Gemini CLI becomes extremely slow (1+ HOURS) / stuck during small code-edit tasks (agent loop + model response delays)

[nitin-stockypro]

What happened?

I'm experiencing severe performance issues when using Gemini CLI for very small tasks (editing 1–3 files). The CLI frequently gets stuck for several minutes even after the edit is already completed.
In one case, Gemini CLI took 13–14 minutes just to answer a simple question ("which model are you using?") while processing a very small code change.
This seems related to the agent verification loop and possibly model routing when using Auto.

Environment :

• Tool: Gemini CLI
• Model setting: Auto (Gemini 3)
• Auth: Browser-based login (not API key)
• Sandbox: disabled (no sandbox)
• Repo size: medium (~few hundred files)
• OS: macOS
• Task scope: 1–3 files

Example Task

I asked Gemini CLI to fix a simple validation issue in a React form.

Scope:
base-form.tsx
create-link-version-form.tsx

The change was very small (changing form validation mode).

Example diff:

const form = useForm({

• mode: 'all',

• mode: 'onChange',
  resolver: zodResolver(LinkVersionFormSchema),
  defaultValues,
  })

Asking a simple question like "which model are you using?" took ~13–14 minutes to receive a response.

What did you expect to happen?

Client information

Client Information

Run gemini to enter the interactive CLI, then run the /about command.

> /about
# paste output here

Login information

No response

Anything else we need to know?

No response

[gemini-cli]

Found possible duplicate issues:

• CLI Agent Intermittent Long Loading Times/Hangs During Tool Execution #17442
• Gemini CLI model gemini-3.1-pro-preview long response time #21143
• [v0.33.0] Infinite "Taking a bit longer" loop despite Web UI/Other models working fine #22080

If you believe this is not a duplicate, please remove the status/possible-duplicate label.

[abulgher]

I am experiencing the same frustrating situation. This message: it is taking a big longer than expected, still working on that has become my nightmare.
Hitting F12 is showing a lot of errors status 429: Too Many Requests. Is this the root cause of the slowness?

headers: { │
│ 'alt-svc': 'h3=":443"; ma=2592000,h3-29=":443"; ma=2592000', │
│ connection: 'close', │
│ 'content-length': '624', │
│ 'content-type': 'application/json; charset=UTF-8', │
│ date: 'Thu, 12 Mar 2026 21:09:38 GMT', │
│ server: 'ESF', │
│ 'server-timing': 'gfet4t7; dur=1568', │
│ vary: 'Origin, X-Origin, Referer', │
│ 'x-cloudaicompanion-trace-id': 'f2e031d2792dbb43', │
│ 'x-content-type-options': 'nosniff', │
│ 'x-frame-options': 'SAMEORIGIN', │
│ 'x-xss-protection': '0'

[hutaoshusband]

Same here it is really unusable. It was not even usable when I was subscribed to Ultra Plan

[gt732]

extremely slow on the ai pro plan as well...not even usable..went right back to claude/codex

[ethanwaldo]

Slow for me too

[jbeers]

If it helps anyone, I was able to resolve this issue by toggling my model selection. I was on gemini-3-pro specifically. When I changed it to auto it started working.

[DevDokus]

I ahve the same issue. For a good hour or 2 it worked very fast. The last question i went away from the pc and it was still busy for a good 1.5 hours and still no response.

I deleted the tmp folder, the .cache and even deleted my old project chats within the .gemini folder and logged back in with my email. Now on a complete fresh instance I executed the comand /memory refresh what should execute within seconds. Now currenly 7 minutes in with no response, and even the tips you see when working do not display any working lines at all, just tips and jokes.

I have enough usage left on all models and have set it even to auto choose model. But still nothing is working one bit, becomes really frustrating, cause this happens daily and it stagnates the complete development.

Hope this will be fixed soon cause if i have to wait 9/10 min or even an hour for a responce, gemini-cli simplu becomes useless.

Edit: I do have to mention that i have a .geniniignore file containing the following data.

