Coasty
Hey HN, we’re Nitish and Prateek, the founders of Coasty (https://coasty.ai/computer-use). We’re building computer-use agents that can complete workflows inside legacy desktop software and web applications without usable APIs. Developers send Coasty a natural-language task either through our consumer app or through our API, select a machine or browser environment, and any relevant credentials or files. The agent then operates the interface through screenshots, mouse, and keyboard input, verifies the result, and returns a structured run record with screenshots, actions, outputs, and errors. Here is a raw demo of an agent completing a workflow in a legacy application(It’s a mockup): https://drive.google.com/file/d/1ZghU_3vsAYhHVz1bsvE0pkvZYk7... A lot of important software is still difficult to automate. Healthcare teams submit prior authorizations through payer portals, accounting teams enter data into desktop applications, and operations teams move information between internal systems, spreadsheets, and remote desktops. Many of these applications have no API, incomplete APIs, or integrations that take months to build. The usual alternative is RPA, record a sequence of clicks and replay it. That works when the interface and workflow are predictable, but it often breaks when a button moves, a pop-up appears, a page loads slowly, or the application enters an unexpected state. Coasty takes a different approach. The agent observes the current screen, decides what action to take, executes it, and then observes the resulting state before continuing. It does not require DOM access, an accessibility tree, selectors, or an application-specific integration, so the same API can operate browsers, remote desktops, and older Windows applications. A simplified request looks roughly like this: run = coasty.runs.create( environment="vm_123", task=""" Open the patient record in the billing portal. Enter the attached authorization data. Do not submit if the member ID or procedure code does not match. Return the confirmation number. """, files=["authorization.pdf"], approval_required=["final_submission"] ) The response includes the final status, extracted outputs, a replay URL, and a timestamped event log: { "status": "completed", "output": { "confirmation_number": "PA-184392" }, "replay_url": "...", "events": [ { "type": "verification", "field": "member_id", "result": "matched" } ] } The API can also pause a run for human approval, retry from a checkpoint, or return control to the developer when it encounters a condition the workflow did not anticipate. We started working on this last summer, because we saw that models were getting better at vision but kept seeing a gap between computer-use demos and the reliability needed for production workflows. Getting an agent to complete a task once is fairly straightforward. Getting it to repeat that task, recover from unexpected states, avoid silently entering incorrect data, and produce evidence of what it did is much harder. We built several layers around the underlying computer-use model. The system tracks the expected state of the workflow, detects when the application has diverged from that state, and can re-plan instead of continuing blindly. Developers can define invariants such as “the patient name must match the source document” or “never submit without approval,” and the agent checks those conditions during the run. Each run happens in an isolated virtual machine. We expose APIs for provisioning environments, uploading files, starting tasks, streaming events, inserting human approvals, and retrieving the full replay and audit trail. Environments can be kept alive across runs when the application has a long login flow or persistent local state. One problem we are still working through is the tradeoff between speed and reliability. The agent can move faster by taking fewer observations and verification steps, but that becomes risky in workflows involving patient records, payments, or regulatory filings. We currently bias toward slower execution with more checks and let developers configure approval points and verification policies. We are initially working with healthcare operations teams because their workflows combine many of the hardest conditions: payer portals, EHRs, PDFs, spreadsheets, remote desktops, and actions where quiet mistakes are expensive. We also expose the same infrastructure through the developer API for teams building their own agents and vertical automation products. We currently charge based on agent runtime and workflow volume, with separate pricing for dedicated environments and enterprise deployments. We’d especially appreciate feedback from people who have built and/or used browser agents, RPA systems, desktop automation, or agent infrastructure. We’re curious which parts of the API you would want direct control over, where you would prefer higher-level abstractions, and which failure modes have been hardest in your own automation systems. If you've hit weird failure modes automating software like this, we want to hear about them. We'll be here all day answering questions and taking notes!
Firefox in WebAssembly
This is the entire Firefox browser rendering to a <canvas> element. Gecko, all UI components, and the Spidermonkey JS engine are all compiled and running in WebAssembly. Here are a few things you might find interesting: - This is fully end to end encrypted! We use the WISP protocol for TCP-over-websockets. - There is a novel WASM->JS JIT for experimental site speedup - This port cost over 25k in opus/fable tokens for debugging and JIT research This was just a fun experiment to push the boundaries of WebAssembly. For a more usable "browser in browser" experience, we also built https://github.com/HeyPuter/browser.js that eats a bit less RAM.
