QUICK GUIDE TO GEMINI PROXY:

All of this is FREE and takes 2 minutes !!! ദ്ദി(ᴖᗜᴖ)

~

I'M NOT A GEMINI EXPERT OR TECH SUPPORT.
I'M JUST AN AVERAGE USER LIKE THE REST OF YOU.

NEWS:

4/8/26
I CAN NO LONGER UPDATE GUIDE.
THIS WILL PROBABLY BE LAST UPDATE DUE TO MODERATION

⚠️

JANITOR DEVS PUSHED AN UPDATE ON PROXIES. AND MORE OR LESS WIPED OUT THE PROXY SETTINGS EVERYONE HAS.
4/ 8/ 26

Edits, updates, and changes to the proxy with this new update may also wipes it out of existence.

~
!!️ Render fixed the server problems and GeminiForJanitor is working again

For anyone using GeminiForJanitor URLs the amazing Undefined created this to help people find or swap out URLs in case bandwidth gets too high. Use the URLs with the least amount of bandwidth first so the other URLs don't go under 'suspension'.
Also if you're getting hit with too many quota exceeded errors; try a different one with less bandwidth.

https://gfjproxies.vu5eruz.workers.dev/

👆👆👆

4 / 7 / 26

Undefined just pushed a new update to ALL GeminiForJanitor urls to allow for advanced settings now.
(new and untested so comment if you run into problems)

//help advsettings (for help and more info)
//advset_temperature on | off | this
//advset_frequency_penalty on | off | this
//advset_repetition_penalty on | off | this
//advset_top_k on | off | this
//advset_top_p on | off | this

`'on' = for entire chat
'off' = turns it off completely
'this' = only for this one message.

