Show

I'm a Type 1 diabetic and software engineer. Last year I went months between endocrinologists with no clinician reviewing my data. I'm an engineer, so I built the tool I needed — and now I'm open sourcing it. GlycemicGPT is a self-hosted platform that connects continuous glucose monitors, insulin pumps, and existing Nightscout instances to an AI analysis layer running on your own infrastructure. Data sources:

Dexcom G7 (cloud API) Tandem t:slim X2 and Mobi pumps (direct BLE) Nightscout (point it at your existing instance and you're running in minutes)

What the AI layer does:

Daily briefs summarizing overnight and 24-hour patterns Meal response analysis Conversational chat with RAG-backed clinical knowledge Predictive alerting with configurable thresholds and caregiver escalation

Important: this is monitoring and analysis only. GlycemicGPT does not deliver insulin, does not control your pump, and is not a closed-loop system. It reads your data and gives you insight on top of it. Your clinical decisions stay between you and your care team. Architecture:

Self-hosted via Docker or K8S — the GlycemicGPT stack runs entirely on your hardware BYOAI — bring your own AI provider. Use Ollama for fully local operation (no data leaves your hardware), or point it at Claude, OpenAI, or any OpenAI-compatible endpoint if you prefer a hosted model. Data flows directly from your instance to the provider you choose; nothing is routed through any centralized service operated by the project. GPL-3.0, no subscriptions, no vendor lock-in

Stack:

Backend API: FastAPI, Python 3.12, PostgreSQL 16, Redis 7 Web Dashboard: Next.js 15, React 19, Tailwind CSS, shadcn/ui AI Sidecar: TypeScript, Express, multi-provider proxy Android App: Kotlin, Jetpack Compose, BLE Wear OS: Kotlin, Wear Compose, Watch Face Push API Plugin SDK: Kotlin interfaces, capability-based, sandboxed

Looking for contributors — especially folks with BLE/Android experience or anyone in the diabetes tech space. Plugin SDK is documented if you want to add support for new devices. GitHub: https://github.com/GlycemicGPT/GlycemicGPT

Hey HN,

My co-founders and I have been building GridTravel, a free iOS app for planning and sharing travel routes with turn-by-turn GPS nav. We just launched yesterday after App Store approval.

We're three 21-year-old cofounders and best friends since middle school. We built GridTravel after years of frustration navigating new cities on every trip we took together.

The idea: most people either search Google for "top 10 places to visit in…" lists or go on social media to get inspiration on where to go. GridTravel is built around user-generated routes — actual paths someone walked, that you can follow, save, download, and discover from other travelers. Users also have the ability to create private routes and collaborate with their friends.

Tech stack: Mapbox (Nav SDK + maps), Supabase (auth, DB, storage), and Swift. Native iOS for now, Android coming soon.

Our two real cost drivers are Mapbox Search (hit when users create routes) and Mapbox Navigation (hit when users use live navigation). Both have free tiers, then scale with MAU. We launched fully free to remove the barrier to entry. Revisiting pricing in Year 2 once nav costs start burning a hole in our pocket.

Current state: we're in the UGC cold-start hole. The app's value scales with route density in a given city, but route density requires users, who require routes. Classic chicken and egg. Our current plan: 1. Manually seed 25–30 routes per city, starting with 5-10 priority cities where we have personal networks rather than spreading ourselves thin. 2. Short-form content as the primary social channel (TikTok, reels, shorts). Doing A/B testing: whether route walkthroughs convert better than informational/skit videos. 3. Partnering with micro-influencers in those cities (5k-50k following) for in-app routes plus cross-posts on their channels

Curious what HN thinks. Especially anyone who's shipped a UGC product. What worked for you on cold start? What do you wish you'd done differently? Happy to answer any questions about the app, costs, etc.

App link: https://apps.apple.com/us/app/gridtravel-local-routes/id6762...

Hey HN, Henry here from Cactus. We open-sourced Needle, a 26M parameter function-calling (tool use) model. It runs at 6000 tok/s prefill and 1200 tok/s decode on consumer devices.

