The best Hacker News stories from Show from the past day

from

All Show

from the past*

day week month

with these search terms:

Create follow Go back

Latest posts:

Show HN: Build your own Bracket City puzzle

Hi HN — Bracket City is the word puzzle game I made earlier this year and (in part thanks to this community, see <a href="https://news.ycombinator.com/item?id=43622719">https://news.ycombinator.com/item?id=43622719</a>) managed to license to the Atlantic in April.<p>The game has been growing a lot and I wanted to share the latest: a tool that lets anyone make a Bracket City puzzle — specifically a “Bracket Suburb”!<p>I made this tool to help me construct puzzles, and I’ve been using it every day for months.<p>After the Atlantic launch, I started to get the occasional inquiry about whether there was a way to make your own puzzle. One guy wanted to make a Bracket City puzzle part of a puzzle hunt he made to propose to his girlfriend (he did it!), and that convinced me it would be fun to make something publicly available.<p>I got the Atlantic on board with the idea, and we are launching it today with an "example" custom puzzle: a Halloween/horror-themed puzzle by my pal Wyna Liu of NYT Connections fame.<p><a href="https://suburbs.bracket.city/wyna" rel="nofollow">https://suburbs.bracket.city/wyna</a><p>And we've got few other fun "celeb" puzzles lined up for later this year.<p>The thought is that folks can use the builder to make custom puzzles for birthday wishes/event invites/insults/proposals/break ups in addition to “normal” Bracket City puzzles.<p>I'm also hoping to learn more about the potential of the format – crossword puzzles have benefited so much from the creativity of constructors – I'm hoping bracket puzzles do the same.<p>The good news is that it’s way easier to construct a bracket puzzle than a crossword. Once you try it, you’ll see why: you have many more degrees of freedom. In a crossword, each added word increases the level of constraint exponentially — every new entry sharply reduces the remaining options for completing the grid. Bracket puzzles are the opposite: as you add clues, you expand the available fodder for new ones.<p>Anyway, I would love any/all feedback and to try puzzles created by folks here. I’m hoping we will figure out a way to highlight the best community puzzles on the Atlantic soon!<p>PS and please keep playing the main game / sending me feedback / denouncing me on the subreddit

Show HN: Why write code if the LLM can just do the thing? (web app experiment)

I spent a few hours last weekend testing whether AI can replace code by executing directly. Built a contact manager where every HTTP request goes to an LLM with three tools: database (SQLite), webResponse (HTML/JSON/JS), and updateMemory (feedback). No routes, no controllers, no business logic. The AI designs schemas on first request, generates UIs from paths alone, and evolves based on natural language feedback. It works—forms submit, data persists, APIs return JSON—but it's catastrophically slow (30-60s per request), absurdly expensive ($0.05/request), and has zero UI consistency between requests. The capability exists; performance is the problem. When inference gets 10x faster, maybe the question shifts from "how do we generate better code?" to "why generate code at all?"

Show HN: Why write code if the LLM can just do the thing? (web app experiment)

I spent a few hours last weekend testing whether AI can replace code by executing directly. Built a contact manager where every HTTP request goes to an LLM with three tools: database (SQLite), webResponse (HTML/JSON/JS), and updateMemory (feedback). No routes, no controllers, no business logic. The AI designs schemas on first request, generates UIs from paths alone, and evolves based on natural language feedback. It works—forms submit, data persists, APIs return JSON—but it's catastrophically slow (30-60s per request), absurdly expensive ($0.05/request), and has zero UI consistency between requests. The capability exists; performance is the problem. When inference gets 10x faster, maybe the question shifts from "how do we generate better code?" to "why generate code at all?"

Show HN: Why write code if the LLM can just do the thing? (web app experiment)

I spent a few hours last weekend testing whether AI can replace code by executing directly. Built a contact manager where every HTTP request goes to an LLM with three tools: database (SQLite), webResponse (HTML/JSON/JS), and updateMemory (feedback). No routes, no controllers, no business logic. The AI designs schemas on first request, generates UIs from paths alone, and evolves based on natural language feedback. It works—forms submit, data persists, APIs return JSON—but it's catastrophically slow (30-60s per request), absurdly expensive ($0.05/request), and has zero UI consistency between requests. The capability exists; performance is the problem. When inference gets 10x faster, maybe the question shifts from "how do we generate better code?" to "why generate code at all?"