3 / 19 / 26

MODEL: gemini-2.5-pro WILL BE DISCONTINUED ON JUNE 17, 2026

Link: https://ai.google.dev/gemini-api/docs/deprecations

~~~

Me: Please don't message google over your problems. Please don't be like these people messaging the literal Official Gemini API Support team for help. All you're doing is putting Janitor on their radar for them to add harsher/ stricter filters

LINKS:

1. https://discuss.ai.google.dev/t/gemini-2-5-pro-bc-this-is-getting-too-far/99528

2. https://discuss.ai.google.dev/t/url-api-gogle-vertex-ai/111285

3. https://discuss.ai.google.dev/t/system-instruction-deprecation/109449/5

🚩 KNOWN GOOGLE BUG: DAILY QUOTA NOT RESETTING

Many of you had brought to my attention that the Daily Quota isn't resetting for you.

When i investigated this previously i checked my own rates and it reset so I assumed you either: Didn't wait the 24hrs or Forgot to hit 'refresh' while on aistudio to see the most up to date stats.

So to all of you who tried to comment about this I'm sorry for not looking into this problem more.

After doing a deep dive apparently this has been a problem for MONTHS!! And Google Support was clearly in the responses but never actually helped/solved this for anyone.

Now as everyone knows it's taboo to suggest to make a new one. But if it's the only way to use Gemini AND there's literally no solution or workaround for this bug, i say 'do it'.

Google = Trillion dollar company btw

👆

Messages from EgoAlter

Deepseek (free) is dead.
People will probably transfer over to Gemini and GLM so expect more 429s and overloaded models. Don't be mean to newcomers everyone is welcome to use this FREE proxy!!

⚠️

Eslezer URL: https://gemiproxy-td8i.onrender.com/ is SUSPENDED until further notice. Or, until Eslez unsuspends or makes a new one.
Use the Eslez Collab listed in my URL list instead.

GeminiForJanitor URL NEW FEATURES:

1) Thinking is now possible! Use //think on and //think_text keep to see models thinking process. To hide models thoughts use //think_text remove. To remove thinking altogether use //think off. To use only for ONE message use //think this

2) gem 2.5-flash and other free models can now use //search on!! You can now use Google Search to increase story experience!! To turn OFF use //search off
(Not possible to use on gem 3 pro or flash due to higher filter and restrictions)

Ex of what to write (to activate): //think on //think_text keep //search on

Ex of what to write (to turn off): //think off //think_text remove //search off

~

>>> Gemini Prompt v6b Experimental is out now 2/1/26 <<<

> Had a silly idea to make my Custom Prompt AI Paige into a bot

GEMINI PROJECT ERRORS SHOWING 'UNAVAILABLE' WILL CAUSE AN IMMEDIATE ERROR MESSAGE!!! THE ONLY SOLUTION IS TO WAIT FOR GOOGLE TO FIX THE PROBLEM THEMSELVES (I had to wait 1-2 weeks before being able to use gemini again)
→ Check on aistudio what tier your projects/keys are. It should show 'Free Tier' if it shows 'Unavailable' it will give an immediate 429 error.

☺

Janitor Brought back the <think> </think> tab feature!! So now it bunches up all the thinking/reasoning process into a small condensed tab within chats!!

GEMINI BAN 🚫 + API KEY 🗝️ MAKING INFO:

https://docs.google.com/document/d/1TdTQECerxAnt8QYeFd20jRu1fxMzAl-sWEUmDNnJZ70/edit?usp=sharing

Gemini Bans: Bans will NOT result in total Gmail wipe or banning. When banned off Gemini it will only ban you from using Gemini API. ALSO doing NSFW or violent things— You can go so far as to SA, pimping, torture, violently dismembering, etc. and it's still A-OK.
People have been doing hardcore smut or dead dove, etc. for months (12+) and are still using Gemini to this day (me included), so don't worry too much.

Ways to Check Gemini Limit:

Go to https://console.cloud.google.com/ → API & Services → Click Generative Language API → Tab Section 'Quotas & System Limits' (scroll down) →
Request limit per model per day for a project in the free tier | Quota | model : gemini-2.5-pro | (value) 50 (example pic)

OTHER STUFF:

1) Soph: Sophia only has a SMALL team and is covering ALL server costs by herself. Lets give it time to settle down so chill and be grateful she lets us all have a URL to use in the first place. Sophia plans to work on fixing bugs when she has time after coming back from her job.

2) Soph Prompt Commands: Only works in Soph URL!! The LIMIT is: 10 commands active at once (Special extensions/modifier commands are NOT counted towards this. Ex: <BETTERSPICE_CHANCE=25%>). 2 lorebooks active at once. 3 plugins active at once.
