block - Moozonian Search

github.com• Jan 15, 2026• 1 min read

Show HN: I'm building an open-source AI agent runtime using Firecracker microVMsHello Hacker News! I'm Mark. I'm building Moru, an open-source runtime for AI agents that runs each session in an isolated Firecracker microVM. It started as a fork of E2B, and most of the low-level Firecracker runtime is still from upstream.It lets you run agent harnesses like Claude Code or Codex in the cloud, giving each session its own isolated microVM with filesystem and shell access.The repo is: https://github.com/moru-ai/moruEach VM is a snapshot of a Docker build. You define a Dockerfile, CPU, memory limits, and Moru runs the build inside a Firecracker VM, then pauses and saves the exact state: CPU, dirty memory pages, and changed filesystem blocks.When you spawn a new VM, it resumes from that template snapshot. Memory snapshot is lazy-loaded via userfaultfd, which helps sandboxes start within a second.Each VM runs on Firecracker with KVM isolation and a dedicated kernel. Network uses namespaces for isolation and iptables for access control.From outside, you talk to the VM thro

bing.com• Jan 14, 2026• 1 min read

Why LOVB Is Must See TV, Just Ask Amber IgiedeLOVB is seeing ongoing spikes in viewership—and for good reason. “Tune into LOVB because it is professional women's sports in the US!” LOVB Houston’s middle blocker Amber Igiede told E! News in an ...

news.ycombinator.com• Jan 13, 2026• 1 min read

Show HN: Verdic Guard – Deterministic guardrails to prevent LLM hallucinationsI’ve been working on Verdic Guard, a validation layer for production LLM systems where prompts, filters, and monitoring aren’t enough.In many real deployments (fintech, enterprise workflows, agentic systems), the failure mode isn’t latency or cost — it’s hallucinations that sound confident and pass surface checks. Prompt engineering helps, but it doesn’t scale once systems grow long-running, tool-using, or multi-agent.Verdic takes a different approach:Define an explicit intent + scope contract for what the model is allowed to outputValidate LLM outputs before execution, not just inputsBlock or flag responses that drift semantically, contextually, or domain-wiseKeep enforcement deterministic and auditable (not “best effort” prompts)It’s designed to sit between the LLM and your application, acting as a guardrail rather than another model.This is still early, and I’m especially interested in feedback on:Where this breaks down in real systemsHow teams currently handle hallucinations beyond

pypi.org• Jan 10, 2026• 1 min read

Show HN: FlowKitX – Fluent async workflows with lazy execution for PythonWhat is FlowKitX?FlowKitX is a Python library that makes writing async workflows more readable and structured.It provides lazy pipelines: operations (like API calls) are queued and executed only when awaited. This lets you chain calls and process data fluently without blocking the event loop.It’s not magic — it doesn’t convert blocking code to async or bypass asyncio fundamentals. It’s built for I/O-bound workflows, especially APIs and data pipelines.Why use it?Traditional async code can get messy:async def fetch_user_data(user_id): async with aiohttp.ClientSession() as session: async with session.get(f"https://api.example.com/users/{user_id}") as response: user = await response.json() async with session.get("https://api.example.com/posts", params={"userId": user_id}) as posts_response: posts = await posts_response.json() return {"user": user, "posts": posts}With FlowKitX:@flowkit.simple def fetch_user_data(user_id): user = flowkit.get(f"https://jsonplaceholder.typicode.com/users").jso