Show HN: Strange Attractors

I went down the rabbit hole on a side project and ended up building this: Strange Attractors(<a href="https://blog.shashanktomar.com/posts/strange-attractors" rel="nofollow">https://blog.shashanktomar.com/posts/strange-attractors</a>). It’s built with three.js.<p>Working on it reminded me of the little "maths for fun" exercises I used to do while learning programming in early days. Just trying things out, getting fascinated and geeky, and being surprised by the results. I spent way too much time on this, but it was extreme fun.<p>My favorite part: someone pointed me to the Simone Attractor on Threads. It is a 2D attractor and I asked GPT to extrapolate it to 3D, not sure if it’s mathematically correct, but it’s the coolest by far. I have left all the params configurable, so give it a try. I called it Simone (Maybe).<p>If you like math-art experiments, check it out. Would love feedback, especially from folks who know more about the math side.

Show HN: Strange Attractors

I went down the rabbit hole on a side project and ended up building this: Strange Attractors(<a href="https://blog.shashanktomar.com/posts/strange-attractors" rel="nofollow">https://blog.shashanktomar.com/posts/strange-attractors</a>). It’s built with three.js.<p>Working on it reminded me of the little "maths for fun" exercises I used to do while learning programming in early days. Just trying things out, getting fascinated and geeky, and being surprised by the results. I spent way too much time on this, but it was extreme fun.<p>My favorite part: someone pointed me to the Simone Attractor on Threads. It is a 2D attractor and I asked GPT to extrapolate it to 3D, not sure if it’s mathematically correct, but it’s the coolest by far. I have left all the params configurable, so give it a try. I called it Simone (Maybe).<p>If you like math-art experiments, check it out. Would love feedback, especially from folks who know more about the math side.

Show HN: Strange Attractors

I went down the rabbit hole on a side project and ended up building this: Strange Attractors(<a href="https://blog.shashanktomar.com/posts/strange-attractors" rel="nofollow">https://blog.shashanktomar.com/posts/strange-attractors</a>). It’s built with three.js.<p>Working on it reminded me of the little "maths for fun" exercises I used to do while learning programming in early days. Just trying things out, getting fascinated and geeky, and being surprised by the results. I spent way too much time on this, but it was extreme fun.<p>My favorite part: someone pointed me to the Simone Attractor on Threads. It is a 2D attractor and I asked GPT to extrapolate it to 3D, not sure if it’s mathematically correct, but it’s the coolest by far. I have left all the params configurable, so give it a try. I called it Simone (Maybe).<p>If you like math-art experiments, check it out. Would love feedback, especially from folks who know more about the math side.

Show HN: Strange Attractors

I went down the rabbit hole on a side project and ended up building this: Strange Attractors(<a href="https://blog.shashanktomar.com/posts/strange-attractors" rel="nofollow">https://blog.shashanktomar.com/posts/strange-attractors</a>). It’s built with three.js.<p>Working on it reminded me of the little "maths for fun" exercises I used to do while learning programming in early days. Just trying things out, getting fascinated and geeky, and being surprised by the results. I spent way too much time on this, but it was extreme fun.<p>My favorite part: someone pointed me to the Simone Attractor on Threads. It is a 2D attractor and I asked GPT to extrapolate it to 3D, not sure if it’s mathematically correct, but it’s the coolest by far. I have left all the params configurable, so give it a try. I called it Simone (Maybe).<p>If you like math-art experiments, check it out. Would love feedback, especially from folks who know more about the math side.