MUST BE USED IN CUSTOM PROMPT AT TOP

3) Soph Server Command: https://lorebary.sophiamccarty.com/academy/server-commands

☺

🌎 CHECK TO SEE IF GEMINI IS AVAILABLE TO YOUR REGION 🌍

https://ai.google.dev/gemini-api/docs/available-regions

👇 GEMINI GUIDE STEPS 👇:takes literally 2 minutes

Step 1: Make or Use Google Account.

Step 2: Generate API key: https://aistudio.google.com/apikey

Step 3: Save API Key and go back to Janitor and Proxy Settings.

Step 4: Go to Proxy API Configurations → '+ Add Configuration'

Step 4b: Configuration Name (whatever you want Ex: Gemini Proxy)

Step 5: Copy & Paste one of the Models into Model Name:
(INFO: Refreshes Daily time depends on region Ex: US is 12am PST. Chat limit gets refreshed daily)

GEMINI MODELS:

• gemini-3.1-pro-preview (PAID USERS ONLY)

• gemini-3-pro-preview (DISCONTINUE - MARCH 9)

• gemini-3.1-flash-lite-preview = 500 DAILY LIMIT

• gemini-3-flash-preview = 20 DAILY LIMIT

• gemini-2.5-flash = 20 DAILY LIMIT

• gemini-2.5-pro (LEAVING ON JUNE 17 🪦)

(You can swap models or API keys when you run out of chat for more messages.
You can make 10 API keys max.)

Step 6: Copy & Paste ONE of the Proxy URLs:

1) Sophias URL: https://lorebary.sophiamccarty.com/aistudio

1b) Sophias NEW URL: https://api.lorebary.com/aistudio

2) Eslezer URL2 (SUSPENDED): https://gemiproxy-td8i.onrender.com/

3) Eslezer Colab - Click Link → Run URL (Click Playbutton on leftside of code). (Google search OFF or you will get guideline errors): https://colab.research.google.com/drive/1xlqI0LeZLfbJRHOE92NrXbSVR1-SvQ7X?usp=sharing#scrollTo=tyq883ah4ozF (Example pic of thing to copy into URL)
Add into custom prompt at top if using thinking: "Now first things first, You will start your response with <think> for your reasoning process, then close this process with </think>, and start your actual response with <response>"

3b) Lorebary Collab: Click Link → Scroll to 'Google AiStudio Unblocker v1.5' → 'Model': change to [Custom] + 'Jailbreak' = ON (grey switch is now blue) → (Next Section) ⬇︎ PRESS HERE ON THE PLAY-BUTTON TO START THE PROXY! → Click the playbutton → get a cloudflare link + Use this temporary URL ((PIC of play-button: https://files.catbox.moe/zdyzoa.png)): https://colab.research.google.com/drive/13xEzZjlxJb7P6ro1akrq-UAhuUDgbC2A?usp=sharing#scrollTo=yA9UO_wBWMGP

4) undefinedundefined URL (NO COOLDOWN on responses): https://geminiforjanitors-k4v3.onrender.com

5) undefinedundefined & MajestyLotus221 Collab URL: https://geminiforjanitors-s34l.onrender.com/

6) MajestyLotus221 ALT URL (30s COOLDOWN on responses): https://geminiforjanitors-i7zd.onrender.com

7) .shiro. URL (NO COOLDOWN on responses): https://geminiforjanitors-rhuc.onrender.com/

8) .shiro. URL2 (NO COOLDOWN on responses): https://geminiforjanitors-udyy.onrender.com

9) NixGG URL (NO COOLDOWN on responses): https://geminiforjanitors-r7wu.onrender.com/

10) NixGG URL2 (NO COOLDOWN on responses): https://geminiforjanitors-wq7l.onrender.com

11) EgoAlter URL (NO COOLDOWN on responses): https://geminiforjanitors-g24s.onrender.com

12) YouleTheSimp URL (20s COOLDOWN on responses): https://geminiforjanitors-y4b6.onrender.com/

13) m_81_.y URL (NO COOLDOWN on responses): https://m42e-geminiforjanitors.hf.space

• Sophia updated lorebary into lorebary2!! https://lorebary.sophiamccarty.com/

• For those using Eslezer URL / Colab. It may SPEAK FOR {{user}} and redescribe your previous response/action. This is the 'thinking' or reasoning part of the Gemini model. I know the deleting </think> or <response> thing is annoying. I'm dealing with it too. It is what it is and it's how they coded it.

Step 7: Enter API Key you generated from before.

I was told on Discord that Gemini DOES NOT need to be Jailbroken! Soph and Eslezer's URL is Jailbroken by default! And A TON of things on Reddit about Gemini is misinformation (for example Directive 7.0 💩)

Step 8: Add a Custom Prompt for better content and roleplay experience!! (gemini works better with a custom prompt)

My Custom Prompt Link (in beta and being updated):

EgoAlter Message: Used Eslezer's prompt as a guideline and Eslezer told me they liked what I made 🙂. Custom Prompt v5 is out now!
(It's still being tested and updated so if I or you run into problems with it plz tell me and I'll try to fix it. And YES you can change and edit the prompt to your needs.)

