Google's Gemini is the second free Janitor AI proxy that actually holds up, after OpenRouter, and it works for one reason that is easy to miss: Google ships an OpenAI-compatible endpoint for the Gemini API. Janitor AI's proxy field speaks the OpenAI format and nothing else, so without that endpoint Gemini would not connect at all. With it, a free key from Google AI Studio and one URL put Gemini's Flash models behind your characters.
As with every page in this series, the word proxy here means a language-model endpoint, not a network proxy that changes your IP address. We run the network kind, which is exactly why we can tell you that it does nothing for a Gemini rate limit, and exactly where it does matter, at the end.
The values to paste
Open a chat, tap the menu, choose API Settings, then Proxy, then Add Configuration.
| Field | What goes in | Notes |
|---|---|---|
| Config Name | Gemini free | a label |
| Proxy URL | https://generativelanguage.googleapis.com/v1beta/openai/chat/completions | Google's OpenAI-compatible base URL plus /chat/completions |
| API Key | your key from Google AI Studio | created at aistudio.google.com; Google reads it as a Bearer token |
| Model Name | gemini-3.7-flash | or gemini-3.5-flash, gemini-3.5-flash-lite; exact spelling |
| Custom Prompt | optional | leave empty at first |
Save and send a short test message. The most common failure at this step is the URL: the base URL Google documents ends in /openai/, and Janitor needs the full path with /chat/completions after it.
Which models are free, in Google's words
Google's pricing page describes the free tier as limited access to certain models with free input and output tokens, and it lists these Gemini 3 models as free of charge for input, output and context caching: Gemini 3.7 Flash, 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite and 3 Flash Preview. On the models page, gemini-3.7-flash is the current Flash model, gemini-3.6-flash the previous generation, gemini-3.5-flash and gemini-3.5-flash-lite the lighter pair, and gemini-2.5-flash is still served. The page also marks what has gone: gemini-2.0-flash and gemini-2.0-flash-lite are shut down, so any older Janitor guide telling you to type those is describing a model that no longer answers.
For roleplay the practical choice is between 3.7 Flash, the most capable of the free set, and the Flash-Lite models, which are faster and cheaper on the free tier's limits. Try 3.7 Flash first; drop to Lite if you run into the per-minute limit during a fast back-and-forth.
The limits are per project, and Google shows them in AI Studio
Google's rate-limit documentation no longer prints a table of free-tier numbers. It says limits depend on your usage tier and can be viewed in Google AI Studio, and it defines the tiers: Free, for an active project or free trial; Tier 1, after linking an active billing account; Tier 2, after $100 paid plus three days; Tier 3, after $1,000 plus thirty days. One sentence on that page matters more than any number: rate limits are applied per project, not per API key.
That settles the two shortcuts people try. A second API key in the same project shares the same limit. A different IP address is not an input at all; the limit follows the project the key belongs to. When a 429 arrives, the choices are to wait, to chat more slowly, to switch to a Lite model, or to link billing and move the project up a tier. Our page on Janitor AI error 429 walks through the same decision for every backend.
Moves the limit
Linking billing to the project
Tier 1 and up; also ends the free-tier data use
A lighter model
Flash-Lite variants for fast exchanges
Slowing the conversation
limits are per minute and per day
Does nothing
A second API key
limits are per project, not per key
A different IP address
not an input to the limit
A shared reverse proxy
someone else's project, someone else's logs
The privacy trade, stated plainly
Google's pricing page puts the free tier's condition next to its price: content used to improve our products. The paid tiers carry the opposite note. For a Janitor AI user that means the roleplay you send through a free Gemini key may be used by Google to improve its products, under the Gemini API terms. Linking billing to the project, even without spending much, moves you to the paid tier where that use stops.
We are not telling you which to choose. We are telling you that the choice exists, that it is written on Google's own page, and that it is the real difference between the free Gemini route and OpenRouter's paid models or DeepSeek's cheap API, where the terms are different again. Pick on purpose.
Age and region, the two walls no URL moves
Google's available regions page lists the countries and territories where the Gemini API and AI Studio are offered, more than 190 of them, and names the three reasons access can be refused: you are outside those regions, you are under 18, or your Google account needs to verify your age. The Gemini app's own age rules are separate and stricter in some places, as our proxies for Gemini post covers, but for the API key that Janitor uses, those three are the walls.
Two of them are about the account and no proxy of any kind moves them. The third, region, is the one place where a network proxy is relevant, and it is the subject of the last section.
The errors you will meet
- 401 Unauthorized. The key is wrong, revoked, or pasted with a space. Make a new one in AI Studio and paste it clean.
- 429 Too Many Requests. The project's rate limit. Wait, slow down, switch to a Lite model, or link billing. Details on the 429 page.
- 400 or a refusal. Usually too much text in one request (bot definition plus memory plus prompt), or Gemini's safety filters declining the content. Shorten first; if it is the filter, another backend may suit the scene better.
- Network error. The URL is missing
/chat/completions, or has a typo. Compare it character by character with the table above. - A region or age message. Not a proxy problem in the Janitor sense; see the section above.
The shared-proxy trap
The Discord "free Gemini proxy" with a URL and password is somebody else's project and somebody else's key. Everything you send passes through their server readable, and their key can be revoked mid-scene. Your own AI Studio key takes two minutes and costs nothing, which leaves no reason to borrow one. The longer warning is in best free Janitor AI proxy, and it applies here in full.
If you actually needed a network proxy
If your country is not on Google's list, or your school or office network filters generativelanguage.googleapis.com or janitorai.com, the request never reaches Gemini and no key or model fixes it. That is the one case for the proxy we sell: a static residential or ISP address in a country where both Google's API and Janitor AI are offered, kept for the length of the session; ours start at $0.44 per GB, billed on use, no KYC. Reaching a service from a region it does not serve is a decision to read against Google's terms, so read them before relying on it.
For the rate limit, the privacy setting and the age rule, the network proxy changes nothing, and neither does anything else on this page except the choices Google itself offers: a lighter model, a slower chat, or a billed project. Paste the URL exactly, use a model from the free list, and you are chatting for free.
Sources
- Google, OpenAI compatibility for the Gemini API: the base URL https://generativelanguage.googleapis.com/v1beta/openai/, the model field, and the Bearer key.
- Google, Gemini API pricing: the free tier's free-of-charge models and the note that free-tier content is used to improve Google's products.
- Google, Gemini models: current, preview and shut-down model ids.
- Google, Rate limits: tiers, limits viewable in AI Studio, and the rule that limits apply per project rather than per API key.
- Google, Available regions: the country list and the three reasons access can be refused.
- Google, Gemini API terms.
- Janitor AI Help Center, The Absolute Beginner's Guide to Using a Proxy with Janitor, Part One: the menu path and field names.