Show HN: Front End Fuzzy and Substring and Prefix Search

Hey everyone, I have updated my fuzzy search library for the frontend. It now supports substring and prefix search, on top of fuzzy matching. It's fast, accurate, multilingual and has zero dependencies.<p>GitHub: <a href="https://github.com/m31coding/fuzzy-search" rel="nofollow">https://github.com/m31coding/fuzzy-search</a> Live demo: <a href="https://www.m31coding.com/fuzzy-search-demo.html" rel="nofollow">https://www.m31coding.com/fuzzy-search-demo.html</a><p>I would love to hear your feedback and any suggestions you may have for improving the library.<p>Happy coding!

Show HN: Front End Fuzzy and Substring and Prefix Search

Hey everyone, I have updated my fuzzy search library for the frontend. It now supports substring and prefix search, on top of fuzzy matching. It's fast, accurate, multilingual and has zero dependencies.<p>GitHub: <a href="https://github.com/m31coding/fuzzy-search" rel="nofollow">https://github.com/m31coding/fuzzy-search</a> Live demo: <a href="https://www.m31coding.com/fuzzy-search-demo.html" rel="nofollow">https://www.m31coding.com/fuzzy-search-demo.html</a><p>I would love to hear your feedback and any suggestions you may have for improving the library.<p>Happy coding!

Show HN: A fast, dependency-free traceroute implementation in pure C

Show HN: ekoAcademic – Convert ArXiv papers to interactive podcasts

Show HN: Pipelex – Declarative language for repeatable AI workflows

We’re Robin, Louis, and Thomas. Pipelex is a DSL and a Python runtime for repeatable AI workflows. Think Dockerfile/SQL for multi-step LLM pipelines: you declare steps and interfaces; any model/provider can fill them.<p>Why this instead of yet another workflow builder?<p>- Declarative, not glue code: you state what to do; the runtime figures out how. - Agent-first: each step carries natural-language context (purpose, inputs/outputs with meaning) so LLMs can follow, audit, and optimize. Our MCP server enables agents to run pipelines but also to build new pipelines on demand. - Open standard under MIT: language spec, runtime, API server, editor extensions, MCP server, n8n node. - Composable: pipes can call other pipes, created by you or shared in the community.<p>Why a domain-specific language?<p>- We need context, meaning and nuances preserved in a structured syntax that both humans and LLMs can understand - We need determinism, control, and reproducibility that pure prompts can't deliver - Bonus: editors, diffs, semantic coloring, easy sharing, search & replace, version control, linters…<p>How we got there:<p>Initially, we just wanted to solve every use-case with LLMs but kept rebuilding the same agentic patterns across different projects. So we challenged ourselves to keep the code generic and separate from use-case specifics, which meant modeling workflows from the relevant knowledge and know-how.<p>Unlike existing code/no-code frameworks for AI workflows, our abstraction layer doesn't wrap APIs, it transcribes business logic into a structured, unambiguous script executable by software and AI. Hence the "declarative" aspect: the script says what should be done, not how to do it. It's like a Dockerfile or SQL for AI workflows.<p>Additionally, we wanted the language to be LLM-friendly. Classic programming languages hide logic and context in variable names, functions, and comments: all invisible to the interpreter. In Pipelex, these elements are explicitly stated in natural language, giving AI full visibility: it's all logic and context, with minimal syntax.<p>Then, we didn't want to write Pipelex scripts ourselves so we dogfooded: we built a Pipelex workflow that writes Pipelex workflows. It's in the MCP and CLI: "pipelex build pipe '…'" runs a multi-step, structured generation flow that produces a validated workflow ready to execute with "pipelex run". Then you can iterate on it yourself or with any coding agent.<p>What’s included: Python library, FastAPI and Docker, MCP server, n8n node, VS Code extension.<p>What we’d like from you<p>1. Build a workflow: did the language work for you or against you? 2. Agent/MCP workflows and n8n node usability. 3. Suggest new kinds of pipes and other AI models we could integrate 4. Looking for OSS contributors to the core library but also to share pipes with the community<p>Known limitations<p>- Connectors: Pipelex doesn’t integrate with “your apps”, we focus on the cognitive steps, and you can integrate through code/API or using MCP or n8n - Visualization: we need to generate flow-charts - The pipe builder is still buggy - Run it yourself: we don’t yet provide a hosted Pipelex API, it’s in the works - Cost-tracking: we only track LLM costs, not image generation or OCR costs yet - Caching and reasoning options: not supported yet<p>Links<p>- GitHub: <a href="https://github.com/Pipelex/pipelex" rel="nofollow">https://github.com/Pipelex/pipelex</a> - Cookbook: <a href="https://github.com/Pipelex/pipelex-cookbook" rel="nofollow">https://github.com/Pipelex/pipelex-cookbook</a> - Starter: <a href="https://github.com/Pipelex/pipelex-starter" rel="nofollow">https://github.com/Pipelex/pipelex-starter</a> - VS Code extension: <a href="https://github.com/Pipelex/vscode-pipelex" rel="nofollow">https://github.com/Pipelex/vscode-pipelex</a> - Docs: [<a href="https://docs.pipelex.com" rel="nofollow">https://docs.pipelex.com</a>](<a href="https://docs.pipelex.com/" rel="nofollow">https://docs.pipelex.com/</a>) - Demo video (2 min): <a href="https://youtu.be/dBigQa8M8pQ" rel="nofollow">https://youtu.be/dBigQa8M8pQ</a> - Discord for support and sharing: <a href="https://go.pipelex.com/discord" rel="nofollow">https://go.pipelex.com/discord</a><p>Thanks for reading. If you try Pipelex, tell us exactly where it hurts, that’s the most valuable feedback we can get.