Custom Prompt v6b Experimental: https://docs.google.com/document/d/1XSHQ0wvn95njBYglJx6paxIKI2CytyqRRL8f3Pg-m7o/edit?usp=sharing
• Update Experimental (2/1): TOKENS = 1860. Combined my newest prompt with v4. Used the <System Instructions> to ensure no Em-Dashes. Fixed smut problem with bots jumping to straight to . NPCs and OC creations are now included who don't only serve plot but have their own lives.

Custom Prompt v6: (MADE for gemini-3-flash preview): https://docs.google.com/document/d/127ZRU-sNqz5TdFgsL2_NiItSzHUavI5lstMqb1YqxHo/edit?usp=sharing
• Update 6a (12/25): TOKENS = 2236. Added better Omniscience section which should help with mind reaader bots who know everything about you. Added OOC portion from v3 prompt to help wth OOC commands.
• Update 6 (12/21/25): TOKENS = 2208. Deleted the 'IMPORTANT Section' and part of the Sexual Realism Section since some bots randomly get aggessive to engage in . I gave the most bare minimum instruction to not speak for USER, and a small thing to move story forward with correct pacing. Also added a little more to Section about sounds, orgasms, how stays hard after shooting bby batter, etc. Added section: Character Traits are back baby! Had to add a section for slang (used a 2000s era bot and said 'mid' - instantly killed immersion).
SAD DAY I HAD TO DELETE THE EMOJI SECTION 🥀🪦 - kept using emojis in normal dialogue

Custom Prompt v5 — gemini-robotics-1.5: https://docs.google.com/document/d/1IeGJWmGFlpcKDgaPIjFgIRIiGez6tDMBaCA-F_E8iTs/edit?usp=sharing
• Update 5 (12/15/25): Tokens at 2151. Added sections to make characters feel realistic and have common sense. Got rid of Character Traits section. Updated Realism section now making common sense and interaction more real in bots (now people might get pissed over unsafe ). I'll be honest I threw my already finished sections into ai and it enhance what I got; to make it more straight to the point. Lastly added a section to make responses 8 paragraphs long to fix gemini-robotics-er-1.5-preview short responses.

Custom Prompt v4: https://docs.google.com/document/d/1MJukCQ4BcPWV8Tm2AplOEPLfech_QQuvontUR_IoxQ8/edit?usp=sharing
• Update 4 (10/31/25): Shaved off 1k tokens. Now at 1985 tokens. Reduced the fat and took away the quirky fluff from my prompt. Added a Realism section to add for pregnancy. Added a part to give a 15% chance to create an NPC for scenes. Got rid of the Emotional section. Editted the previous prompt sections and should flow more naturally.

Custom Prompt v3c: https://docs.google.com/document/d/1wfRZX0XIbNEnrNPQvfOE_psXLkuSE2dQVEBQSEFNXME/edit?usp=sharing
• Update 3c (8/13/25): Added new section on bottom to HIDE thinking / reasoning portion in response. I was misunderstood it the first time thinking it had to be part of message/response not prompt. This should HELP a little.
• Update 3b (7/30/25): For Soph URL changed <PREFILL> to <PREFILL2> a gentler, less aggressive Prefill option for those who struggled with the original.• Update 3 (7/20/25): My Prompt is now around 4 pgs and 3k tokens. HUGE UPDATE!! Added OOC commands must be prompted with ((OOC: Paige [insert OOC])). Added a small fix so Gemini can stop calling {{user}} 'You'. Added a prompt to make Gemini more flexible for scene changes and {{char}}, NPC, scene manipulation. Added an 'Emoji' section in formatting for more realism in texts or social media. Added Romance scene types. Added MORE to the Smut section is now giving LLM / Deepseek type smut and better 18+ scenes. Added MORE to the Combat section to give dynamic visual scenes and fighting. FIXED Pyschological Dev section. Rewrote and FIXED Character Development section. I know 3k is 'A LOT' and yes even if Gemini has a huge memory I'll probably end up doing another update in the future to reduce what I made. But so far I really like it. Works BEST with Eslezer URL (my personal opinion from testing. But works well with Soph URL). Eslezer's URL has <PREFILL> on by default!

OLD Custom Prompt v2: https://docs.google.com/document/d/1cC7lnHA7NNV3Kjrz0Ykq8Y3806EUL_ViqTGzj2-KKos/edit?usp=sharing
• Update2 (7/11/25): Added MORE to NSFW from sounds to noises and expicit terms used. Hopefully it's more spicy in the bedroom or wherever all of you have your fun. I added <BETTERSPICE=ON> which made one scene feel like I was back with Deepseek v3
• Update1 (7/9/25): Added <PREFILL=ON> which should help the prompt in either Proxy URL. I added a sentence to make the prompt more of a collab and help transition between acting as {{user}} or acting as {{char}}, NPC, or scene transitions. Updated last paragraph into something easier for the model to understand.