`# CRITICAL: Add these at the top to stop the 10-minute hang
.venv/
venv/
env/
.git/
.gemini/
pycache/

Dependencies

node_modules/
bower_components/
vendor/

Build outputs

dist/
build/
out/
.next/
.nuxt/
target/
bin/
obj/

Version Control & IDEs

.git/
.svn/
.hg/
.vscode/
.idea/
*.swp
*.swo

Environment & Secrets

.env
.env.*
*.pem
*.key
auth.json

Large Locks & Metadata

package-lock.json
yarn.lock
pnpm-lock.yaml
.eslintcache
.stylelintcache

Logs & Temp

.log
npm-debug.log
yarn-debug.log*
yarn-error.log*
.gemini/tmp/
tmp/
temp/

Media & Assets

*.png
*.jpg
*.jpeg
*.gif
*.svg
*.mp4
*.pdf`

[arkarn]

same issue

[DevDokus]

Okey for those that have isuses with it keep in mind the following. Even if you have enough usages left on your models, we are still competing with everyone that used the Gemini-Cli Api. When many people use this at the same time, it van happen that the "slots" available area ll in use. You have to wait until a new slot comes available.

furthermore, when you are working large projects we are making a lot of Api calls as well as using many told, like the ode inspector assistant, editing files ect. When you use these a lot, no matter if you have enough room left on your models, you WILL be rate limited. This mostly happens when you use the --yolo tag starting Gemini, as now gemini will automatically call upon the tools at its own will. This will result in many many calls per minute what will easily trigger the rate limitation on your account. This makes the Gemini CLI stop reacting at all, or it takes a long time until your rate limit is finished before it continues " as long a slot is available.

Also keep in mind that the Cli will make also a .cache what stores your data and client runtime information. This cache keeps growing each time you interact with the AI. The larger this cache folder, the slower the AI becomes to respond. Clearing the cache will help it to speed it up again.

Then you have the % context stored. When this gets too big, each time you make a new request the AI needs to read all recent chat history to understand what you want to do next, this also increases the response duration quite significant. recommended is to often use /compress so it compresses all your chat history within a quick summory what will speed up the responses again.

Then you also have the memory usage that will increase at each request. this can quickly go into the gigabytes, the more you have in memory, again the slower the Ai becomes as it tracks all the data stored to make the best response.

Then you also have the amount people that uses the AI at the same time. This reduces your response time because the max performance of the AI is shared between all users. So if their servers are heavily used, we all experience a longer delay in the responses.

Then you also have the .geminiignore file. if you do not include this, Gemini will index ALL your project files incl depencencies like node_modules that contain 1000s of small files what take a long time to index. Make sure to have the .geminiignore file within the root folder of you project. keep in mind that adding the ignore file AFTER you created your project, the AI often fails including this new file especially when you where already working within your project and the Gemini.md file was already created.

Then you have also the following. if you install the latest version of NodeJS gemini can start to show some issues. They say to use NodejS version 20+ but when installing 24+ issues can popup. i would downgrade your NodeJS and NPM package manager as this does not contain the bugs that can stagnate gemini cli.

Last but not least, when you have tried to do a /reload - /refesh within a project that did not have the ignore file yet it will index all files even the node_modules as well as you 'venv' '.venv' and other folders the application creates on runtime. This increases the amount of files ( depending how much you have ) up to a hour to index it all. This can cause the AI to trip into a infinite loop, and if you cancel this the cache can become corrupt, or internally the infinite loop was never stopped and still running within the background. Due to that all other requests will either time out, or also fall into a continues loop.

Tip: Follow these steps and it will work again for you.

• Uninstall Gemini CLi: sudo npm uninstall -g @google/gemini-cli
• Uninstall Node: sudo apt-get purge -y nodejs npm and sudo apt-get autoremove -y
• Get the right version of node curl -fsSL https://deb.nodesource.com | sudo -E bash -
• Install NodeJS: sudo apt-get install -y nodejs
• Install Gemini: sudo npm install -g @google/gemini-cli

May You get errors

• Break the circular link: sudo apt-get purge -y nodejs then sudo apt-get autoremove -y
• Purge NodeJS: sudo apt-get purge -y nodejs then sudo apt-get autoremove -y
• Install Node and NPM: curl -fsSL https://deb.nodesource.com | sudo -E bash - then sudo apt-get install -y nodejs
• Install Gemini: sudo npm install -g @google/gemini-cli

Warning:
The latest version of gemini 0.33.1 has sometimes some issues with this infinite loop, specially on Linux. Downgrading 1 or 2 versions will stop this from happening. Just make sure that you disable auto updates within the settings. Then when new updates get released do not instantly update!! Wait for a couple weeks to make sure they worked out all the kinks and bugs out of the new version.