Show HN: Pipelex – Declarative language for repeatable AI workflows

We’re Robin, Louis, and Thomas. Pipelex is a DSL and a Python runtime for repeatable AI workflows. Think Dockerfile/SQL for multi-step LLM pipelines: you declare steps and interfaces; any model/provider can fill them.<p>Why this instead of yet another workflow builder?<p>- Declarative, not glue code: you state what to do; the runtime figures out how. - Agent-first: each step carries natural-language context (purpose, inputs/outputs with meaning) so LLMs can follow, audit, and optimize. Our MCP server enables agents to run pipelines but also to build new pipelines on demand. - Open standard under MIT: language spec, runtime, API server, editor extensions, MCP server, n8n node. - Composable: pipes can call other pipes, created by you or shared in the community.<p>Why a domain-specific language?<p>- We need context, meaning and nuances preserved in a structured syntax that both humans and LLMs can understand - We need determinism, control, and reproducibility that pure prompts can't deliver - Bonus: editors, diffs, semantic coloring, easy sharing, search & replace, version control, linters…<p>How we got there:<p>Initially, we just wanted to solve every use-case with LLMs but kept rebuilding the same agentic patterns across different projects. So we challenged ourselves to keep the code generic and separate from use-case specifics, which meant modeling workflows from the relevant knowledge and know-how.<p>Unlike existing code/no-code frameworks for AI workflows, our abstraction layer doesn't wrap APIs, it transcribes business logic into a structured, unambiguous script executable by software and AI. Hence the "declarative" aspect: the script says what should be done, not how to do it. It's like a Dockerfile or SQL for AI workflows.<p>Additionally, we wanted the language to be LLM-friendly. Classic programming languages hide logic and context in variable names, functions, and comments: all invisible to the interpreter. In Pipelex, these elements are explicitly stated in natural language, giving AI full visibility: it's all logic and context, with minimal syntax.<p>Then, we didn't want to write Pipelex scripts ourselves so we dogfooded: we built a Pipelex workflow that writes Pipelex workflows. It's in the MCP and CLI: "pipelex build pipe '…'" runs a multi-step, structured generation flow that produces a validated workflow ready to execute with "pipelex run". Then you can iterate on it yourself or with any coding agent.<p>What’s included: Python library, FastAPI and Docker, MCP server, n8n node, VS Code extension.<p>What we’d like from you<p>1. Build a workflow: did the language work for you or against you? 2. Agent/MCP workflows and n8n node usability. 3. Suggest new kinds of pipes and other AI models we could integrate 4. Looking for OSS contributors to the core library but also to share pipes with the community<p>Known limitations<p>- Connectors: Pipelex doesn’t integrate with “your apps”, we focus on the cognitive steps, and you can integrate through code/API or using MCP or n8n - Visualization: we need to generate flow-charts - The pipe builder is still buggy - Run it yourself: we don’t yet provide a hosted Pipelex API, it’s in the works - Cost-tracking: we only track LLM costs, not image generation or OCR costs yet - Caching and reasoning options: not supported yet<p>Links<p>- GitHub: <a href="https://github.com/Pipelex/pipelex" rel="nofollow">https://github.com/Pipelex/pipelex</a> - Cookbook: <a href="https://github.com/Pipelex/pipelex-cookbook" rel="nofollow">https://github.com/Pipelex/pipelex-cookbook</a> - Starter: <a href="https://github.com/Pipelex/pipelex-starter" rel="nofollow">https://github.com/Pipelex/pipelex-starter</a> - VS Code extension: <a href="https://github.com/Pipelex/vscode-pipelex" rel="nofollow">https://github.com/Pipelex/vscode-pipelex</a> - Docs: [<a href="https://docs.pipelex.com" rel="nofollow">https://docs.pipelex.com</a>](<a href="https://docs.pipelex.com/" rel="nofollow">https://docs.pipelex.com/</a>) - Demo video (2 min): <a href="https://youtu.be/dBigQa8M8pQ" rel="nofollow">https://youtu.be/dBigQa8M8pQ</a> - Discord for support and sharing: <a href="https://go.pipelex.com/discord" rel="nofollow">https://go.pipelex.com/discord</a><p>Thanks for reading. If you try Pipelex, tell us exactly where it hurts, that’s the most valuable feedback we can get.