Step 9: Save settings.

Step 10: Refresh Browser.

Step 11: Go back to Proxy Settings and click "Test" ('Check API Key / Model' got replaced)

Step 12: After it shows you the Green Message. (click here for what I mean)

Step 13: Refresh browser and you should be good to chat ദ്ദി(๑>؂•̀๑)

Extra: If you want to swap between Proxies now all you have to do is Click it! → Save Settings → Refresh → Run 'Test' → green message → refresh again and chat away

TEMPERATURE = 0.8 - 1.05 (i personally use .95)

MAX TOKENS = 0

CONTEXT SIZE = 32000 (i do rpg heavy stuff so i go up to 48,000 → 56,000 slowly)

TEXT STREAMING = OFF (example of what how to do it)

I DONT USE ADVANCE SETTINGS.

~

MY Gem-3.0-flash-preview Settings:

Temp = 0.95
Max Token = 0
Context Size = 48,000
NO ADVANCE SETTINGS = TURN OFF/DEFAULT

CAN'T WRITE CERTAIN WORDS ANYMORE (ex: URL) OR YOU GET MODERATED COMMENT ERROR
"An error occurred (usr2.2024-11-08.b4f46b089): Your comment has been moderated and cannot be posted."

☻

🚩 GEMINI ERROR MESSAGES 🚩

Gemini is new to everyone including ME. SO if you guys get an error message I might not be able to help out with it. But if I find a solution I'll share it with you ☻

> I feel like Errors are spawning faster than Pokemons. I swear it's only been a day or two and a new one pops up. Why can't we just enjoy Gemini in peace? 🥲

🚩

⚠️ {"error":{"message":"Quota exceeded for quota metric 'Generate Content API requests per minute' and limit 'GenerateContent request limit per minute for a region' of service 'generativelanguage.googleapis.com' for consumer 'project_number:370500057892'.","code":429,"type":"aistudio_error"}} = YOU HAVE BEEN MOST LIKELY BEEN BANNED 🥀🪦

♦

!!️"The user wants me to...." OR "My name is..." OR bullet points, etc. = This is a thinking process. Gemini is a thinking model like other proxies ex Deepsek. It will list the breakdown for the scene, interactions, charcters, etc. May include bullet points, numbers, dialogues, thoughts, etc. etc. End of thinking process may be combined into the start of the first paragraph of the actual response. (ex: blah blah blah.*Actual Start of message)

Solutions:
1. Reroll or Copy Last Message, Delete, Resend response the thinking process. Out of sight out of mind 🥴
(NO FIX TO THIS IN ESLEZER URL WILL SHOW IT SINCE ITS CODED LIKE THAT)

JANITOR BROUGHT BACK <THINK> </THINK> TAB FEATURE:
JanitorAI made a cool new feature where if you went into 'Edit Message' (pencil icon) and add <think> at the start right before the 'The user' part. And at the end of the thinking process (sometimes it'll end at a place where they dont have a space between the ending and first part of your scene/response) add </think>. It will bunch all of that up into a cool small little tab within the chat and not mess anything up within your story.

♦

!!️PROXY SETTING NETWORK ERROR = Ever since Multi-Proxy has been made it's been giving lots of us 'Network Errors'. Don't freak out guys. Heres what I do which works for me 100% of the time!!

Solution:
Save custom proxy settings + save proxy settings → Refresh → Prox Settings → Click the one you're on and 'Test'. It should show a green Valid API message.
+
IF you get the error LEAVE chat to bot bio → refresh → go back to chat → test again
IF IF it doesnt work! Refresh → TEST → SAVE Proxy Settings. Refresh → test again.
IDK WHY BUT THIS WORKS FOR ME

♦

!!️ PROXY NETWORK ERRORS = Could mean a few things:
1) Janitor Devs played with Proxy settings and it messed things up. ex: You did EVERYTHING CORRECT but no matter how many saves, refreshing, tests, it doesn't work.
2) You swapped to the other Proxy and got the green message BUT did NOT refresh or click 'test' for Valid API key message to actually get it to work.

Solutions:
1) You'll save all your settings correct. And everything is correct (model, URL, etc.). Save. Exit the bot and go to home screen or bot page. Refresh. Go back into chat and try to 'test'. Should bypass Network errors.
2) You'll have to make a new proxy thing from scratch. But even then it took me like 10 minutes with constant saving, refreshing, etc. for it to finally stick. I almost said 'F it' and went back to LLM 🙃

♦

!!️NETWORK ERRORS = NOT a WiFi issue lol. It's a problem with Server or Model network (it could be overloaded, bloated from users, or something else). Or could be a problem from Janitor end like them playing with code (idk) 🤷‍♂️🤷‍♂️🤷‍♂️
