Look today34
2026-09
S mcp-extensions OpenAI's officially open-sourced ChatGPT plugin development kit, bringing MCP capabilities into ChatGPT as native features—a platform-level entry point. With 208 stars in a day and OpenAI's backing, it will almost certainly become a de facto standard dependency for the ChatGPT plugin ecosystem, with clear monetization (plugin marketplace, API billing). The risk is ecosystem policy changes and cross-vendor standard fragmentation.
Build plugins that feel like native, first-class features of ChatGPT.
S Show HN: Raven – The harness of harnesses, built for RSI Raven calls itself a 'harness of harnesses,' building a self-evolving multi-agent ecosystem to support RSI (recursive self-improvement), with 4,807 stars and 53 HN points—one of the strongest early momentum in this batch. If the RSI narrative pans out, it could become the infrastructure layer for agent orchestration. The risk is that the concept is ahead of its time, engineering feasibility is doubtful, and it competes head-on with mature orchestration frameworks like LangGraph.
S Show HN: HN.watch – Videos of all Hacker News posts Turns any article into an HTML-rendered explainer video in seconds at about $0.04 each, pushing video from dollars-per-minute to cents-per-second and creating a new content cost structure. 118 points plus MCP, Chrome extension and ChatGPT plugin distribution validate demand, and Scrimba is a mature YC-backed company that could sell B2B PR/documentation video subscriptions; the risk is rough visuals and easy copying by Loom, Canva and other large players.
Hi HN, I’m Per, founder of Scrimba (YC S20). We’ve spent the last decade teaching people how to code with an HTML-based video format. We’ve now plugged an LLM into it, so that people can create explai…
S PrunedCTC Cuts the activation memory of CTC loss from growing linearly with vocabulary size while keeping the loss and first-order gradients equivalent, directly benefiting large-vocabulary speech recognition training; the claim is small and the description accurate. The limit is that conclusions come from a single setting with no empirical comparison against mainstream CTC implementations.
Memory-efficient CTC loss with exact loss and first-order gradient equivalence under vocabulary reduction, and activation memory that no longer scales linearly with vocabulary size.
S jevgrep Provides semantic code retrieval for coding agents, recalling files and context by what the code does rather than keywords, targeting wasted agent context windows. 743 stars in 1.8 days is the strongest signal in this batch, showing rapid adoption of agent tooling; commercialization could be an open-source CLI plus enterprise internal-index subscription, or being built into agent frameworks as a standard dependency. The risk is Sourcegraph, grep.app and IDE-native semantic search already occupying the space, and reliance on the Jev model's popularity.
Find code by asking what it does. A CLI for coding agents that uses Jev to discover relevant files and source context.
S Show HN: Reladraw – A diagram language where you decide where to place things Fills the gap between Mermaid/Graphviz's lack of layout control and Draw.io's inefficient manual dragging: textual but position-controllable, and designed for agent editing. 232 stars in two days plus 151 HN points is the strongest signal in this batch, and the playground and Claude skill show real adoption. The business path is a design tool and standard entry point for agent drawing; the risk is Mermaid ecosystem inertia and Visio-like vendors following.
I love making diagrams to help understand, plan, etc. However, the options are (A) auto-placement languages like Mermaid or Graphviz (which don't let me decide how the diagram looks), or (B) software …
S HaloBattery Shows battery levels for Logitech, Razer and other wireless mice, headsets and Bluetooth devices in one Windows tray, solving a very specific small annoyance. 170 stars in two days shows the pain is real; limits are Windows-only support and vendor protocol coverage you must read the README to learn.
Unified battery level indicators for wireless mice, headsets and Bluetooth devices in the Windows tray
S Show HN: Whiteboard (YC W26) – An open-source IDE for thoughtful software design A CodeOSS-based desktop whiteboard where agents link architecture diagrams, sequence diagrams and code locations bidirectionally via SDK and Rust semantic diff folds noise, already used by Salesforce and Modal, hitting the cognitive debt of agentic coding. YC W26, MIT open source and a clear paid hosted path, with the risk of competing head-on with Cursor and Claude Code's built-in review.