We were always frustrated by the little effort made towards building agentic models that run on budget phones, so we conducted investigations that led to an observation: agentic experiences are built upon tool calling, and massive models are overkill for it. Tool calling is fundamentally retrieval-and-assembly (match query to tool name, extract argument values, emit JSON), not reasoning. Cross-attention is the right primitive for this, and FFN parameters are wasted at this scale.

Simple Attention Networks: the entire model is just attention and gating, no MLPs anywhere. Needle is an experimental run for single-shot function calling for consumer devices (phones, watches, glasses...).

Training: - Pretrained on 200B tokens across 16 TPU v6e (27 hours) - Post-trained on 2B tokens of synthesized function-calling data (45 minutes) - Dataset synthesized via Gemini with 15 tool categories (timers, messaging, navigation, smart home, etc.)

You can test it right now and finetune on your Mac/PC: https://github.com/cactus-compute/needle

The full writeup on the architecture is here: https://github.com/cactus-compute/needle/blob/main/docs/simp...

We found that the "no FFN" finding generalizes beyond function calling to any task where the model has access to external structured knowledge (RAG, tool use, retrieval-augmented generation). The model doesn't need to memorize facts in FFN weights if the facts are provided in the input. Experimental results to published.

While it beats FunctionGemma-270M, Qwen-0.6B, Granite-350M, LFM2.5-350M on single-shot function calling, those models have more scope/capacity and excel in conversational settings. We encourage you to test on your own tools via the playground and finetune accordingly.

This is part of our broader work on Cactus (https://github.com/cactus-compute/cactus), an inference engine built from scratch for mobile, wearables and custom hardware. We wrote about Cactus here previously: https://news.ycombinator.com/item?id=44524544

Everything is MIT licensed. Weights: https://huggingface.co/Cactus-Compute/needle GitHub: https://github.com/cactus-compute/needle

I used to study by rereading notes, and then I blanked in the exam hall.

Did some research, and found that my experience isn’t isolated, and that passive review doesn’t force retrieval, so nothing sticks, and I knew I had to do something about it.

That’s why I built Professor Goose. You pick a topic, explain it out loud to a goose, and he keeps probing until he understands you. Never gives you the answer, just keeps asking follow ups until a sound understanding is reached, which in turn makes you figure stuff out or realise you never understood your topic in the first place.

Free to try, no account needed, upload your syllabus for exam board specific questions. Curious whether this approach resonates with others, it sure has for me.

Every privacy-focused DNS service requires an account: NextDNS, Cloudflare for Families, Apple's iCloud Private Relay (paid, iOS-only). The protocol that doesn’t require one - ODoH - had basically one well-known public relay operator (Frank Denis on Fastly Compute, default in dnscrypt-proxy). I built a second one and the client to talk to it.

An attempt at a single pass LLVM frontend in ~3000 lines of C without external dependencies, malloc, or an AST. Included are some graphical examples. The IR isn't perfect, and the README touches on one particular downfall

A few months ago I stumbled on obra's superpowers repository https://github.com/obra/superpowers. I really liked the approach and idea that you enforce discipline for your agent through a skill-based workflow. Even though coding agents (copilot included) have become a lot better at natively handling complex tasks, they still wander off and lose track of things. I really liked how superpowers fixed this and how it enabled long-running sessions without the agent losing its "focus". So I decided to build a Copilot tailored skill suite around the core idea of superpowers. I didn't just want to port superpowers to Copilot, I took inspiration from it and improved on it. JDS enforces a strict think -> plan -> execute pipeline where nothing gets skipped. It leverages Copilot's built-in sql todo dependencies and provides a live task graph visualizer which helps visualize the agentic workflow and its parallelism. Curious whether others have tried similar approaches, and what's worked or not.

Hi HN, I’m Namanyay from Gigacatalyst (link: https://gigacatalyst.com/). Gigacatalyst allows sales, CS, and users to build one-off features, so your SaaS can support long-tail customer workflows and engineers aren’t pulled away from the roadmap.