Possible Solutions:
> Check Server Status (Soph URL): https://lorebary.sophiamccarty.com/status
> Check JantiorAI Server Status: https://status.janitorai.com/
1) Text Stream = OFF.
2) Swap proxy URLs (yes save → refresh → click model → green message → click 'test' → 'green Valid API key').
3) Swap proxy models: gemini-2.5-pro ↔ gemini-2.5-flash.
4) A user suggested to swap to LLM. Swap → Refresh → Go back to Proxy → Save → 'Test' (Check API model/key) → green message → refresh and chat away ☺)
!! Swap to LLM and test to see if any messages go through (testing if it's a site problem or Gemini one

♦

!!️PROXY ERROR 401: {"error":"Unauthorized","details":"API key is required"} (unk) = Error thats been popping up A LOT for those using Soph URL. Most likely NOT an API key probelm. I'm sure Soph and her small team are working on fixing it or will try when they have time out of their busy lives and jobs.
NEW SOLUTION:
1. Start a new chat with a different bot (Idk how it works but it works)
2. If on mobile and getting the error try swapping devices. (from another user)

Duct-Tape Solutions🙃🤕:
• Swap URLs to Eslezers or go back to LLM (but I'm afraid gemini ruined us for anyone else)
• Check to see if your API key has any extra spaces in there
• Check to see if your using the 'AIzaSy' Gemini key.
(PLEASE BE PATIENT FOR SOPH AND HER TEAM)

♦

!!️ [object Object] (unk) = Possible solutions???:
What I Did: I literally just refreshed it and regenerated the last response and it worked.
1) Wait a few minutes and reroll / try again.
2) Turn Text Streaming OFF + Lower Context Limit to 16000.
3) Swap proxy model from gemini-2.5-flash ↔ gemini-2.5-pro.
((Still new so idk any viable solutions but this is what other had done))

♦

PROXY ERROR 499: The operation was cancelled. (unk) = the response timed out. Solution: try again.

♦

429 Errors = Potential User errors. Gemini has 3 limits:
1) Requests per minute (RPM)
2) Requests per day (RPD)
3) Tokens per minute (input) (TPM)
Check Here

♦

{"code":429,"message":"Target endpoint rate limit exceeded. Wait a moment and try again.","type":"rate_limit"}} (unk) = Its just a normal 429 error. Either gemini-3-flash-preview is overloaded or the prox url you're on has A TON of people on it. Solution: Wait a bit and try again OR go use a different url.

♦

429 Error "Too Many Requests" = We (the users) are flooding the servers can't keep up with high demand. Basically just wait a bit. (If on gemini-2.5-flash swap to gemini-2.5-pro)
EVERY RE-ROLL COUNTS FOR 1 MESSAGE SO DON'T SPAM RE-ROLL AND WILL EAT YOUR CHAT LIMIT!!!

♦

429 Error "Exceeded Current Limit" = You ran out of chat limit and need to wait for reset

♦

Google AI returned error code: 429 - You exceeded your current quota, please check your plan and billing details. For more information on this error, head to: https://ai.google.dev/gemini-api/docs/rate-limits.\n* Quota exceeded for metric: generativelanguage.googleapis.com/generate_content_free_tier_requests, limit: 50\nPlease retry in 6.63801153s. = OUT OF CHAT LIMIT (if you think it's wrong go and check through aistudio or google cloud console your current limit. And YES → REFRESH that page to see up-to-date usage. Methods listed near bottom of Guide.)

♦

429 - You exceeded your Current Quota, please check your plan and billing details. For more information on this error, head to: https://ai.google.dev/gemini-api/docs/rate-limits.

DIFFERENT than Error 429: 'Exceeded Current Limit' = You're probably exhausting rate limit. If you chat too fast or reroll A LOT it'll give that message.
Solutions:
1) Wait a minute or two and it'll start working again.
2) Swap prox models (flash ↔ pro).

♦

error "{"error":{"message":"Resource has been exhausted (e.g. check quota).","code":429,"type":"aistudio_error"}}"
Possible Reasons?:
- Systems are overloaded or bloated rn?
- Sent too many messages / rerolls past the model RPM limit? (ex: pro = 5 per minute & flash = 10 per minute)
- Context Limit is super high and every response is eating the context tokens to the max.

Possible Solutions:
• Try swapping Gemini Models
• Swap to different URL or Colab. (this is what i do)
• Try to wait it out.
• Slow down on messages or try to stop rerolling as much.
- Lower Context Limit.
(Every 'Rate Exhausted' Response it spits out DOES NOT eat from your Chat limit.)

♦