Hello! We’re Sid, Alex, Ketan, and Milan. We’re building Whiteboard ( https://whiteboard.dev.fast/), an open-source desktop app where humans and agents can architect software together in a common work…
S golive-skill Packages the dirtiest part of shipping agent-written code into an Agent Skill: domains, databases, email and payments all through the user's own accounts, with a zero-dependency CLI doing a reviewable detect→plan→apply flow. 874 stars in 1.4 days shows the ecosystem lacks this last mile, but Vercel and Cloudflare official Skills and built-in deployment directly cover it, making charging hard.
Take your agent-built product live: hosting, database, domain, email, payments — on your own accounts. Open-source Agent Skill + zero-dependency Node CLI: detect → plan → approve → apply → verify. No …
S Show HN: An open-source manufacturing ERP/MES/QMS An open-source self-hosted manufacturing ERP+MES+QMS suite filling the gap for small factories bound by expensive closed SAP and Oracle systems; open-core and per-line licensing is a mature industrial-software path, and 51 points shows real demand. Risks are heavy implementation, consultant-dependent delivery, and thin compliance modules being squeezed by existing MES vendors.
S jev-chat-jarvis Turns screen reading, intent judgment and candidate replies into a conversation copilot on the phone, solving the pain of not knowing how to reply in WeChat and Feishu, with no injection or packet interception reducing ban concerns; 3,723 stars in 1.5 days comes from strong话题 appeal. Monetization could be a subscription input-method-style entry point. Biggest risks: platform ToS and privacy compliance, and suspected star farming among sibling repos undermining credibility.
A conversation copilot installed on your phone: understand the other person in WeChat / QQ / X / Feishu, get candidate replies, and fill them into the input box with one tap — you decide whether to se…
S Show HN: Foremerge – Catch intent conflicts between parallel coding agents Builds on git with intent and scope declarations to predict conflicts when multiple agents modify code in parallel, moving semantic conflicts from the PR stage to before writing code, with built-in MCP and acceptance gates. Multi-agent parallel work in one repo is a real pain point, and 451 stars plus 33 points shows engineering-circle resonance, with a chance to become a collaboration protocol standard; the risk is git vendors or Cursor building coordination in.
At, GPTree, we run several coding agents across our team on one repo using parallel worktrees. Apart from wasted time reviewing and fixing conflicts at PR time, the failures that hurt the most are whe…
S preflight A local proxy that scans LLM inputs and attachments for secrets before requests go out, addressing the frequent incident of employees pasting credentials into AI and leaking them. The author is influential and enterprise AI security budgets are clear, with a chance to become a DLP standard layer; competitors are vendors' built-in privacy filters and security gateways like Zscaler.
A local proxy that scans LLM requests and attachments for secrets before they reach the model.
S Show HN: CUA-S1 – A System One Model for Computer Use Trains a 700K-parameter small model for the many local decisions in computer-use (checking boxes, skipping, filling values), pulling System1-specific judgment out of the general LLM loop at 7ms versus 260ms; backed by the 24K-star cua, open-source drivers and benchmarks, and HN's 56 points show agent vendors are genuinely watching; the risk is that jev-style services become the standard first.
Hello HN! We're Dillon and Francesco from Cua. We were wondering how many computer use tasks actually need a full general purpose LLM (e.g. gpt-6-astra, claude-opus-5 etc.) to think through all their …
S Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash Compresses tool calling and structured JSON output into 8–29MB 2-bit models that hit 4k tok/s on Raspberry Pi and browser WASM, cutting edge-agent deployment costs. The third iteration earned 156 votes, showing the team keeps delivering on feedback. It could license an edge inference SDK. Risk: narrow capability, squeezed by edge runtimes like llama.cpp and MLX.
Hey HN, Henry from Cactus here. We submitted Needle 2 here a few weeks ago, and the feedback in the discussion thread was incredibly valuable, thanks! Thanks to all that feedback, we’ve been able to m…
S Show HN: Swift-Qwen3.8-27B, -58.3% thinking, x1.95 speed, accuracy of xhigh Post-training Qwen 27B with on-policy distillation that penalizes tokens tied to overthinking halves think length with almost no accuracy loss, directly cutting inference cost. It drew 80k downloads in 3 days plus community quantization and independent evals, strong momentum; monetization is API and licensing (only charging companies above $10M), but the risk is that base models' built-in effort control will absorb this value.