When you sell software to large businesses, you realize that each customer needs their own workflow and features. Traditionally, this either means long engineering roadmaps or the customers end up using workarounds.

But what if everyone could build their critical missing features just by talking to an AI? That’s what we do at Gigacatalyst. We provide an AI customization layer for your customers, CS team, and sales team to build these missing critical workflows without needing any engineers at all. Think Lovable, but built on top of YOUR platform.

We connect to your product's APIs, learn your data model and design system, and let non-technical users build governed apps via natural language - inside your product, under your brand.

Here’s what it looks like in action: https://www.youtube.com/watch?v=_taSpSphH6E

One of our customers, a Series B company, saw their users (not engineers - managers, ops people, facility directors) build critical workflows like:

- Parts stockout prevention: A maintenance manager typed "show me which parts will run out in the next 2 weeks based on usage over the last 90 days, accounting for vendor lead times." The app tracks consumption velocity, forecasts stockouts, and alerts before it's too late. He says it's prevented ~$500K in emergency downtime.

- Invoice OCR from phone photos: Technicians kept losing paper invoices. The prompt: "upload a photo of the invoice, extract vendor name, date, amount, and line items, then match it to the purchase order and flag discrepancies." Now techs snap a photo on-site to automatically add to the system of record.

- Restaurant emergency triage: A pizza chain's facilities manager was drowning in maintenance requests. He built a priority matrix: "walk-in freezer not cooling" auto-routes as CRITICAL, "dining room light flickering" goes to LOW. He's now able to manage backlogs with the correct priority.

How Gigacatalyst works under the hood:

1. Agentic API discovery: Our agents go through your app and parse your endpoints, query params, request/response shapes, and sample data to build the base layer.

2. Generation and Validation: When a user describes what they want our AI generates an app. We set up multiple validation steps, including static checks, runtime error analysis, and LLM-as-a-judge.

3. Sandboxing and Compilation: We wrote our own compilation and sandboxing framework to get the fastest speeds and lowest costs. This means that users can interact with the built app in seconds.

4. Proxy layer: We create a proxy layer for all APIs to handle auth, tenant isolation, and rate limiting. Everything the agent has access to is controlled, logged, observed, and version controlled.

After 2000+ daily users, 900+ apps built, and 70% 30-day retention, today we're opening a public demo.

Try it: https://app.gigacatalyst.com/ - enter your SaaS product's API URL (or just the homepage) and start prompting.

If you're serving a variety of use cases, you probably deal with a lot of custom requests and Gigacatalyst will save you time and increase your bottom line. Book a meeting at https://gigacatalyst.com/#contact and I'll help your team and customers build new functionality on top of your platform.

I've been reading Hacker News since I was 12 years old. I'm proud to launch for all of you and I want to hear your feedback on my product and comments!

Hi HN, we’re Sai and Aayush, and we’re building Hypercubic (https://www.hypercubic.ai/), bringing AI tools to the mainframe and COBOL world. (We did a Launch HN last year: https://news.ycombinator.com/item?id=45877517.) Today we’re launching Hopper, an agentic development environment for mainframes.

You can download it here: https://www.hypercubic.ai/hopper, and you can also request access and immediately get a mainframe user account to play with.

There's also a video runthrough at https://www.youtube.com/watch?v=q81L5DcfBvE.

Mainframes still run a surprising amount of critical infrastructure: banking, payments, insurance, airlines, government programs, logistics, and core operations at large institutions. Many of these systems are decades old, but they continue to process enormous transaction volumes because they are reliable, secure, and deeply embedded into business operations.

A lot of that software is written in COBOL and runs on IBM z/OS. The development environment looks very different from modern cloud or Unix-style development. Instead of GitHub, shell commands, package managers, and CI pipelines, developers often work through TN3270 terminal sessions, ISPF panels, partitioned datasets, JCL, JES queues, spool output, return codes, VSAM files, CICS transactions, and shop-specific conventions.

TN3270 is the terminal interface used to interact with many IBM mainframe systems. ISPF is the menu and panel system developers use inside that terminal to browse datasets, edit source, submit jobs, and inspect output. It is powerful and reliable, but it was designed for expert humans navigating screens, function keys, and fixed-width workflows, not AI agents.

