Built In AI Assistant seems dumber than outside version
I have the paid version of AI assistant in PyCharm set to ChatGpt4o, but consistently find myself using my ChatGPT4o version directly from OpenAI.
My experience is that PyCharm AI (PAI) seems to have a very short attention span and doen't remember what was already tried and doesn't work. I'm getting led around in a giant loop, so to speak…
Initially I thought that is due to the project having non-programming aspects (mechanical design etc). But every time I tried to get back to PAI, I get frustrated by the same thing.
Am I the only one?
Oh, let me share a funny experience: a few weeks ago, after finding out about this, I shared my thoughts with PAI: “ You don't seem to be as smart as your twin from outside PyCharm” . PAI promptly hung up on me, unwilling to continue programming. I had to start a new chat! Sensitive too! 😂
请先登录再写评论。
Hello Abhishek;
I don't have sample queries that I kept for 8 weeks unfortunatley. But all this was before PyCharm revamped its whole design. For the last week or so, I have been using gpt5 and it has performed significantly better than gpt4. In fact, it made only one small error all week until yesterday. Now it seems to be going down a rabbit hole again, like it used to do. Because of its good performance, I have not tried GPT5 on the openai platform. But I will do so immediately, because last night JetBrains advised me that I had consumed all of my weekly tokens of the PRO platform. I find this rather rich, given that I have a Pro subscription for PyCharm AND a Pro subscription to AI AND pay for chatgpt on top of it… Today is Monday, and in most calendars that is a NEW WEEK, so I should be allowed back on. So far I'm still locked out, despite two messages to support… makes me wonder if I should just continue with openai and go to basic level at PyCharm.
Hello Jr
Let me address your questions one by one:
Quality of response using the same prompt
When sending a request in chat, AI Assistant passes up to 10 previous messages from the same conversation to the LLM. The exact number may vary depending on the total context size and whether it fits within the token limit, which means that in some cases the AI Assistant may lose previous context during an active conversation.
You can find more information by using the following link:
SUPPORT-A-1650 Effective token management in AI Assistant. How to not exceed the token limit?
Quota limit
Please note that AI Assistant works under a fair usage policy, which defines quota limits for all JetBrains AI plans.
The quota resets every 30 days, starting from the date when you first activated the AI Assistant.
You can check your personal quota renewal date in the AI widget, located at the top-right corner of the IDE window.
You can find more information by using the following links:
Thanks Wladyslaw,
That is nice and clear and congruent with my experience. It seems sometimes AI Assistants loop back into methods proven not to work.
Noted on the 30 day limit. There is some contradictory messaging in PyCharm though:
The notice board says I've reached 85% and it might affect my usage.
And AI says I've exceeded my WEEKLY quota, when it really is monthly.
ChatGPT 5 directly from openAI, seems to have a better memory, although my recent experience in PyCharm was quite good.
BTW, my AI Pro license expires on 20 Feb 2026, which would mean that it was issued 21 Feg 2025. We are now day 22 of the month, so my block should have been lifted. I still cannot access AI Pro.
Cheers!
Hello Jr,
Regarding the message you encountered: You have reached the weekly fair use volume
This is a known UI bug. Our development team is aware of the issue and working on a fix. You can find more details by using the following link:
LLM-18597 Incorrect message for AI Free/AI Pro in case quota is exhausted—You have reached the weekly fair use volume ... contact support to allocate extra volume
As for the quota renewal date, could you please create a support ticket by using the following link https://www.jetbrains.com/support/?
This will allow us to investigate the issue further.