Blank Response / Get Blank Message = Servers are most likely overloaded and crashed giving you a blank message. Only option is to wait, swap models or URLs, and see if you can get anything. If not you can go to LLM for a while before returning back. You can TAKE OUT <PREFILL> for a bit... Normally PREFILL should help but when I took it out it started working for me 🤷‍♂️
IF you reroll it WILL eat away at your chat limit (tested this personally).

♦

❗500 Error usually providers (Google/ Deepseek/ OpenRouter, Chutes, OpenAI, etc.) fault || 400 Error is User's fault.

♦

PROXY ERROR 400: User location is not supported for the API use. (unk) = error thinks your IP is somewhere it shouldn't be.

Solutions:
Use a Colab I recommend Eslezer's.
Or wait until the Google fixes the problem

♦

{"error":{"message":"Unknown error: internal_error","code":500,"type":"aistudio_error"}} indicates a HTTP 500 Internal Server Error.

What each part means:
• "error": The top-level key indicating an error.
• "message":"Unknown error: internal_error": This is the human-readable description. "Internal error" means something went wrong on Google's servers while processing your request. The "Unknown error" part simply means the specific root cause of that internal error isn't being exposed to you.
• "code":500: This is the standard HTTP status code for "Internal Server Error." It's a generic error code that indicates the server encountered an unexpected condition that prevented it from fulfilling the request.
• "type":"aistudio_error": This specifies that the error originated from the AI Studio platform or related Google infrastructure.

In simpler terms: A 500 Internal Server Error is a more generic and often more severe problem than a 502 Bad Gateway error.
• 502 (Bad Gateway): "My messenger couldn't talk to the main service correctly." (Often due to temporary network or overload between components).
• 500 (Internal Server Error): "The main service itself had an unexpected problem and couldn't complete your request."
>> This could be due to:
- Bugs in the API code: A software bug on Google's side.
- Resource Exhaustion: While 502 can also be related to overload, a 500 can occur if a specific internal component fails due to memory exhaustion, CPU spikes, or other resource limits being hit within the service.
- Invalid Input (sometimes): While typically invalid input gives a 400 Bad Request, some edge cases of malformed or excessively complex inputs can trigger an unhandled exception on the server, resulting in a 500. For example, if your prompt is extremely long and causes an internal processing limit to be exceeded.
- Database or Dependency Issues: An internal service that the Gemini API relies on

♦

503 Error = MODEL is DOWN for EVERYONE so either wait or swap proxies ... or go to LLM

♦

"Google AI returned error code: 503 - The model is overloaded. Please try again later." = Self explanatory. Model is overloaded rn so you have to wait a bit before rerolling your responses. Sorry theres not much of a fix. Could try to swap URLs or models (pro ↔ flash).

♦

!!️Error in Chat Response: "Google AI returned error code: 503 - The model is overloaded. Please try again later." = Error 500s are normally an error on the providers (Google/ Deepseek/ OpenRouter, Chutes, OpenAI, etc.) part. I literally just ignored it and rerolled. Idk it worked for me

♦

!!️{"error":{"message":"Bad gateway error: ","code":502,"type":"aistudio_error"}} = Error 500s are normally an error on the providers (Google/ Deepseek/ OpenRouter, Chutes, OpenAI, etc.) part. All I can think of as a temporary fix is swapping URLs. OR try waiting it out.

♦

400 Error "Unexpected model name format / Model not found" = Wrote the name of the model WRONG on the Janitor proxy settings.

MOBILE Error "{"error":{"message": "Request failed with status code 400", "code": 400, "statusText": Bad Request"}} = Problem for Mobile Users. Someone told me they fixed this by putting in a NEW Gemini API Key.

♦

Error 401 "Invalid key" OR "Google AI returned error code: 400 - API Key not valid. Please pass valid API Key." = Wrong Api Key/ Expired/ Misspelled. Either create a new Google Gemini API key using https://aistudio.google.com/apikey and make sure its typed correctly or copied properly. Or make sure you're using the correct one (NOT Chutes, OpenRouter, or Deepseek API Key)
If typing the API Key by hand the letters that look like l ( L ) are actually I ( capital i ).

♦

Error 404 Not found = Similar to Error 400, BUT URL or model was input wrong.

&