Hi everybody, we post-trained Qwen 3.8 27B to be more efficient by figuring out which tokens were linked to overthinking and penalizing them without "attacking" the reasoning length directly then fixe…
S knoxmap It converts real locations into playable maps for Project Zomboid B42 and explicitly credits upstream Knoxify; a rare genuine need for PZ modders and players who want to fight zombies outside their home, limited by dependence on the upstream toolchain and output quality that varies with geographic data.
KnoxMap: turn any real place into a playable Project Zomboid Build 42 map (built on Knoxify)
S Show HN: Capsule – Single-file web apps that save their data into SQLite It packages HTML apps, assets and SQLite data into a single distributable file, solving the long-standing pain of static pages having nowhere to store data, with a permission model and local AI. 267 upvotes is the strongest momentum this period, and the 1.0 open format could become a standard; commercialization could be an app marketplace and enterprise distribution, with the risk of WASM sandbox alternatives.
Hey HN, I always had the problem that building HTML pages is really simple now, but trying to save data required hosting it somewhere, and sharing it afterwards was not easy. Over the last few months,…
S DeepSelect DeepSeek's official open-source sparse-attention TopK and sampling CUDA kernels directly determine DSA inference compute and memory costs, so any team reproducing or deploying the model needs them. Official backing could make it a de facto standard absorbed by frameworks like vLLM, though it only fits their own architecture.
DeepSelect: TopK kernels for DeepSeek Sparse Attention (DSA) and Samplers
S maskit A local gateway that masks requests before they reach the LLM and restores them in streaming replies, pluggable into any tool that lets you change the Base URL; one of the few privacy tools that works right after install, limited by rule-based masking that still misses complex entities.
Data Maskit — a local privacy masking gateway built for large models (requests automatically masked, responses streamed back restored; supports any tool with a configurable Base URL such as Cursor / C…
S Show HN: Geiger – See every AI agent on your machine and what it can touch Addresses the difficulty of tracking proliferating AI agents with a read-only command that inventories agents, MCP servers and plugins. Its 44 HN points and 68 stars show strong security-community interest; commercialization means part of a security audit or asset management tool or a SaaS offering, but the AI ecosystem changes fast and operating systems may build similar features.
S Show HN: Self-hosted company OS, Claude Code and Codex agents in departments A self-hosted multi-tenant AI agent platform where teams collaborate with Claude Code/Codex inside sandboxes. It gained 95 stars in 0.3 days and 37 HN points, strong momentum; commercialization resembles a company OS with team subscriptions or private deployment, but competition is fierce and dependence on Anthropic/OpenAI subscriptions exposes it to platform policy.
Hi HN, I am Dimitris. This is a company OS that I built and use to run my business and anyone can install it and self host it for free. Think of it as Claude Code, Cowork and the cloud sessions in one…
S Show HN: Give your AI agent on-screen guides that show users where to click Solves AI agents not understanding their own product UI by using a browser agent to learn app structure and generate on-screen guidance. Its 17 HN points show interest, from a YC startup; commercialization means selling an assist API as SaaS billed per call, but RAG keeps improving, enterprises may self-train, and browser-agent security and maintenance costs matter.
Hey HN. I'm Christian, one of the founders of Frigade (YC W23). I've noticed that a lot of in-app AI agents struggle to actually understand the products they exist in. For instance, let's say a user a…
S glooprint An Unreal Engine plugin that auto-lays out and rewires Blueprint node graphs with one keypress, tidying the whole graph. Messy Blueprints are a daily pain for UE developers, so the angle is small and precise; limits are it only serves Blueprints, and stability needs testing in the editor.
An Unreal Engine plugin for automatic Blueprint graph formatting and clean/performant wire routing. Press F to tidy up your entire graph.
S Show HN: Copperhead – Cursor for circuit boards This uses AI to assist circuit board design, like Cursor but for hardware, and 218 Hacker News points show hardware engineers want it. Commercialization could be a subscription PCB design tool with a large market; risks include competition from EDA giants, and competitors are Altium and JLC's AI features.
