?+
Beta is live now for the first 200 users. Full launch coming later this year. Join the waitlist at bodegaone.ai to be first in line.

Related reading: Cursor alternatives / GitHub Copilot alternatives.

A meter that cannot move.

One-time purchase. Runs local. Air-gap mode built in. 23 tools and an autonomous agent that verifies its own work via the Quality Enforcement Layer. Your code stays on your machine, not Google's training set, not anyone else's.

Join the Waitlist

$79 Personal. $149 Pro. One-time. Windows, macOS, Linux.