Hey guys! I've been using Cortex.cpp (https://github.com/janhq/cortex.cpp) which is basically a backend that runs different models. It's a new project and it lacks API libraries to interact with using programming languages (you should make http requests directly and there's no layer of abstraction). Any way I am spending some time now developing a kotlin based library which provides easy access to it. I'd appreciate support for the library and contributors with better knowledge of Ai and Cortex.cpp (since I'm a physics student I don't necessary know everything about it). Here's my project: https://github.com/thisismeamir/Kortex.cpp

Hey everyone! I love Kotlin. I built my whole company on top of it, but as the LLM world had gotten into full swing, it is disheartening to see that my company is pressured more and more into relying on python. I personally feel like the ergonomics and readability of Kotlin are light years ahead of python, but the market is not looking good for kotlin in this quickly evolving space. I fear that Kotlin will lose relevance if Python eats it’s lunch

i would love to see JB take AI more seriously

Hi everyone! Working on a personal research project on Kotlin's potential for developing AI applications. It would be great if some of you who tried to work on such project could take maybe 10 minutes to fill out this survey. SURVEY LINK HERE Thank you all!

I love Kotlin, and despite lack of tooling, I believe it's actually one of the most potent languages when it comes to AI based development, where my claudine AI agent is writing majority of the code. Kotlin code generated by Anthropic's models is usually flawless, allowing me to generate complete Kotlin based products in the span of minutes. The fact that Kotlin is a compiled language, allows to detect problems very early, and lets LLM correct them. When working in typical TDD workflow: • I load the whole project source code into the token window. • Then I ask AI to generate a suite of comprehensive test cases for the requested feature, under my supervision, and Kotlin based DSLs are unbelievably practical in this case. • Then we proceed with implementation, which might involve thousands lines of code. With each iteration AI agent can execute the build and verify the outcome in the feedback loop. Sometimes it's going in circles. Sometimes I need to intervene. Sometimes it is forgetting some test case and removing it for no reason from the code. Working with version control during iterations is crucial. However, overall, it is order of magnitude more efficient workflow than me implementing a complete product or library myself. I wonder how are you using AI in your Kotlin projects?

Hottest OpenAI-compatible Client Library for Kotlin? 🔥 K I was using

aallam/openai-kotlin

here but its maintenance has dropped kodee sad - trying to decide whether to jump ship or Fork it 🛠️

Has anybody here been granted access to the Junie EAP? Assuming you are able to discuss your experience so far, what are your initial impressions? (Particularly from the perspective of Kotlin development).

I seem to constantly run into an issue when using Jetbrains AI, or testing out Junie, that the underlying models downgrade various kotlin libraries or other various dependencies simply because they don't know new versions exist. I wonder if there were a way to get these products to have them check various repositories to validate the versions before sending a response back. For example, I used the KMP project generator to stub out a new project for me. The first thing Junie did was to downgrade it from kotlin 2.1.0 to 1.9

This is due to knowledge cutoff. Depending if Claude Sonnet 3.5 or 3.7 is in use, it will be April or October 2024 respectively. To certain extent it can be fixed with proper information in system prompt or token window: 1. which time is now (not sure if Junie is adding that, my Claudine is): https://github.com/xemantic/claudine/blob/main/src/commonMain/kotlin/Environment.kt 2. loading file containing current versions (e.g. libs.versions.toml) into the token window before doing any refactoring (preferred anyway - to load all the source code into the token window first) 3. providing either the newest version specs in system prompt, or adding a tool for the model to check them. I would say that most of the time 1 + 2 works the best. When I work with my Claludine, I instruct it to load the whole source code of the project into the token window first (and hopefully cache it), and only then continue with changes. I am pretty sure that Junie can be prompted to do the same, however, if it is loading a single file instead of multiple files in a single step via a tool, it will be still quite inefficient.

