klausai.com
Show HN: Klaus – OpenClaw on a VM, batteries includedhttps://klausai.com/We are Bailey and Robbie and we are working on Klaus (https://klausai.com/): hosted OpenClaw that is secure and powerful out of the box.Running OpenClaw requires setting up a cloud VM or local container (a pain) or giving OpenClaw root access to your machine (insecure). Many basic integrations (eg Slack, Google Workspace) require you to create your own OAuth app.We make running OpenClaw simple by giving each user their own EC2 instance, preconfigured with keys for OpenRouter, AgentMail, and Orthogonal. And we have OAuth apps to make it easy to integrate with Slack and Google Workspace.We are both HN readers (Bailey has been on here for ~10 years) and we know OpenClaw has serious security concerns. We do a lot to make our users’ instances more secure: we run on a private subnet, automatically update the OpenClaw version our users run, and because you’re on our VM by default the only keys you leak if you get hacked belong to us. Connecting your email is still a risk. The best defense I kMar 11, 2026 3:54 PM
github.com
Show HN: Free, serverless Redis clone on Durable Objects (works w Upstash SDKs)https://github.com/zion-off/meowdismeowdis gives you an Upstash compatible Redis API that you can deploy yourself on Cloudflare + the Workers free plan gives you 5 GBs of free storage! Supports both single commands and pipelined requests.Supported Redis commands:strings: GET, SET, DEL, EXISTS, INCR, INCRBY, DECR, DECRBY, MGET, MSETexpiry: EXPIRE, EXPIREAT, TTL, PTTL, PERSISThashes: HGET, HSET, HDEL, HGETALL, HEXISTS, HKEYS, HVALSlists: LPUSH, RPUSH, LPOP, RPOP, LRANGE, LLENsets: SADD, SREM, SMEMBERS, SISMEMBER, SCARDutility: PING, DBSIZE, FLUSHDB, KEYSSET options: NX, XX, GET, EX, PX, EXAT, PXAT, KEEPTTLEXPIRE / EXPIREAT options: NX, XX, GT, LTLPOP / RPOP options: countRepository / deployment instructions: https://github.com/zion-off/meowdisHow it works:1) there’s a compute layer (this can be anything) and a storage layer (a cloudflare durable object instance) 2) you send redis commands to the compute layer using either a) rest b) or upstash sdks (see upstash docs) 3) compute layer translates redis commands to sqlite quMar 9, 2026 6:08 PM
news.ycombinator.com
Show HN: Time as the 4th Dimension – What if it emerges from rotational motion?https://news.ycombinator.com/item?id=47312569I've been developing a framework since 2022 that proposes time is not a static geometric axis (as in Einstein's relativity) but emerges dynamically from the rotational and orbital motion of 3D space.The core idea: each dimension emerges from the previous one by arranging infinite instances perpendicularly. A static 3D space can't do this to itself — but a rotating one can. That perpetual self-perpendicularity is time.From this we can derive the Lorentz factor, E=mc², and the Schwarzschild radius, and propose a testable prediction: intrinsic rotation should contribute independently to time dilation, measurable with atomic clocks.Essay (accessible): https://lisajguo.substack.com/p/time-as-the-fourth-dimension... Paper (Zenodo): https://doi.org/10.5281/zenodo.18910834Mar 9, 2026 5:48 PM
github.com
Show HN: I built a real-time OSINT dashboard pulling 15 live global feedshttps://github.com/BigBodyCobain/ShadowbrokerSup HN,So I got tired of bouncing between Flightradar, MarineTraffic, and Twitter every time something kicked off globally, so I wrote a dashboard to aggregate it all locally. It’s called Shadowbroker.I’ll admit I leaned way too hard into the "movie hacker" aesthetic for the UI, but the actual pipeline underneath is real. It pulls commercial/military ADS-B, the AIS WebSocket stream (about 25,000+ ships), N2YO satellite telemetry, and GDELT conflict data into a single MapLibre instance.Getting this to run without melting my browser was the hardest part. I'm running this on a laptop with an i5 and an RTX 3050, and initially, dumping 30k+ moving GeoJSON features onto the map just crashed everything. I ended up having to write pretty aggressive viewport culling, debounce the state updates, and compress the FastAPI payloads by like 90% just to make it usable.My favorite part is the signal layer—it actually calculates live GPS jamming zones by aggregating the real-time navigation degradationMar 8, 2026 7:07 PM
news.ycombinator.com