S Show HN: Sparrow-2 – Noise cancellation isn't designed for conversational AI It addresses turn-taking and conversational flow understanding for voice AI in noisy environments, modeling non-verbal signals such as breathing and ambient sound; with 11 HN points and a claimed Turnbench first place, it comes from the professional Tavus team. It can commercialize directly as an API or model embedded in customer service and voice assistants; the risk is competitors like OpenAI Realtime and Deepgram plus inference costs.
Hey there, I’m Brian. I've been shipping conversational models at Tavus for the past two years. I want to tell you about our latest audio-understanding/turn-taking model: Sparrow-2! It’s a new categor…
S okf-agent-memory This gives AI coding agents Git-native persistent memory with OKF v0.2 and sub-millisecond BM25 retrieval plus an embedded MCP server; 348 stars per day show it hits the agent memory pain point. It could become a standard dependency for tools like Claude Code, but competes with Mem0 and Letta, and the OKF standard is unsettled, so watch the ecosystem.
Git-native persistent memory for AI coding agents. Implements Google OKF v0.2 with sub-300µs in-memory BM25 search, embedded MCP server, and progressive disclosure. Slashes token bloat by 80% with zer…
S Show HN: ZSvirt – A lightweight, scalable open source virtualization platform A lightweight, extensible open-source virtualization platform that replaces heavyweight options for SMB IaaS needs. It hit the Hacker News front page with 71 points and 1563 stars, showing strong community approval; private cloud or cloud-provider integration is a clear path, but the architecture is complex and KVM/Proxmox competition is fierce.
S commerce-agents A reference blueprint for building shopping and merchant agents with Claude, covering retail and e-commerce scenarios. Anthropic's official backing plus a Shopify example brought 300+ stars fast; it could become a standard dependency for custom enterprise agents, but early versions are unstable and LangGraph-style frameworks compete.
Reference blueprint for building shopping and merchant agents with Claude. Examples in retail, commerce, telecom, and entertainment included.
S claude-for-commerce-examples Shopify's official implementation of the Commerce blueprint, providing reference code for store shopping and merchant agents, deeply tied to the Shopify ecosystem and practical for enterprises. Its 34 stars are few but growing well, monetization is clear as a standard Shopify app reference, and risks are dependence on Shopify APIs and Anthropic model evolution.
Shopify implementations of Anthropic's commerce-agents blueprint: a storefront shopping agent over UCP and Sign in with Shop, and a merchant agent over the Admin API
S Show HN: Aura – a Rust agent that investigates and fixes production incidents A Rust-written SRE agent platform that investigates and fixes production incidents, with human-in-the-loop and permission controls. It is validated in real production and tackles context overflow and approval fatigue; strong DevOps automation demand and Rust performance make an enterprise SRE product a clear path, but market entry is slow and PagerDuty-style tools compete.
We run a SaaS that handles petabytes of data. Our SRE team experimented with using claude, openclaw, langchain, etc. within our incident response workflows. We struggled with overflowing context, leth…
S agent-fleet-manager A general engine for large-scale repeated information gathering that manages multiple worker agents collecting in parallel. Its 168 stars and growth show real demand, with a clear path to SaaS or internal enterprise data platforms, but it competes with other agent orchestration frameworks and must solve data quality and anti-scraping compliance.
A general-purpose engine for large-scale, repeated information gathering by a fleet of workers.
S data-deidentification A pure front-end file de-identification tool that locally detects and masks personal data in PDF/Word/Excel for review, then restores it via a CSV code table, with data never leaving the machine. Practical for HR, medical and legal scenarios that can't use the cloud; limits are detection rule coverage and front-end performance on large files.
A pure front-end file de-identification tool: automatically detects personal data in PDF/Word/Excel/TXT/Markdown, mask preview, manual review, restorable via CSV encoding table, data never leaves the …
S Show HN: 49 IDE – 2D Canvas for Agents It solves context fragmentation across parallel AI CLIs, using a 2D canvas to unify terminals, git, and issues. A hot HN post plus 411 stars shows it hits a real pain point. It could become a standard AI development workbench, commercialized via subscription. The risk is a fragmented AI CLI ecosystem, with competitors like Cursor and terminal multiplexers.
Track agentic CLIs across multiple providers and repos, from any device and across multiple machines, see git trees, terminals, usage, issues, and UI all on one workspace. I have been trying to parall…