Hey everyone I'm building Build That Idea – a platform that lets anyone build GPT wrappers in 60 seconds How it works: - Define your GPT - Choose a base LLM (OpenAI, Claude, DeepSeek, etc.) - Upload knowledge base - Set pricing and start making money Right now, it's just a waitlist, but I'd love feedback on the idea. Would you use something like this? What features would be useful?

alphabet white question Any tools aimed at running an LLM processing prompt over multiple files in a more formal fashion? I've been having a frustrating time trying to get Agentic AI to perform 'simple' batch-processing tasks. Example: Getting an AI to take on the job of cleaning up unwanted comments, improve a few simple code cleanliness issues that are just a cut above what IDE reformatting can do. Each AI solution I've tried so far: • Warp AI (Best so far despite being aimed at command line, required much manual intervention to approve changes, eventually gave up) • Junie ('Plans' to do all files but gives up easily after a couple no matter how much you implore it to do all) • Block Goose (Hopeless, dumb as a sack of bricks at nearly every task asked of it) ...the common theme is that many current agents are really poor at completing batches of changes even when it's an ostensibly simple loop over a few files. Feeling like building something, but this must be a solved problem(?)

https://github.com/openai/openai-java fun fact, the OpenAI SDK for Java is made in Kotlin

Hi, Anyone interested in helping me with a Kotlin KMP implementation of the new Google A2A protocol (https://github.com/google/A2A)? Just started https://github.com/a2a-4k/a2a-4k so any help or feedback would be appreciated.

Anyone else here building MCPs and have found a good debugging workflow? The inspector tool is alright and I know it's early days, just wanting faster feedback that is easier to automate and eventually have tests for. Coming to the conclusion that I'll just have to build the tools I want to use

How could I choose a different Ollama model in KOOG? For example "qwen3:14b"

Hey, at Kotlinconf Duncan and Dmitry had this auto generated commit messages with Junie is this already available in public Junie version or ?

Is there a KMP library for running LLM inference (locally)?

If I create an agent and call multiple times to

agent.runAndGetResult(it)

it returns always the first response. Is it a bug?

AI Agent versus MCP ——— An AI agent is a software program that can interact with its environment, gather data, and use that data to achieve predetermined goals. AI agents can choose the best actions to perform to meet those goals. Key characteristics of AI agents are as follows: 1. An agent can perform autonomous actions without constant human intervention. Also, they can have a human in the loop to maintain control. 2. Agents have a memory to store individual preferences and allow for personalization. It can also store knowledge. An LLM can undertake information processing and decision-making functions. 3. Agents must be able to perceive and process the information available from their environment. ——— Model Context Protocol (MCP) is a new system introduced by Anthropic to make AI models more powerful. It is an open standard that allows AI models (like Claude) to connect to databases, APIs, file systems, and other tools without needing custom code for each new integration. MCP follows a client-server model with 3 key components: 1. Host: AI applications like Claude 2. MCP Client: Component inside an AI model (like Claude) that allows it to communicate with MCP servers 3. MCP Server: Middleman that connects an AI model to an external system Over to you: Have you used AI Agents or MCP? ——— Source: ByteByteGo

Is there a specific place to ask questions about Junie?

Hi! Koog

0.2.0

has just been released with all the latest updates and fixes. Thank you everyone for your contributions! Full changelog https://github.com/JetBrains/koog/releases/tag/0.2.0 Some notable changes: • media types support (image, audio, document) • token counting support • Lots of Ollama-related bug fixes + dynamic model discovery support for Ollama • Fixed structured data bugs with property serialization • Fixed missing support for Double and List in the annotation-based tools API • … and many other bug-fixes and improvements (see the full CHANGELOG) Slack Conversation

I start to believe that Junie (or Clause as the LLM behind) can't to web searches. It fails to add https://github.com/whyoleg/cryptography-kotlin to my project. It uses a wrong / non-existing artifact & version numbers. I asked it to read the GitHub documentation, but it seemingly can't.

🚀 New release of nirmato-ollama v0.2.0 is out! Ollama API client for Kotlin multiplatform. github - https://github.com/nirmato/nirmato-ollama

Hey guys! I tried Junie for Android dev and it’s quite impressive. Do you know if it is limited/well trained mostly for Kotlin/Android/KMP or will it also be as good in python coding? I wonder if some of you tried it out elsewhere.

Hey, do you know if JetBrains AI can modify the files inside the AS?

i just learned about https://github.com/embabel/embabel-agent today. has anyone played with it?

🎉 Exciting news! My at KotlinConf is now publicly available! Koog is a powerful framework for building AI agents in Kotlin, making it easy to integrate AI into any type of application — whether it’s mobile, backend, or web. If you’re curious about bringing AI into your Kotlin projects, I’d love for you to check it out. Let’s build smarter products together — let’s cook some AI! 🍳🤖

does anyone here know what JB is using to build Junie?

Is there a way in Junie to save the entire transcript of a task? Is it stored on disk?