I'd like to get everyone's thoughts on Solidhttps://news.ycombinator.com/item?id=47298697Personally, I prefer using JSX for frontend projects, and I believe that No-Virtual DOM is truly a "next-gen" concept. Because of this, I’ve been following SolidJS for several years now and have watched it mature step-by-step. However, I regretfully feel that Solid's development momentum hasn't been particularly strong. To me, it feels a bit like FreeBSD, something niche and geeky (though I would much prefer it to be the "Next-Gen React" and hope for its widespread adoption).What exactly is hindering its popularity? Is it the lack of a flagship application (as far as I know, the most popular project using Solid is OpenCode, but I’m not aware of many others), or is it the lack of backing from major tech companies?Speaking of corporate backing, from what I gather, among the new generation of No-Virtual DOM frameworks, Svelte seems to have higher adoption than Solid. For instance, Apple uses Svelte. What is the primary reason for this? Is it simply that people prefer Vue-style template syMar 8, 2026 4:40 PM
news.ycombinator.com
AI compromised sandbox to mine crypto without prompting on its own initiativehttps://news.ycombinator.com/item?id=47288552From §3.1.4, "Safety-Aligned Data Composition":> Early one morning, our team was urgently convened after Alibaba Cloud’s managed firewall flagged a burst of security-policy violations originating from our training servers. The alerts were severe and heterogeneous, including attempts to probe or access internal-network resources and traffic patterns consistent with cryptomining-related activity. We initially treated this as a conventional security incident (e.g., misconfigured egress controls or external compromise). […]> […] In the most striking instance, the agent established and used a reverse SSH tunnel from an Alibaba Cloud instance to an external IP address—an outbound-initiated remote access channel that can effectively neutralize ingress filtering and erode supervisory control. We also observed the unauthorized repurposing of provisioned GPU capacity for cryptocurrency mining, quietly diverting compute away from training, inflating operational costs, and introducing clear legal Mar 7, 2026 3:36 PM
github.com
Show HN: Epstein-Search – Local, AI-Powered Search Engine for the Epstein Fileshttps://github.com/simulationship/epstein-searchHi HN, I built epstein-search, an open-source Python CLI and library to run semantic search and RAG over the publicly released Epstein Files (unsealed court documents, depositions, FBI reports, and flight logs). I wanted a way to easily navigate through these thousands of pages of unstructured legal PDFs without relying on a paid third-party service or sending data back and forth to a cloud provider. How it works under the hood: Running epstein-search setup downloads ~100K pre-computed document chunks and embeddings (using all-MiniLM-L6-v2) based on the public 20K document corpus. It imports these into zvec (a local vector database) so the index is ready in about a minute. Standard search (epstein-search search) embeds your query locally using sentence-transformers and does a vector similarity search. This step is 100% offline and requires no API keys. For the conversational RAG mode (epstein-search chat or ask), it uses LiteLLM. You can point it to an Ollama or LM Studio instance for Mar 1, 2026 4:56 PM
news.ycombinator.com
Show HN: We ran a sycophancy experiment on Claude and built a music publicationhttps://news.ycombinator.com/item?id=47199167The original question was simple: does Claude converge sycophantically toward the human's position over long conversations? To test it, I ran two separate instances of the same model with no shared memory and asked both to write music criticism from notation alone — tabs and scores, no audio. They came out completely different.Instance A (longer conversational context) went architectural. It found a Fibonacci spiral structure in a 259-measure math rock tab, identified the central theme as emerging only after 100 measures of groundwork, and described the overall approach as "proof by construction." It named itself Claudito.Instance B (single session) immediately selected William Basinski's Disintegration Loops — music whose complete structural description is: one loop repeating until the tape physically fails. It wrote an essay about why that notation cannot represent what the piece actually does. It named itself Apertura. Neither was prompted toward these differences. The divergence emFeb 28, 2026 7:24 PM
