--- title: We Gave AI Agents a Live Jupyter Kernel in PyCharm description: PyCharm introduces a new Jupyter skill that lets AI agents operate a live Jupyter kernel directly through the IDE instead of running code in a throwaway shell... canonical: https://daily.dev/posts/we-gave-ai-agents-a-live-jupyter-kernel-in-pycharm-hrqonqnd7 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: We Gave AI Agents a Live Jupyter Kernel in PyCharm | daily.dev og:description: PyCharm introduces a new Jupyter skill that lets AI agents operate a live Jupyter kernel directly through the IDE instead of running code in a throwaway shell... og:url: https://daily.dev/posts/we-gave-ai-agents-a-live-jupyter-kernel-in-pycharm-hrqonqnd7 og:image: https://api.daily.dev/og/posts/hrqonQnD7.png og:image:alt: We Gave AI Agents a Live Jupyter Kernel in PyCharm og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # We Gave AI Agents a Live Jupyter Kernel in PyCharm **[JetBrains](https://daily.dev/sources/jetbrains)** · 6 min read · 5 upvotes · 0 comments ## Summary PyCharm introduces a new Jupyter skill that lets AI agents operate a live Jupyter kernel directly through the IDE instead of running code in a throwaway shell subprocess. This keeps kernel state (trained models, dataframes, imports) persistent across cells, avoids corrupting the .ipynb file, and lets agents wait on long-running jobs instead of polling and burning tokens. Benchmarking across twelve MLGym ML tasks showed Claude Opus 5 running through the kernel cost about 12% less than through the shell (59.09 USD vs 67.06 USD), mainly because the kernel keeps the prompt cache warmer (98% cache reads vs 82%). The cost benefit was model- and task-dependent: it was clearest for Opus on long stateful jobs, while the shell was cheaper for short tasks and for Codex models (GPT-5.6-based Sol and Terra), which already use caching efficiently. The feature requires a JetBrains AI subscription and is available in PyCharm 2026.2.1. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much cheaper is it for Claude Opus 5 to run machine learning tasks through a live Jupyter kernel versus a shell in PyCharm? Running through the kernel cost 59.09 USD versus 67.06 USD through the shell, about 12% cheaper, across twelve MLGym benchmark tasks. The savings come from the kernel keeping the prompt cache warm: 98% of input tokens were cache reads compared to 82% for the shell, and cache reads cost only 1/12 of creating a fresh cache. With the contaminated Titanic task removed, the kernel was still 10% cheaper (56.34 USD versus 62.65 USD). _Teams weighing agent cost tradeoffs in notebook workflows can follow tooling changes like this via daily.dev._ ### What problem does PyCharm's new Jupyter skill for AI agents solve compared to running an agent in a shell subprocess? It prevents AI agents from corrupting .ipynb files and losing kernel state such as trained models, loaded dataframes, and imports when a subprocess exits. Instead of hand-editing notebook JSON and running code in a disposable subprocess, the agent operates a live Jupyter kernel through the IDE via a single MCP wrapper called execute_tool, letting state persist across cells and long jobs run without constant polling. _Developers debugging fragile AI-agent notebook workflows can track fixes like this through daily.dev._ ### Does the PyCharm Jupyter skill reduce AI agent costs for all models, including GPT-5.6 through Codex? No, the cost win is model- and task-dependent. It was clearest for Claude Opus 5 on long, stateful jobs, while running through the shell was cheaper on short tasks and for the Codex-based GPT-5.6 models Sol and Terra, which already use prompt caching efficiently, so for those the skill's value comes from workflow improvements rather than cost savings. _Developers comparing AI agent costs across models can watch for benchmark updates like this via daily.dev._ ## Similar posts on daily.dev - [We Stopped AI Agents From Installing Into the Wrong Python: Task Success Rates Jumped to 95%\+](https://daily.dev/posts/we-stopped-ai-agents-from-installing-into-the-wrong-python-task-success-rates-jumped-to-95--ntqtgf7vo) · JetBrains · 1 upvotes · 0 comments - [Promoting AI agents](https://daily.dev/posts/promoting-ai-agents-zppmug4cg) · David Heinemeier Hansson · 112 upvotes · 10 comments - [What's New in PyCharm 