Show HN: Pipelex – Declarative language for repeatable AI workflows

We’re Robin, Louis, and Thomas. Pipelex is a DSL and a Python runtime for repeatable AI workflows. Think Dockerfile/SQL for multi-step LLM pipelines: you declare steps and interfaces; any model/provider can fill them.<p>Why this instead of yet another workflow builder?<p>- Declarative, not glue code: you state what to do; the runtime figures out how. - Agent-first: each step carries natural-language context (purpose, inputs/outputs with meaning) so LLMs can follow, audit, and optimize. Our MCP server enables agents to run pipelines but also to build new pipelines on demand. - Open standard under MIT: language spec, runtime, API server, editor extensions, MCP server, n8n node. - Composable: pipes can call other pipes, created by you or shared in the community.<p>Why a domain-specific language?<p>- We need context, meaning and nuances preserved in a structured syntax that both humans and LLMs can understand - We need determinism, control, and reproducibility that pure prompts can't deliver - Bonus: editors, diffs, semantic coloring, easy sharing, search & replace, version control, linters…<p>How we got there:<p>Initially, we just wanted to solve every use-case with LLMs but kept rebuilding the same agentic patterns across different projects. So we challenged ourselves to keep the code generic and separate from use-case specifics, which meant modeling workflows from the relevant knowledge and know-how.<p>Unlike existing code/no-code frameworks for AI workflows, our abstraction layer doesn't wrap APIs, it transcribes business logic into a structured, unambiguous script executable by software and AI. Hence the "declarative" aspect: the script says what should be done, not how to do it. It's like a Dockerfile or SQL for AI workflows.<p>Additionally, we wanted the language to be LLM-friendly. Classic programming languages hide logic and context in variable names, functions, and comments: all invisible to the interpreter. In Pipelex, these elements are explicitly stated in natural language, giving AI full visibility: it's all logic and context, with minimal syntax.<p>Then, we didn't want to write Pipelex scripts ourselves so we dogfooded: we built a Pipelex workflow that writes Pipelex workflows. It's in the MCP and CLI: "pipelex build pipe '…'" runs a multi-step, structured generation flow that produces a validated workflow ready to execute with "pipelex run". Then you can iterate on it yourself or with any coding agent.<p>What’s included: Python library, FastAPI and Docker, MCP server, n8n node, VS Code extension.<p>What we’d like from you<p>1. Build a workflow: did the language work for you or against you? 2. Agent/MCP workflows and n8n node usability. 3. Suggest new kinds of pipes and other AI models we could integrate 4. Looking for OSS contributors to the core library but also to share pipes with the community<p>Known limitations<p>- Connectors: Pipelex doesn’t integrate with “your apps”, we focus on the cognitive steps, and you can integrate through code/API or using MCP or n8n - Visualization: we need to generate flow-charts - The pipe builder is still buggy - Run it yourself: we don’t yet provide a hosted Pipelex API, it’s in the works - Cost-tracking: we only track LLM costs, not image generation or OCR costs yet - Caching and reasoning options: not supported yet<p>Links<p>- GitHub: <a href="https://github.com/Pipelex/pipelex" rel="nofollow">https://github.com/Pipelex/pipelex</a> - Cookbook: <a href="https://github.com/Pipelex/pipelex-cookbook" rel="nofollow">https://github.com/Pipelex/pipelex-cookbook</a> - Starter: <a href="https://github.com/Pipelex/pipelex-starter" rel="nofollow">https://github.com/Pipelex/pipelex-starter</a> - VS Code extension: <a href="https://github.com/Pipelex/vscode-pipelex" rel="nofollow">https://github.com/Pipelex/vscode-pipelex</a> - Docs: [<a href="https://docs.pipelex.com" rel="nofollow">https://docs.pipelex.com</a>](<a href="https://docs.pipelex.com/" rel="nofollow">https://docs.pipelex.com/</a>) - Demo video (2 min): <a href="https://youtu.be/dBigQa8M8pQ" rel="nofollow">https://youtu.be/dBigQa8M8pQ</a> - Discord for support and sharing: <a href="https://go.pipelex.com/discord" rel="nofollow">https://go.pipelex.com/discord</a><p>Thanks for reading. If you try Pipelex, tell us exactly where it hurts, that’s the most valuable feedback we can get.