Another Tip:
Even if you reinstall the gemini cli, reindexing your old project will still cause the AI to ignore the .geminiignore file. This will result again into a long index that can trigger that some infinite loop. I would suggest to create a new folder, and add in the .geminiignore file as the ONLY file within your folder. Then open gemini and execute the /init command. This will create the Gemeni.md file once again. The file will be empty accept the the information that Gemini needs to always take the ignore file into account, this will now be written within the instructions of the Gemini.md file.

Then you can add back in your project and simply ask gemini to look over the project, index it and update the Gemini.md file to reflect your current project. Only this time it WILL abide by the ignore rules and thus it will skip all folders and files you have specified.

Make sure to refresh the chat when you made changes to the Gemini.md or the ignore file so it loads in the newly added entries you made.

This should at least prevent the issue where Gemini stops responding to your requests, though keep in mind this does NOT fix when you get rate limited due to calling upon tools to often. if you want to prevent that you should not run gemini with the --yolo tag, as this massively increase the tool calls automatically what will quicker send you into a rate limit.

I made to .sh file you can execute in your terminal to clear gemini's cache and other data that could stagnate your client.

File 1: Gemini reset: This will terminate all gemini background services, then it will delete the global brain data like the .gemini folder as well as its gemini_config. Then it wipes the temp data of gemini. After that it will uninstall Gemini and all its data and dependencies.

GemReset.sh

File 2: This does basically the same but in a slight different way may gemini did not got removed with the first file.

nuclear_wipe.sh

After that you simply run the command sudo npm install @google/gemini-cli. How ever this will install the lastest version, if you experience the same issues it might be the 0.33.1 bug. I would recommend installing a older version

You can see a full list of all gemini version with npm view @google/gemini-cli versions --json

Then choose one of those version and use it like sudo npm install -g @google/gemini-cli@0.33.0.
This is just 1 version downgrade, if that does not work either you can ofcause downgrade even more.

VERY IMPORTANT!!!!!!!!!!
When executing the clean up commands shown above WILL also remove all chat history. Make sure to save your chat history by /chat save name and then extract this folder from the ~/.gemini folder. Then later you can place the chat back in when you reinstalled it, but adviced is just to start off with a new clean slate.

Hope it helps you guys. I have it working here again with responses of mere seconds.

[andreknieriem]

Nearly one time a week something gets sooo shitty. History lost, slow times and so on. I am using the pro model and it just does nothing

[DevDokus]

Nearly one time a week something gets sooo shitty. History lost, slow times and so on. I am using the pro model and it just does nothing

I think that could be as they are updating the models. I had the slow response issue for days but i have found my specific issue. Often its the .geminiignore file not getting used as it should.

Best you can do is creating a new folder, then create a .geminiignore and place in every file and folder Gemini needs to ignore indexing. When you can run the /init command within the empty folder only containing the ignore file, gemini will explicit set the rules for the ignore file.

Then import your project files and prompt to gemini "I have made changes to my project, please scan the entire project and update the Gemini.md file so it reflects the state of my current project". Then gemeni will index again and update this file. You can also add this to the /memory command and then refresh.

The issue that i found is that if the ignore file is created after the indexing of the project, gemini fails use the ignore correctly. The amount of files gemini is indexing is so much that it can take up to a hour to index. Clearing the cache can help also.

But now i have created a new folder, made the ignore file, then used the init command, then imported the project and asked for a new index, it was done within seconds. All my gemini responses are now in mere seconds and not minutes to hours anymore.

The only other reasons gemini might refuse to talk back is when the servers slots are all in use so it throws you in a queue, or simply cancels the request. Make sure to safe the /chat when exiting and open a new session then use /resume. If the memory is filling up its responses become slower and slower.

Also when the --yolo parameter is used gemini makes A LOT of tool calls what will throw your account into a temp lock very fast.

But now i made the ignore file as the first thing within the folder, then i used the /init and then i imported my project. I have no issues anymore with slow responses "at all".