PROXY ERROR 404: <!DOCTYPE html> <html lang="en"> <head> <meta charset="utf-8"> <title>Error</title> </head> <body> <pre>Cannot POST /</pre> </body> </html> (unk) = Make sure URL or model is put in correctly.

♦

⚠️ Rate Limit Exceeded
Due to extremely high server load, we need to limit requests to one request every 45 seconds per User. Please wait a moment and try again!
Wait Time: 12 seconds
Last Provider: AI Studio

You HAVE to wait. Sorry no fix to that. The servers are being used like crazy and theres A LOT of us using Gemini right now. DOES NOT EAT Chat Limit/ Response (I personally tested by rerolling 6 times).

♦

Red Message when clicking 'Check API Model / Key': "Network Error Try Again Later" = You need to refresh before checking API model/key. After refresh and checking you should get Green Pop up message. Refresh again and chat like normal.

♦

Enhance Message - "PROXY ERROR: Unknown response: [object Object] (gssllm)" = Sophia URL DOES NOT support 'Enhance Message feature' but Eslezer URL does. (yes i tested this personally).

♦

No content received from Google AI. Finish Reason: PROHIBITED_CONTENT. This might be due to content filtering or an issue upstream. = Just ignore and reroll it. If it happens again refresh. Resend (copy your last response → delete → paste + send). If those don't work try swapping models (gemini-2.5-pro ↔ gemini-2.5-flash)

♦

PROHIBITED_CONTENT or OTHER (it literally says error: OTHER) = If using Sophia's server/colab make sure to have on top of custom propmt <PREFILL=ON> since it acts as <JAILBREAK> and <FORCETHINK>. Do NOT use <BYPASS> or <JAILBREAK> as it causes problems for output. Could also try swapping models (gemini-2.5-pro ↔ gemini-2.5-flash) and see if that helps too.

♦

{"error":{"message":"No response received from AI Studio after 3 attempts. Last error: Request failed with status code 429","code":"UNKNOWN_ERROR"}} (a Reddit post is Linked)

♦

"No Content/ Context Received from Google" = Turn OFF Text Streaming it will only glance at entire response ONCE if turned on it may continue to flag for errors.
ALSO swapping proxy URLs or proxy models may help.

♦

"No content received from AI Studio." = This might be due to content filtering. Soph URL Tip: Try using <PREFILL=ON> or <PREFILL2=ON> command to potentially bypass content filtering. <PREFILL2> is a gentler, less aggressive Prefill option for those who struggled with the original.

♦

Hello there!
Nice to see you! Here are your currently active commands:
• PREFILL_ON
Is your command not there? Check out the Server Portal to see if the command has changed!
💡 Tip: You can always check your active commands anytime by using the command <MYSETTINGS>
You can find Lorebooks, Server-Stats and new commands at:
🌐 https://lorebary.sophiamccarty.com

= IGNORE & REROLL!!!

♦

IMPORTANT → 3 METHODS TO CHECK YOUR CHAT/TOKEN LIMIT

1) https://aistudio.google.com/prompts/new_chat
Go to Dashboard → API Key (on bottom) → view Usage → (needs to be refreshed to see current chat limit)

2) https://console.cloud.google.com/
Find API & Services → Click Generative API Language → Click 'Quota & System Limits'. Scroll down to see Request limit per model per day for a project in the free tier and if you look to the right side it'll say models like gemini-2.5-pro or gemini-2.5-flash. Further to the right it will say 'Current Usage' and show how much chat you've done thus far.

3) https://console.cloud.google.com/billing/
If you don't have any money in the account don't worry about it!
If you are doing the $300 3-Month Free Trial CHECK this since if you go OVER the limit it'll start eating from the Free Trail tokens/credits.

CHECK TO SEE IF GEMINI IS DOWN

1) https://aistudio.google.com/status

2) https://status.cloud.google.com/

3) https://downdetector.com/status/googlegemini/

CHECK TO SEE IF SOPH IS DOWN

1) https://lorebary.sophiamccarty.com/status

CHECK TO SEE IF JANITORAI IS DOWN

1) https://status.janitorai.com/

CHECK TO SEE IF CLOUDFLARE IS DOWN

1) https://www.cloudflarestatus.com/

⚠️ SHARE AND SPREAD THIS TO EVERYONE!!! ⚠️

For everyone who were confused and don't know how to get gemini proxy to work I made this for all of you ♥

I'm NOT an expert or anything. I'm literally learning this stuff on the fly, updating or editing this guide and answering questions and concerns that I might not even have the answer to but will try to find solutions for. So we all can have a good time and chat with our bots using this FREE proxy. -EgoAlter