news.ycombinator.com• Jan 6, 2026• 1 min read

Show HN: Kafka-compatible streaming on S3 with stateless brokersHi HN,KafScale is Kafka-compatible streaming, k8s native, where S3 is the source of truth and brokers hold no persistent state. Written in Go, runs on Kubernetes.Built this after years of operating Kafka and hitting the same walls: broker failures that take hours to recover, partition rebalancing that blocks deploys, disk capacity planning that never ends.How it works:- Producers and consumers use standard Kafka clients - Brokers buffer in memory, flush to S3 - etcd stores metadata and consumer group state - Recovery means restarting a pod and reading from S3 - Optional Iceberg processor reads segments directly from S3, bypasses brokers entirely for batch/analytical workloadsWhat you give up: latency is 400-500ms (S3 round-trip), no transactions, no compacted topics. It's not a 100% replacement.What you get: brokers are disposable, scaling is just replica count, no disk management, direct access to streamed data over S3 ACLLicense: Apache 2.0 GitHub: https://github.com/novatechflow/kaf

bing.com• Jan 5, 2026• 1 min read

12 highly anticipated films coming out in 2026In 2026, cinema promises an explosive cocktail of franchise comebacks, sci-fi blockbusters, popular comedies, and XXL epics. Between superheroes, interstellar travel, and family adventures, studios ...

news.ycombinator.com• Jan 4, 2026• 1 min read

People who use only their laptop screens for work – How do you do it?Full stack developer, but currently working on backend only. During COVID I tried all sorts of setups, from triple 27" monitors to one 43" monitor, but for the past couple years I've been really trying to simplify things. My current setup is a single 27" 4K monitor and using MacBook's keyboard + trackpad with the built-in display disabled, and I close the lid partway so that it doesn't block the view of the monitor. I find that with Spaces and using third party apps for app switching, I'm able to work effectively with a single monitor.Every now and then, I try the experiment of working solely on my laptop, and I'd like to try again. Though I really only work from home or the office, I like the idea of being able to work from anywhere, not being tethered to a specific location or a specific room in the house. Whenever I do research about this I face the obvious reality that laptops are not ergonomic and this can't really be done ergonomically. But then there are posts from a small minor

authority.bhaviavelayudhan.com• Jan 3, 2026• 1 min read

Show HN: An authority gate for AI-generated customer communicationsMany teams now allow AI systems to draft customer-facing messages across support, CRM, and billing workflows.This introduces a specific failure mode: AI can generate text that constitutes an irreversible business commitment (refunds, credits, billing changes, contractual promises).Once emitted, the commitment exists. Detection after delivery is irrelevant.This project implements a hard authority boundary.Model -------- AI systems propose messages. They do not decide whether those messages are allowed to commit the company.A gateway enforces that decision.AI drafts message ↓ Authority Gateway ↓ Send | Block → ApprovalBehavior --------- For each outbound message:Inspect text for commitment signalsClassify outcome as reversible or irreversibleIf reversible → allowIf irreversible → block and require explicit approvalLog decision and evidenceNo attempt is made to assess advice quality, intent, or correctness. Only enforceability is considered.API surface ------------ /v1/messages/send Enfor

github.com• Dec 30, 2025• 1 min read

Show HN: SU_N an Adaptive Mesh Refinement EngineMaybe somebody out here has a need for this. It features:Linear octree storage with Norton ordered blocks! It's kinda like p4est. It even has runtime partitioning! It has an MPI backend! Turn that "home lab" into a "HPC". It supports gauge lattices! Model a nuke! In zig! Does it work? Download and find out!Tests and benchmarks will show you how to make a kernel, API usage, and some examples for running a model.

app.kensho.zone• Dec 28, 2025• 1 min read

Show HN: Terminal Aesthetic Productivity AppHey HN,I keep building things and failing at launch. This time I built something that enforces the discipline I lack: a productivity system that physically blocks you from working when you've done too much focus without rest.What it does: - Focus timer + meditation tracker + task manager in one terminal-aesthetic interface - Hard 10:1 focus:rest enforcement — work 100 minutes, you OWE 10 minutes of rest before the timer unlocks again - Task completion requires reflection logs (1 reflection per 5 tasks) — no backlog journaling allowed - ASCII progress bars, monospace everything, brackets for every action: `[ START_FOCUS ]` - No gamification, no streak anxiety, no "you can do it!" cheerleading Why I built it: I'd work 4-hour blocks, burn out by Wednesday, then doomscroll for 3 days. Every productivity app let me keep going. I needed something that would stop me and enforce recovery.The aesthetic is intentional — brutalist terminal UI to strip out dopamine triggers. Black background, func

bing.com• Dec 22, 2025• 1 min read

NAC protein complex slows early synthesis to optimize cellular protein productionProteins are among the most important molecular building blocks of life. They are chains of amino acids assembled in our cells by ribosomes, the molecular "protein factories" of our bodies. The ...