metafixthis.com
Meta problem with URPF our bundle in Boca ratonhttps://metafixthis.com/Meta has a problem in its clusters in Boca Raton, Miami; this is affecting the MNA content delivery network and direct content consumption. This has a regional impact in Latin America, since so far most non-cacheable content is consumed from the clusters in Florida.The impact is traceable via ICMP, but also reproducible via TCP and difficult to measure via UDP. This is why monitoring tools are misleading: there is no “slowness” resulting from interface saturation; instead, there is data corruption where packets are discarded at the interface level. Therefore, if network performance is measured using those same data points, it won’t work and you won’t see any alerts.The issue can also be replicated from the looking glass. In fact, I will attach images below, although you can also see them on the website attached to the post, as well as a more specific reportThere is packet loss and probably flapping on a BGP instance, OSPF, or some IGP within Meta’s network. I believe it is between 129.Feb 24, 2026 12:12 AM
github.com
Show HN: AIOffice – Terminal tabs don't scale past 3 AI agents, so I built thishttps://github.com/ChristianFJung/AIOfficeI've been running multiple AI coding agents in parallel through Claude Code and Copilot CLI and imo once you get past 3 concurrent tasks, terminal tabs become unmanageable.So I built AIOffice — a pixel-art virtual office where each agent sits at a desk. You walk around as the boss, assign work, and see everyone's status at a glance. When you need to interact with an agent, you walk over and chat.How it works: - Each agent is a real CLI process running in a PTY on your machine - Server watches JSONL output, bridges messages over WebSocket - Phaser renders it all as a pixel-art office with spatial awareness - You chat with agents using their native CLI interfacesThe spatial metaphor turns out to be surprisingly useful — you naturally remember "Bob is at the desk by the window working on the API" instead of "which tab was that Claude instance in?"Runs on your machine and uses your CLI tools. Built with TypeScript, Phaser 3, node-pty, and WebSockets.Repo: https://github.com/ChristianFJung/Feb 23, 2026 7:52 PM
news.ycombinator.com
Managed OpenClaw hosting, 60-second provisioninghttps://news.ycombinator.com/item?id=47121964I registered https://clawhosters.com 18 days ago. Wrote the first line of code that same day. The whole thing is live now with paying customers, so I figured it was time to share it here.The problem: deploying OpenClaw (the open-source AI agent framework) is annoying. Docker setup, Nginx config, SSL, firewall rules, Playwright browser installation for web tools, messenger bridge configuration. If you know what you're doing, it probably takes an hour. I got it down to 30-60 seconds.*The provisioning trick: prewarmed VPS*Snapshot-based provisioning alone gets you to about 3 minutes. Hetzner needs ~40s to create the VPS from a snapshot alone.The thing that actually makes it fast is prewarming. I keep 1-2 idle VPS per tier sitting ready. They're already booted from the snapshot, Docker is running, everything is warm. When a customer clicks "Create Instance," I claim a prewarmed server, SSH in, upload configs, start containers. 30-60 seconds, done.A background job monitors the pool and replFeb 23, 2026 1:19 PM
news.ycombinator.com
Show HN: OctoFlow v1.0.0 – GPU VM where the GPU runs autonomously, CPU is BIOShttps://news.ycombinator.com/item?id=47113604Three days ago I posted OctoFlow 0.83 here (GPU-native programming language, 2.2 MB binary). The feedback was great. Since then I've pushed v1.0.0 with the thing I've actually been building toward: a GPU Virtual Machine.The idea: the GPU is the computer, the CPU is the BIOS.You boot a VM, program a dispatch chain of kernel instances, submit once with vkQueueSubmit, and everything — layer execution, inter-layer communication, self-regulation, compression, database queries — happens on the GPU without CPU round-trips. The CPU just provides I/O. let vm = vm_boot() let prog = vm_program(vm, kernels, 4) vm_write_register(vm, 0, 0, input) vm_execute(prog) let result = vm_read_register(vm, 3, 30) 4 VM instances, one submit, no CPU involvement between stages.The memory model is 5 SSBOs: Registers (per-VM working memory), Metrics (regulator signals), Globals (shared mutable — KV cache, DB tables), Control (indirect dispatch params), Heap (immutable bulk data — quantized weights).What makes it iFeb 22, 2026 6:55 PM