Show HN: Pipelex – Declarative language for repeatable AI workflows

We’re Robin, Louis, and Thomas. Pipelex is a DSL and a Python runtime for repeatable AI workflows. Think Dockerfile/SQL for multi-step LLM pipelines: you declare steps and interfaces; any model/provider can fill them.<p>Why this instead of yet another workflow builder?<p>- Declarative, not glue code: you state what to do; the runtime figures out how. - Agent-first: each step carries natural-language context (purpose, inputs/outputs with meaning) so LLMs can follow, audit, and optimize. Our MCP server enables agents to run pipelines but also to build new pipelines on demand. - Open standard under MIT: language spec, runtime, API server, editor extensions, MCP server, n8n node. - Composable: pipes can call other pipes, created by you or shared in the community.<p>Why a domain-specific language?<p>- We need context, meaning and nuances preserved in a structured syntax that both humans and LLMs can understand - We need determinism, control, and reproducibility that pure prompts can't deliver - Bonus: editors, diffs, semantic coloring, easy sharing, search & replace, version control, linters…<p>How we got there:<p>Initially, we just wanted to solve every use-case with LLMs but kept rebuilding the same agentic patterns across different projects. So we challenged ourselves to keep the code generic and separate from use-case specifics, which meant modeling workflows from the relevant knowledge and know-how.<p>Unlike existing code/no-code frameworks for AI workflows, our abstraction layer doesn't wrap APIs, it transcribes business logic into a structured, unambiguous script executable by software and AI. Hence the "declarative" aspect: the script says what should be done, not how to do it. It's like a Dockerfile or SQL for AI workflows.<p>Additionally, we wanted the language to be LLM-friendly. Classic programming languages hide logic and context in variable names, functions, and comments: all invisible to the interpreter. In Pipelex, these elements are explicitly stated in natural language, giving AI full visibility: it's all logic and context, with minimal syntax.<p>Then, we didn't want to write Pipelex scripts ourselves so we dogfooded: we built a Pipelex workflow that writes Pipelex workflows. It's in the MCP and CLI: "pipelex build pipe '…'" runs a multi-step, structured generation flow that produces a validated workflow ready to execute with "pipelex run". Then you can iterate on it yourself or with any coding agent.<p>What’s included: Python library, FastAPI and Docker, MCP server, n8n node, VS Code extension.<p>What we’d like from you<p>1. Build a workflow: did the language work for you or against you? 2. Agent/MCP workflows and n8n node usability. 3. Suggest new kinds of pipes and other AI models we could integrate 4. Looking for OSS contributors to the core library but also to share pipes with the community<p>Known limitations<p>- Connectors: Pipelex doesn’t integrate with “your apps”, we focus on the cognitive steps, and you can integrate through code/API or using MCP or n8n - Visualization: we need to generate flow-charts - The pipe builder is still buggy - Run it yourself: we don’t yet provide a hosted Pipelex API, it’s in the works - Cost-tracking: we only track LLM costs, not image generation or OCR costs yet - Caching and reasoning options: not supported yet<p>Links<p>- GitHub: <a href="https://github.com/Pipelex/pipelex" rel="nofollow">https://github.com/Pipelex/pipelex</a> - Cookbook: <a href="https://github.com/Pipelex/pipelex-cookbook" rel="nofollow">https://github.com/Pipelex/pipelex-cookbook</a> - Starter: <a href="https://github.com/Pipelex/pipelex-starter" rel="nofollow">https://github.com/Pipelex/pipelex-starter</a> - VS Code extension: <a href="https://github.com/Pipelex/vscode-pipelex" rel="nofollow">https://github.com/Pipelex/vscode-pipelex</a> - Docs: [<a href="https://docs.pipelex.com" rel="nofollow">https://docs.pipelex.com</a>](<a href="https://docs.pipelex.com/" rel="nofollow">https://docs.pipelex.com/</a>) - Demo video (2 min): <a href="https://youtu.be/dBigQa8M8pQ" rel="nofollow">https://youtu.be/dBigQa8M8pQ</a> - Discord for support and sharing: <a href="https://go.pipelex.com/discord" rel="nofollow">https://go.pipelex.com/discord</a><p>Thanks for reading. If you try Pipelex, tell us exactly where it hurts, that’s the most valuable feedback we can get.