github.com• Dec 22, 2025• 1 min read

Show HN: SiteIQ–Automated security tests for LLM APIs(prompt inj,jailbreaks,DoS)Hi HN, I'm an 11th grader learning cybersecurity. I built SiteIQ, an open-source security testing tool that includes 36 automated tests specifically for LLM-powered APIs. Why this matters: Most security scanners focus on traditional web vulnerabilities (SQLi, XSS). But if you're shipping an LLM-powered feature, you need to test for prompt injection, jailbreaks, and LLM-specific DoS attacks. I couldn't find a good open-source tool for this, so I built one. What it tests: - Prompt Injection – Direct, indirect, RAG poisoning - Jailbreaks – DAN-style, persona continuation, "grandma exploit", fictional framing - Encoding Bypass – Base64, ROT13, nested encodings, custom ciphers - Refusal Suppression – Attacks that block the model from saying "I cannot" - Hallucination Induction – Tries to get fake library names/CVEs (package hallucination attacks) - ASCII Art Jailbreaks – Visual text that bypasses keyword filters - Recursive Prompt DoS – Quine-style prompts, Fibonacci expansion, tree generat

github.com• Dec 20, 2025• 1 min read

Show HN: Skill capsules" for LLMs, a "poor man's continual learning""Continual learning" is considered one of the "blockers" for LLMs: they can't learn on the job, don't improve over time, etc. In particular, Dwarkesh Patel describes it as a number of problem which has to be solved to get to AGI.Many academic article propose some kind of a memory system for LLM which might be considered a form of "continual learning". But most evals focus on memorizing facts which is just not very useful (it's better to fetch facts via tool use than to store it in neural memory) and these proposals might not fit well into common LLM API use patterns.In this article I'm proposing a "new" method called "skill capsules" which is highly pragmatic, easy to understand and evaluate and might integrate well into existing tooling.Skill capsule is a concrete object - it's a bunch of vectors, basically. You can insert it somewhere into a middle of LLM context and it improves performance on a particular skill, e.g. get tool calls more reliable, use particular writing style, coding

news.ycombinator.com• Dec 17, 2025• 1 min read

Google's AI quoted me verbatim 12 days laterI recently posted a comment[1] on Reddit identifying errata in a diagram Asus published for their Pro WS WRX90E-SAGE SE motherboard.Less than two weeks later, I was mildly surprised when I performed a Google search and found its AI quoting my words nearly verbatim, and presenting them with the confidence of fact: https://i.imgur.com/pgImAWh.pngI took the opportunity to play with search terms, to see just how generic a query could lead to a response lifted from my post. At the moment, googling "Asus Pro WS WRX90E-SAGE SE diagram has errata" does it [2], and to some extent even just "Does ASUS block diagram have errata" [3].In this case, the comment is (mostly) correct - in fact, ASUS is working on fixing the diagram.But what if I had made up the post, or worse, were a competitor trying to sabotage their reputation?I'm not sure whether I should be impressed or scared. I'm amazed how quickly one obscure comment can be surfaced to look authoritative. The AI further amplified my sentiments

bing.com• Dec 17, 2025• 1 min read

Henry James’s Venice Is Still HereIn a quiet, almost empty part of Venice stands a Renaissance palazzo with an unusually large garden. The garden is invisible from the outside, blocked by a high brick wall that I recognized when I saw ...