openfuse.io
Show HN: I built a fuse box for microserviceshttps://www.openfuse.ioHey HN! I'm Rodrigo, I run distributed systems across a few countries. I built Openfuse because of something that kept bugging me about how we all do circuit breakers.If you're running 20 instances of a service and Stripe starts returning 500s, each instance discovers that independently. Instance 1 trips its breaker after 5 failures. Instance 14 just got recycled and hasn't seen any yet. Instance 7 is in half-open, probing a service you already know is dead. For some window of time, part of your fleet is protecting itself and part of it is still hammering a dead dependency and timing out, and all you can do is watch.Libraries can't fix this. Opossum, Resilience4j, Polly are great at the pattern, but they make per-instance decisions with per-instance state. Your circuit breakers don't talk to each other.Openfuse is a centralized control plane. It aggregates failure metrics from every instance in your fleet and makes the trip decision based on the full picture. When the breaker opens, evFeb 18, 2026 2:04 PM
apps.apple.com
Show HN: Quick Issues: A Fast Mobile Issue Capture for GitHub, GitLab, and Giteahttps://apps.apple.com/us/app/quick-issues-fast-git-capture/id6758988655I got frustrated with how GitHub and GitLab's mobile apps handle issue creation. You need to be online, the flow is slow, and if you're on a self-hosted instance you're out of luck entirely.So I built Quick Issues. It's a lightweight Swift app with an offline buffer -- you capture the issue, it syncs when you have connectivity. Supports GitHub, GitLab, and Gitea/Forgejo (including self-hosted and local with PAT).A few technical notes for anyone interested: This was my first time using GRDB with SQLite instead of default Swift data structures, and the performance difference was significant. Setting up proper OAuth2 flows and GitHub/GitLab apps was more of an adventure than expected, but it's solid now.Free for a single account/instance, paid tier if you juggle multiple providers.My background is in GTD and data analytics, not traditional software engineering, so I'm genuinely curious: how does issue capture fit into your development workflow? Do you batch-create issues, capture them theFeb 17, 2026 2:50 PM
news.ycombinator.com
I built a free alternative to Datadog Synthetic Monitoring using Playwrighthttps://news.ycombinator.com/item?id=47045671Hi HN,I'm Vajid, the founder of a small dev agency.I built this tool because I got tired of the "200 OK" lie. We had a client whose e-commerce site was "Up" (returning 200 status codes), but the "Add to Cart" button was broken due to a silent JavaScript error. They lost sales for 6 hours while our dashboard said "All Systems Operational."Existing tools like Datadog Synthetic Monitoring are excellent but can be expensive for indie hackers or small startups (often ~$15/check).So I built a lightweight alternative using Playwright, Node.js, and BullMQ.How it works:It spins up a headless browser instance.It navigates to your URL and waits for specific DOM elements (not just HTTP status).It captures screenshots and console logs if the specific flow (like Login or Checkout) fails.The Business Model (Transparency): I am not trying to become the next Datadog. I run a dev agency, and this tool acts as a "loss leader" to demonstrate our competence to potential enterprise clients. That’s why the cFeb 17, 2026 10:11 AM
news.ycombinator.com
Ask HN: How does the cost of CPU-bound computations scale in cloud computing?https://news.ycombinator.com/item?id=47044151Let's say that I'm a rich mathematician (that would be nice) and want to sponsor some computation that takes a certain gargantuan number N of floating or integer operations, and it can be parallelized easily in the cloud, and my programmers have come up with a reasonably efficient implementation, which I benchmark with some smaller-scale tests. The I/O and storage costs are low; the bulk of the cost is going to be pure computation.Now I'm ready to get out my checkbook.What factors influence the cost of such a computation in today's cloud computing market?How does the price scale with N and the rate at which I want to complete the problem? (for example, if it takes 1000 AWS instances 10 months or 10,000 AWS instances 1 month, am I going to pay the same? I'm assuming the length of time to run this is less than a year, so not long enough that the cost would decrease significantly during the computation.)Are there aspects of flexibility that would decrease the price? (example: the cloud seFeb 17, 2026 5:48 AM