Show HN: Quibbler – A critic for your coding agent that learns what you want

Show HN: Quibbler – A critic for your coding agent that learns what you want

Show HN: Quibbler – A critic for your coding agent that learns what you want

Show HN: Butter – A Behavior Cache for LLMs

Hi HN! I'm Erik. We built Butter, an LLM proxy that makes agent systems deterministic by caching and replaying responses, so automations behave consistently across runs.<p>- It’s a chat completions compatible endpoint, making it easy to drop into existing agents with a custom base_url<p>- The cache is template-aware, meaning lookups can treat dynamic content (names, addresses, etc.) as variables<p>You can see it in action in this demo where it memorizes tic-tac-toe games: <a href="https://www.youtube.com/watch?v=PWbyeZwPjuY" rel="nofollow">https://www.youtube.com/watch?v=PWbyeZwPjuY</a><p>Why we built this: before Butter, we were Pig.dev (YC W25), where we built computer-use agents to automate legacy Windows applications. The goal was to replace RPA. But in practice, these agents were slow, expensive, and unpredictable - a major downgrade from deterministic RPA, and unacceptable in the worlds of healthcare, lending, and government. We realized users don't want to replace RPA with AI, they just want AI to handle the edge cases.<p>We set out to build a system for "muscle memory" for AI automations (general purpose, not just computer-use), where agent trajectories get baked into reusable code. You may recall our first iteration of this in May, a library called Muscle Mem: <a href="https://news.ycombinator.com/item?id=43988381">https://news.ycombinator.com/item?id=43988381</a><p>Today we're relaunching it as a chat completions proxy. It emulates scripted automations by storing observed message histories in a tree structure, where each fork in the tree represents some conditional branch in the workflow's "code". We replay behaviors by walking the agent down the tree, falling back to AI to add new branches if the next step is not yet known.<p>The proxy is live and free to use while we work through making the template-aware engine more flexible and accurate. Please try it out and share how it went, where it breaks, and if it’s helpful.