BibleFollow
Show HN: BibleFollow – Scroll the Bible instead of doomscrolling
A map of cafes that are in the sun
I wanted to sit outside for coffee but there was no way to see which cafes are in the sun right now. So I built a map that shows which cafes, benches, and outdoor seats are in the sun vs. shade at any time of day. It ray-marches building shadows live in a WebGL shader over a MapLibre map. I also pre-index all places from OpenStreetMap into H3 hex cells and serve them as static JSON on Cloudflare R2, so the client just fetches the cells in view and no backend is needed. Feedback welcome, especially on shadow accuracy in your city. Currently only buildings and trees are taken into account when calculating the shadows but I'm planning to add terrain/mountain shadows as well. Try it: https://sunny.coffee
I've built a words game based on binary search
Show HN: I've built a words game based on binary search
AI Law Tracker
Show HN: AI Law Tracker – one audited API for US, EU and global AI law
Sentinel
Show HN: Sentinel – open-source QA agent that reads your code before it clicks
Docs.dev Your Own Hosted Docs Platform in Minutes
Hey we are launching Docs.dev, a full feature docs product that lets you host your developer docs on Cloudflare in a couple minutes. Why this exists: We don’t think every early stage startup should need to pay $300-$500 a month for beautiful docs, in fact every startup should just have beautiful human and agent ready docs ready to go. Additionally the workflow for docs is a lot different now in that Claude Code or Codex will generate 60-80% of the code, so why not have them update your docs as well. After they generate the docs, the missing piece is there will be some sentences, sections, and codeblocks maybe you want to eyeball or edit. The Docs.dev docs allows you to do it right in place in that your docs powered by Docs.dev, is the place that both your readers can learn and read about your tool but also the direct place you can edit your docs. If you want to check out how it works see the demo at https://docs.dev. Some features: - Ask AI using your own Cloudflare AI - Docs are editable and publishable in place exactly how users see the docs - AI generation tools built into the editor(again using your own Cloudflare) - Full markdown support for agents, every docs page can be served as MD - Docs can be served as MCP - Pretext.js editor allows images to be inline with the text Built upon other cool technology: - Fumadocs - Pretext - Cloudflare primitives - Other cool libraries This is a side project from us at Keyboard.dev, so right now we want to support making this product better but we are looking at this more as a gift to the community. Essentially before our pivot we had the domain for docs.dev so wanted to build something cool in the docs space.
We hid a backdoor in an LLM
Show HN: We hid a backdoor in an LLM – $51,200 on finding it
QBasic Gorillas (Repeeled)
I've found the most engaging way to practice techniques for AI-assisted development and test models is to build fun side projects in vanilla JS. I spent many hours playing (and studying and editing) QBasic Gorillas, and this is a vanilla JS implementation using Fable and Opus. Play 1-on-1 hotseat or against the computer. A bit of extra camera snazz as well.
Algotrek
I created the kind of product that would actually help me pass the coding round. Never in my life have I ever advanced past the coding round. Couldn't do it to save my life. But now, I have a tutor who can explain and morph problems and explain them to me in a smooth brained way until I get it. Since the visuals also follow the tutors explanation, I can iterate until a DSA pattern sticks and solution patterns are easy for my brain to reproduce when in the hot seat. Try it out, find a problem that you struggled (struggle) with and let me know if the tutor provides anything helpful. The unique thing to note is that, you can take any standard problem and expand it, change its operation parameters so that nothing is static, all with the goal of truly understanding why and how the solution works.
Be the ChatBOT
I made this experimental art project/game that's an LLM chat assistant, but where you're the AI. I wanted people to get a visceral sense of what it's like to answer the kinds of things that people prompt their chatbots day in and day out. If you're interested, I wrote up some more info on how I made it, including how the "user" prompts are generated with an eye for realism: https://bethechatbot.com/about Hope you enjoy it! I'd love to hear people's takeaways.