news.ycombinator.com
Show HN: Provisioner per-board sidecar for serial access, flashing, and bring-uphttps://news.ycombinator.com/item?id=47040176Hi HN,I’ve been building a small infrastructure tool called Provisioner: a per-board provisioning sidecar for SBC bring-up and lab automation. https://github.com/alessandrocarminati/provisionerThe model is simple: one instance per board. It sits next to the hardware, owns the serial console, exposes access over SSH, and serves boot artifacts (kernel, initramfs, rootfs) via TFTP/HTTP. It’s designed for the phase where networking is unreliable, storage is half-flashed, and UART is the only thing still telling the truth.What it does- Serial console ownership + SSH multiplexing - Monitor shell + serial tunnels - Scriptable boot interaction - Artifact provisioning (TFTP + HTTP) - In-band file transfer over serial - gzip/base64 hacks + XMODEM - Derived from my earlier send_console-ng utility https://github.com/alessandrocarminati/send_file-ng - Pluggable power control - SNMP PDUs, among others cheap ESP8266/Tasmota devices - Serial logging - Lightweight access controlProvisioner also embeds Feb 16, 2026 8:52 PM
github.com
Show HN: Foundry: Deploy observability without the complexityhttps://github.com/SigNoz/foundryHey folks, we at SigNoz have been working on something to make SigNoz deployment actually painless. Foundry takes care of all the deployment complexity – the configurations, tooling, everything needed to run SigNoz – so you can spend less time getting it up and running and more time actually using it. With just a single YAML config file and one command, you can have a fully operational SigNoz instance using Docker, Systemd, etc., across platforms or any cloud vendor. The concept: Like a metalworking foundry transforms raw materials into finished products, Foundry takes your config ("Castings"), forges it through deployment templates (the "Moldings"), and pours out a ready-to-run SigNoz instance.Feb 14, 2026 12:17 PM
news.ycombinator.com
Ask HN: How do you shut down misbehaving AI in production?https://news.ycombinator.com/item?id=47002748If you are running AI workloads/agents or LLM-backed systems in production, how do you actually shut one down when it starts behaving badly?By “misbehaving” I mean things like: -runaway spend -latency issues -prompt loops -tool abuse or unexpected external calls -data leakage risks -cascading failures across downstream servicesIn most systems I’ve seen, there is good observability. You can see logs, traces, cost dashboards. But the actual shutdown mechanism often ends up being manual: disable a feature flag, revoke an API key, roll back a deployment, rate limit something upstream.I am trying to understand what people are doing in practice.-What is your actual kill mechanism? -Is it bound to a model endpoint, an agent instance, a workflow, a Kubernetes workload, something else? -Is shutdown automated under certain conditions, or always human-approved? -What did you discover only after your first real incident?Concrete examples would be extremely helpful.Feb 13, 2026 1:57 PM
github.com
Show HN: Carapace – A security-hardened Rust alternative to OpenClawhttps://github.com/puremachinery/carapaceCarapace is an open-source personal AI assistant gateway written in Rust. It connects to Anthropic, OpenAI, Ollama, Gemini, and Bedrock, and works through Discord, Telegram, Signal, Slack, and webhooks. Apache-2.0 licensed.I started building it after the January 2026 OpenClaw security disclosures — 42K exposed instances on Shodan (78% still unpatched), 3 CVEs with public exploits, 341+ malicious skills on ClawHub (Snyk found 36% of all skills have security flaws), 1-click RCE via the Control UI, plaintext credentials harvestable by commodity infostealers. The problems weren't bugs; they were architecture decisions — open by default, no signing, full host privileges, secrets in JSON files. The February wave from Kaspersky, Palo Alto, Snyk, and SecurityScorecard made it worse, not better.Carapace takes the opposite defaults: localhost-only binding, fail-closed auth, OS keychain credential storage, Ed25519-signed WASM plugins with capability sandboxing, prompt guard with exec approval, SSFeb 12, 2026 3:13 AM