A simple COBOL change might require finding the right source member, checking copybooks, locating compile JCL, submitting a job, reading JES/SYSPRINT output, interpreting condition codes, patching fixed-width source, and resubmitting.

Much of this work is so well-defined and repetitive that it's a good fit for agentic AI. To get that working, however, a chatbot next to a terminal is not enough. The agent needs to operate inside the mainframe environment.

Hopper combines three things: (1) A real TN3270 terminal, (2) Mainframe-aware panels for datasets, members, jobs, and spool output, and (3) An AI agent that can operate across those z/OS surfaces.

For example, here is a tiny version of the kind of thing Hopper can help debug:

  COBOL:

   IDENTIFICATION DIVISION.
   PROGRAM-ID. PAYCALC.

   DATA DIVISION.
   WORKING-STORAGE SECTION.
   01  CUSTOMER-BALANCE     PIC 9(7)V99.

   PROCEDURE DIVISION.
       ADD 100.00 TO CUSTOMER-BALNCE
       DISPLAY "UPDATED BALANCE: " CUSTOMER-BALANCE
       STOP RUN.


  JCL:

    //PAYCOMP  JOB (ACCT),'COMPILE',CLASS=A,MSGCLASS=X
    
    //COBOL    EXEC IGYWCL
    
    [//COBOL.SYSIN](https://cobol.sysin/) DD DSN=USER1.APP.COBOL(PAYCALC),DISP=SHR
    
    [//LKED.SYSLMOD](https://lked.syslmod/) DD DSN=USER1.APP.LOAD(PAYCALC),DISP=SHR

A human would submit this job, inspect JES output, open `SYSPRINT`, find the undefined `CUSTOMER-BALNCE`, map it back to the source, patch the member, and resubmit. Hopper is designed to let an agent operate through that same loop autonomously.

Hopper is not trying to hide the mainframe behind a generic abstraction, and it's not a chatbot. The design principle is simple: preserve the fidelity of the mainframe environment, but make it accessible to AI agents.

Sensitive operations require approval, and the terminal remains visible at all times.

Once agents can operate inside the mainframe environment, new workflows become possible: faster job debugging, automated documentation, safer code changes, test generation, migration planning, traffic replay, and modernization verification.

We’re curious to hear your thoughts! especially from anyone who has worked with mainframes, COBOL or has done legacy enterprise modernization.

Agentic problem solving in its current state is very brittle. I fell in love with it, but it creates as many problems as it solves.

I'm Ben Cochran, I spent 20+ years in the trenches with full-stack Engineering, DevOps, high performance computing & ML with stints at NVIDIA, AMD and various other organizations most recently as a Distinguished Engineer.

For agents to work reliably you either need massive parameter counts or massive context windows to keep the solution spaces workable. Most people are brute forcing reliability with bigger models and longer prompts.

What if I made the problem smaller instead of making the model bigger?

I took a different approach by using smaller models: models in the 13-20B parameter range and set them to task solving real SWE-bench problems. I constrained the tool and solution spaces using formal state machines. Each state in the machine defines which tools the model can access, how many iterations it gets and what transitions are valid. A planning state gets read-only tools. An implementation state gets edit tools (scoped to prevent mega edits) and write friendly bash tools. The testing state gets bash but only for testing commands. The model cannot physically skip steps or use the wrong tool at the wrong time. It is enforced via protocol, not via prompts.

The results were more promising than I would have expected. Across multiple model families irrespective of age (qwen-coder, gpt-oss, gemma4) and the improvements were consistent above the 13B parameter inflection point. Below that, models can navigate the state machine but can't retain enough context to produce accurate edits. More on the research bit: https://statewright.ai/research

Surprisingly this yielded improvements in frontier models as well. Haiku and Sonnet start to punch above their weight and Opus solves more reliably with fewer tokens and death spirals. Fine tuning did not yield these kinds of functional improvements for me. The takeaway it seems is that context window utilization matters more than raw context size - a tightly scoped working context at each step outperforms a model given carte blanche over everything. Constraining LLMs which are non-idempotent by using deterministic code is a pattern that nobody is currently talking about.

