Stanislav
@khromov.se
🇸🇪 Web developer and obscure trivia enjoyer
🌱 Find me at https://stanislav.garden/
🔵 Starting over on Bluesky
🌱 Find me at https://stanislav.garden/
🔵 Starting over on Bluesky
🔥 Before and after SvelteKit Remote Functions!
(by Snaæk in the official Svelte Discord - join us at discord.com/invite/svelte )
#svelte #sveltekit #javascript
(by Snaæk in the official Svelte Discord - join us at discord.com/invite/svelte )
#svelte #sveltekit #javascript
October 4, 2025 at 8:45 AM
🔥 Before and after SvelteKit Remote Functions!
(by Snaæk in the official Svelte Discord - join us at discord.com/invite/svelte )
#svelte #sveltekit #javascript
(by Snaæk in the official Svelte Discord - join us at discord.com/invite/svelte )
#svelte #sveltekit #javascript
It's not every day you see the Swedish central bank's website at the top of Hacker News! 🇸🇪 🫎
October 3, 2025 at 11:14 PM
It's not every day you see the Swedish central bank's website at the top of Hacker News! 🇸🇪 🫎
👀 Open weights LLMs are catching up to commercial models in their Svelte 5 skills! Kimi K2, DeepSeek v3.2 and GLM-4.6 are in 4th, 5th and 6th place, far surpassing closed models by OpenAI and Google! #ai #svelte
All results at khromov.se/svelte-bench
All results at khromov.se/svelte-bench
October 1, 2025 at 8:45 AM
👀 Open weights LLMs are catching up to commercial models in their Svelte 5 skills! Kimi K2, DeepSeek v3.2 and GLM-4.6 are in 4th, 5th and 6th place, far surpassing closed models by OpenAI and Google! #ai #svelte
All results at khromov.se/svelte-bench
All results at khromov.se/svelte-bench
🎞 Putting the finishing touches on my next video! Is Wuchale the perfect library to translate your Svelte and SvelteKit apps? Find out tomorrow! #svelte #javascript
September 15, 2025 at 8:45 AM
🎞 Putting the finishing touches on my next video! Is Wuchale the perfect library to translate your Svelte and SvelteKit apps? Find out tomorrow! #svelte #javascript
📝 One of the challenges with MCP is for the LLM to know when and how to use MCP tools. To solve that, the Svelte MCP server now has a recommended system prompt. Add it to to your CLAUDE.md, AGENTS.md files or wherever your coding agents prefers! #ai #svelte
svelte-llm.stanislav.garden
svelte-llm.stanislav.garden
September 9, 2025 at 8:45 AM
📝 One of the challenges with MCP is for the LLM to know when and how to use MCP tools. To solve that, the Svelte MCP server now has a recommended system prompt. Add it to to your CLAUDE.md, AGENTS.md files or wherever your coding agents prefers! #ai #svelte
svelte-llm.stanislav.garden
svelte-llm.stanislav.garden
Reddit decided to randomly flip my avatar using CSS. Now my face points to the right. Why? Don't know! 🤷
September 4, 2025 at 8:45 AM
Reddit decided to randomly flip my avatar using CSS. Now my face points to the right. Why? Don't know! 🤷
We complain a lot about Google search lately, but sometimes they still deliver
August 30, 2025 at 3:38 PM
We complain a lot about Google search lately, but sometimes they still deliver
Anthropic is changing their privacy policy and now wants to train on our data. While I didn't think the "we never train on inputs" stance was tenable, I do wish they would give us something in return for our data (eg. free tiers/discounts like Google does with AI Studio).
#ai #claude
#ai #claude
August 29, 2025 at 8:45 AM
Seems like Anthropic changed how MCP support works in Claude Desktop recently. If you've been using the Svelte MCP with Claude Desktop, you need to update the URL. The instructions on the site are updated! svelte-llm.stanislav.garden #svelte
August 21, 2025 at 2:27 PM
Seems like Anthropic changed how MCP support works in Claude Desktop recently. If you've been using the Svelte MCP with Claude Desktop, you need to update the URL. The instructions on the site are updated! svelte-llm.stanislav.garden #svelte
👀 It turns out LLMs like Claude are amazing at adhering to coding standards when you can put your whole codebase in context. The context-coder MCP improves generation for small-to-medium-sized projects and lets you prompt in netural language. www.npmjs.com/package/cont... #mcp #ai
August 14, 2025 at 8:45 AM
👀 It turns out LLMs like Claude are amazing at adhering to coding standards when you can put your whole codebase in context. The context-coder MCP improves generation for small-to-medium-sized projects and lets you prompt in netural language. www.npmjs.com/package/cont... #mcp #ai
The DigitalOcean AI platform could have been an alternative to OpenRouter, but instead these are the models they offer their users. 🤷 Really hope they start taking AI seriously because the current offering is not good.