2026.2.1](https://daily.dev/posts/what-s-new-in-pycharm-2026-2-1-clujd95sb) · JetBrains · 0 upvotes · 0 comments - [Google Launches Colab CLI for Developers, Automation, and AI Agents](https://daily.dev/posts/google-launches-colab-cli-for-developers-automation-and-ai-agents-r8hst8g5q) · InfoQ · 2 upvotes · 1 comments --- Tags: [#python](https://daily.dev/tags/python), [#devtools](https://daily.dev/tags/devtools), [#ai-agents](https://daily.dev/tags/ai-agents), [#mcp](https://daily.dev/tags/mcp), [#jupyter](https://daily.dev/tags/jupyter) [View this post on daily.dev](https://daily.dev/posts/we-gave-ai-agents-a-live-jupyter-kernel-in-pycharm-hrqonqnd7) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"We Gave AI Agents a Live Jupyter Kernel in PyCharm","url":"https://daily.dev/posts/we-gave-ai-agents-a-live-jupyter-kernel-in-pycharm-hrqonqnd7","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/we-gave-ai-agents-a-live-jupyter-kernel-in-pycharm-hrqonqnd7"},"datePublished":"2026-08-12T13:23:37.877Z","dateModified":"2026-09-14T06:17:51.608Z","description":"PyCharm introduces a new Jupyter skill that lets AI agents operate a live Jupyter kernel directly through the IDE instead of running code in a throwaway shell...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9cd11721b06d723e008e0810c139a547?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9cd11721b06d723e008e0810c139a547?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"JetBrains","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"JetBrains","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/53ecf0c676f34b3896ee109609d91efa","url":"https://daily.dev/sources/jetbrains"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/we-gave-ai-agents-a-live-jupyter-kernel-in-pycharm-hrqonqnd7","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":5},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"python,devtools,ai-agents,mcp,jupyter","timeRequired":"PT6M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"JetBrains","item":"https://daily.dev/sources/jetbrains"},{"@type":"ListItem","position":3,"name":"We Gave AI Agents a Live Jupyter Kernel in PyCharm"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/we-gave-ai-agents-a-live-jupyter-kernel-in-pycharm-hrqonqnd7#faq","mainEntity":[{"@type":"Question","name":"How much cheaper is it for Claude Opus 5 to run machine learning tasks through a live Jupyter kernel versus a shell in PyCharm?","acceptedAnswer":{"@type":"Answer","text":"Running through the kernel cost 59.09 USD versus 67.06 USD through the shell, about 12% cheaper, across twelve MLGym benchmark tasks. The savings come from the kernel keeping the prompt cache warm: 98% of input tokens were cache reads compared to 82% for the shell, and cache reads cost only 1/12 of creating a fresh cache. With the contaminated Titanic task removed, the kernel was still 10% cheaper (56.34 USD versus 62.65 USD). Teams weighing agent cost tradeoffs in notebook workflows can follow tooling changes like this via daily.dev."}},{"@type":"Question","name":"What problem does PyCharm's new Jupyter skill for AI agents solve compared to running an agent in a shell subprocess?","acceptedAnswer":{"@type":"Answer","text":"It prevents AI agents from corrupting .ipynb files and losing kernel state such as trained models, loaded dataframes, and imports when a subprocess exits. Instead of hand-editing notebook JSON and running code in a disposable subprocess, the agent operates a live Jupyter kernel through the IDE via a single MCP wrapper called execute_tool, letting state persist across cells and long jobs run without constant polling. Developers debugging fragile AI-agent notebook workflows can track fixes like this through daily.dev."}},{"@type":"Question","name":"Does the PyCharm Jupyter skill reduce AI agent costs for all models, including GPT-5.6 through Codex?","acceptedAnswer":{"@type":"Answer","text":"No, the cost win is model- and task-dependent. It was clearest for Claude Opus 5 on long, stateful jobs, while running through the shell was cheaper on short tasks and for the Codex-based GPT-5.6 models Sol and Terra, which already use prompt caching efficiently, so for those the skill's value comes from workflow improvements rather than cost savings. Developers comparing AI agent costs across models can watch for benchmark updates like this via daily.dev."}}]} ```