news.ycombinator.com• Dec 3, 2025• 1 min read

Show HN: YapYap – Post to X/LinkedIn without seeing the timeline (macOS)Building in public doesn't come naturally to me. I overthink, stress about engagement, get lost in timelines. Tried schedulers, web blockers, willpower - nothing stuck.So I built YapYap, a macOS menu bar app with global keyboard shortcuts: - Option+X → X compose window (no feed) - Option+L → LinkedIn compose window (no feed) - Cmd+Enter to post, and back to workTechnical approach: It uses a webview to load the native compose UI. No API calls (avoiding X's $200/month pricing), no data stored on my servers. Your credentials stay in the app's sandboxed session.The webview detects when you navigate away from compose (after posting) and auto-closes. Next time you press the shortcut, it's already loaded and ready.Stack: - Electron + Vite - Global shortcuts via Electron's globalShortcut API - Persistent sessions so you stay logged in - Code signed + notarized for macOS - Auto-updates via electron-updater€5 one-time purchase, macOS only.https://yapyap.dev?ref=hn

news.ycombinator.com• Dec 1, 2025• 1 min read

Teaching LLMs to compose math symbolically, not execute itRight now LLMs cannot be counted on to perfectly perform math. The solution I propose is to teach LLMs to instead of executing the math, just compose the mathematical equations correctly and leave the execution to a post-processing step.My core method would be: Use a single special token ᶜ (U+1D9C) before each element that needs computation, and then compute the result afterwards. For known math that doesn't need to be computed the ᶜ is not added or present.Thus we would see in the output:Normal (already computed): 847 * 293 = 248171Requesting computation: ᶜ847 ᶜ* ᶜ293 ᶜ= ᶜxThe Core Mechanic: Post-Process ComputationThis is what makes everything work: Model generates output with ᶜ-marked expressions (fast, no blocking) Generation completes Parse all ᶜ-marked expressions Execute computations with perfect precision Substitute results back into the output Show user the final result with normal mathematical notationThe model never waits for computation results. It reasons symbolically with

bing.com• Dec 1, 2025• 1 min read

Will's Game-Changing 'Stranger Things' Reveal Has an Even Deeper Meaning for the Rest of the ShowMichael Block is a 14 time GLAM Award nominated writer, producer, and host of the podcast Block Talk. Throughout his time in the entertainment industry, he has worked on and off Broadway as a stage ...

news.ycombinator.com• Nov 30, 2025• 1 min read

A three-layer memory architecture for long-running agentsAnthropic's recent piece on effective harnesses for long-running agents hit close to home. We've been wrestling with the same problems — agents that try to one-shot everything, declare victory prematurely, and leave chaos for the next session to clean up. But we solved some of these problems differently. Here's what's working for us, what isn't yet, and where we respectfully disagree with the proposed solutions.The Memory Problem: Three Layers Beat One Anthropic's solution is a progress.txt file plus git history. It works, but it's flat. We use three layers instead: Layer 1: Model actualization. A semantic memory system that helps the orchestrating agent understand "what are we building and why." This is the soft layer. Layer 2: Think Jira meets Git, but for AI agents. Structured storage of tasks with metadata: blockers, decision paths, dependencies, progress state. The agent doesn't just know what to do next — it understands the logic of how we got here and where we're going. Layer 3:

github.com• Nov 30, 2025• 1 min read

Show HN: Supabase Testcontainers for RustI utilize Supabase as the basis for many of the applications that I create. I find it valuable to pursue a test driven development strategy for both my organic coding as well as my AI assisted coding. This makes me lean on Testcontainers often. While there are Testcontainers available for many of the foundational building block services we all use, there was not one for Supabase (at least in Rust). So I created one. Hope this helps someone out there. Thanks.Full disclosure I also use AI to assist me in doing this in a timely manner. If you find any issues or broken code, please submit an issue. I've added tests, but sometimes things fall through the cracks. Thank you.