So, I built Statewright. Its core is a Rust engine that evaluates state machine definitions: states, transitions, guards and tool restrictions. Its orchestration doesn't use an LLM, just enforces the state machine. On top of that is a plugin layer that integrates with Claude Code (and soon Codex, Cursor and others) via MCP. When you activate a workflow, hooks enforce the guardrails per state automatically. The model sees 5 tools available instead of dozens, gets clear instructions for the current phase and transitions when conditions are met. Importantly it tells the model when it's attempting to do something that isn't in scope, incorrect or when it needs to try something else after getting stuck.

You can use your agent via MCP to build a state machine for you to solve a problem in your current context. The visual editor at statewright.ai lets you tweak these workflows in a graph view... You can clearly see the failure paths, the retry loops and the approval gates. State machines aren't DAGs; they loop and retry, which is what agentic work actually needs.

Statewright is currently live with a free tier, try it out in Claude Code by running the following:

/plugin marketplace add statewright/statewright

/plugin install statewright

/reload-plugins

Then "start the bugfix workflow" or /statewright start bugfix. You'll need to paste your API key when prompted. The latest versions of Claude may complain -- paste the API key again and say you really mean it, Claude is just being cautious here.

Feedback is welcome on the workflow editor, the plugin experience, and tell me what workflows you'd want to build first. Agents are suggestions, states are laws.

Most AI chat applications (such as ChatGPT or Claude) stream their responses to the client as markdown text. As each new chunk of text arrives, the front end typically re-parses the entire markdown document to render the updated message. This works, but it can quickly slow down the UI for long responses.

I’ve been obsessing over ways to make this more efficient, so I wrote a markdown parser that can parse streaming markdown (semi) incrementally. Instead of re-processing the whole document each time, it only parses what’s new, processing each line only once. Block‑level nodes are buffered until they’re complete (for example, once a paragraph is done and won’t be extended by more text). This also makes parsing the markdown on server possible. The main demo does exactly that. Additionally, animating markdown blocks becomes much simpler and efficient, as a result.

Here’s a demo if you’d like to see it in action: https://markdownparser.vercel.app/experimental

Feel free to type 'Render a table with 10 rows' to see each table row animate in.

I’ve spent a lot of time thinking about this problem, so if you’re working on similar issues, I'd love to chat.

Permission-first config kit for Claude Code and Google Antigravity IDE. Installs CLAUDE.md + GEMINI.md + 10 specialist agents + 28 skills with one command. MIT license.

It supports: - Image, SF Symbol, and autoplaying video pages - Optional skip behavior - Custom theming - Completion gating - Snapshot-tested SwiftUI UI

Robotics is advancing really fast lately, with AI inference, different controllers, software, and parts always changing. I wanted a place that supports many device types, Raspberry Pi, NVDA Jetson, Arduino, ESP32, hardware sources, and maximizes for printability. Instructables, Github, and Thingiverse are currently popular but aren't really focused on robotics, So I built orobot.io to try and make printing robots as standardized and accessible as possible. It uses a lot of Agent built content custom to each project, and every project is designed to be used by humans or your agent.

Features:

- Photos and Estimated Prices for all projects

- Links back to source GitHub projects

- LLMs write descriptions and tips on how to build

- View + Download 3d printable STL files in browser

- BOM purchase links are kept up to date with LLMs checking Amazon link health

- LLMs write Javascript install and controller wrappers custom to each project so a single one-click install works across many frameworks and controller types

- Public skill files, clis, and prompts let your agent do everything it needs to walk you through the complexity.

It's still pretty new, so somethings are broken, and there's a lot more I want to build. But I'm very interested to have people try it out let me know if they want to use something like this and give me feedback about where they ran into problems so I can fix it. Thank you HN!

The two from parameter golf (one I trained, one was the baseline) are just 16MB each! They produce barely plausible English