August 12, 2025 at 12:21 PM
The DigitalOcean AI platform could have been an alternative to OpenRouter, but instead these are the models they offer their users. 🤷 Really hope they start taking AI seriously because the current offering is not good.
✨ New video coming out tomorrow! We'll be diving (delving? 😄 ) into the Devin agentic AI platform to see what it can do!
August 11, 2025 at 9:08 PM
✨ New video coming out tomorrow! We'll be diving (delving? 😄 ) into the Devin agentic AI platform to see what it can do!
🎨 I am a Tailwind skeptic but there's no denying that it enables you to fit much more code into an LLM context window. An example of a project with vanilla CSS versus Tailwind = ~18% token reduction. PS. If you want to try the project you can find it at ai.stanislav.garden
August 10, 2025 at 8:45 AM
🎨 I am a Tailwind skeptic but there's no denying that it enables you to fit much more code into an LLM context window. An example of a project with vanilla CSS versus Tailwind = ~18% token reduction. PS. If you want to try the project you can find it at ai.stanislav.garden
Playdate in shambles right now. 😅
August 9, 2025 at 8:45 AM
Playdate in shambles right now. 😅
📈 Here are the SvelteBench results for gpt-5. Overall it places in 12th spot. However, it's a huge improvement over the other GPT models (last image), and seems like a decent choice for Svelte 5. Mini and Nano do not inherit the good scores. 😿 khromov.se/svelte-bench #svelte
August 7, 2025 at 10:29 PM
📈 Here are the SvelteBench results for gpt-5. Overall it places in 12th spot. However, it's a huge improvement over the other GPT models (last image), and seems like a decent choice for Svelte 5. Mini and Nano do not inherit the good scores. 😿 khromov.se/svelte-bench #svelte
☔️ A lesser talked about issue on OpenRouter is third party providers that serve quantized models. This causes worse model performance unbeknownst to the user. For example even though providers claim to serve GLM-4.5 at fp8, performance scores are worse on OpenRouter compared to the official Zai API
August 7, 2025 at 9:55 AM
☔️ A lesser talked about issue on OpenRouter is third party providers that serve quantized models. This causes worse model performance unbeknownst to the user. For example even though providers claim to serve GLM-4.5 at fp8, performance scores are worse on OpenRouter compared to the official Zai API
💬 There has been some discussion on who is making the mysterious horizon-alpha/beta model on OpenRouter. People say it's OpenAI, but based on the SvelteBench benchmark scores (where it does very well) I think it's more likely that it's another company/group.
August 6, 2025 at 8:45 AM
💬 There has been some discussion on who is making the mysterious horizon-alpha/beta model on OpenRouter. People say it's OpenAI, but based on the SvelteBench benchmark scores (where it does very well) I think it's more likely that it's another company/group.
🍩 Finally a useful use case for AI!
www.reddit.com/r/StableDiff...
www.reddit.com/r/StableDiff...
August 4, 2025 at 11:06 AM
🍩 Finally a useful use case for AI!
www.reddit.com/r/StableDiff...
www.reddit.com/r/StableDiff...
One of the annoying antipatterns in AI tooling that resells other providers frontier models, is that they introduce friction that prevent you from using the models too much due to cost. Like in this case with Copilot. That's why I prefer provider-made tooling like Claude Code.
August 1, 2025 at 8:45 AM
One of the annoying antipatterns in AI tooling that resells other providers frontier models, is that they introduce friction that prevent you from using the models too much due to cost. Like in this case with Copilot. That's why I prefer provider-made tooling like Claude Code.
What does this "unsaved changes with extra square" icon do and how can I get rid of it? 🫠
July 31, 2025 at 8:45 AM
What does this "unsaved changes with extra square" icon do and how can I get rid of it? 🫠
😸 When I have a programming problem I always make sure to rubber duck with my cat, they're great helpers!
July 30, 2025 at 8:45 AM
😸 When I have a programming problem I always make sure to rubber duck with my cat, they're great helpers!