... Quick build-in-public update from Sponkly (real-time voice translation for chat, calls and meetings in 32 languages). This week: NVIDIA Inception: accepted AssemblyAI Startup Program: accepted AWS Activate: $10,000 in credits issued today What it changed right away: we moved Video Studio's speech recognition to AssemblyAI and raised the video limit from 5 minutes to 1 hour. A 30-minute video is now transcribed in ~10 seconds and is ready with subtitles in under 2 minutes ... ... usage grows. Tip for other bootstrapped founders: AWS rejected my first two Founders-tier applications (the first one because my company email domain did not match my AWS account email). What finally worked was applying through a partner program (NVIDIA

... excited to announce that AtaForge Inc., the company behind Sponkly, has officially been accepted into the NVIDIA Inception Program! A new milestone in our journey to break language barriers with AI-powered real-time communication. Thank you, NVIDIA

... Tejas, a Computer Engineering student at Waterloo and a research engineer at NVIDIA, where I work on post-training: fine-tuning, synthetic data, and agents. A couple of highlights: AEVAL (ICML 2026): I led our work on making AI agents stop grading their own homework, so agent testing becomes deterministic. Data Efficient Fine Tuning: I worked on a loop that finds where a vision model fails and fixes it with targeted real and synthetic data. It matches full-data fine ... ... tuning on a fraction of the data, and it s open source in NVIDIA

... Quick build-in-public update from Sponkly (real-time voice translation for chat, calls and meetings in 32 languages). This week, as a solo bootstrapped founder: NVIDIA Inception: accepted AssemblyAI Startup Program: accepted AWS Activate: $10,000 in credits LiveKit Startup Program: approved today, $17,000 in credits + a year of the Scale plan What I will use it for: Sponkly's calls and meetings are peer to peer today (WebRTC mesh). Great for 1:1, heavy beyond ...

... week for Sponkly: right after NVIDIA Inception, we have been accepted into the AssemblyAI Startup Program . Every Sponkly feature starts with speech recognition. Before a call, a meeting or a face-to-face conversation can be translated, the app has to understand exactly what was said, in any accent and in real time. With AssemblyAI's support we will make Sponkly's live voice translation faster and more accurate, from live calls to meeting transcripts and video subtitles. Thank ...

... building apps and games with AI agents on your own computer. Here's what changed: Parallel agents are back in the chat. You can follow each worker's progress while they work on the same project. The OpenRouter and NVIDIA model picker now separates models verified with your account from those still pending a check or unavailable. Availability is checked again every six hours. V-CORE can bring Luanti mods from ContentDB into the game workflow. This release also adds ...

... attended a VoiceAI event held by NVIDIA at their Bengaluru, India office. It was very interesting to see all the startups that presented. Adi Margolin , Product Manager at NVIDIA

... shipped four releases (1.4.8 to 1.5.1). The biggest changes: - Bring your plan, not just your keys. Sign in with your Claude or ChatGPT account and the agents run on the subscription you already pay for. API keys, OpenRouter, NVIDIA and Ollama still work, with zero markup. - Ollama installs itself. Pick a local model and Airuncode sets up and manages the server for you. - Pull requests inside the app. CI status, checks, reviews and diffs, then checkout or merge. - One-click ...

... agents, device collaboration, security and new ways for computers and phones to share capabilities. GPU Borrow demonstrated one of those ideas: A device doesn't necessarily need to own a capability to temporarily use it. We demonstrated a real NVIDIA RTX 3050 Laptop GPU being exposed through Blink Bridge, discovered and requested by an AI orchestrator, explicitly authorized by the user, used for a bounded workload, verified, and then released. Since then, our experiments have gone further. Android ... ... demonstrated an Android device participating in a workflow where a data block travelled through Blink Bridge, was transformed using an NVIDIA

... about or indexed Kanverse GPU Borrow and Kanverse ProofStamp thanks to Product Hunt The part of GPU Borrow that matters most to me is the control model: Discover Ask Permission Borrow Execute Verify Release The demo uses a real NVIDIA RTX 3050 Laptop GPU across devices, with explicit human authorization before a bounded GPU workload is allowed to run. But the idea is not simply remote GPU access. It is part of a broader principle we are exploring at Kanverse ...

... When I was reading TechCrunch today, I came across Jensen Huang s (NVIDIA CEO) view that AI safety should be handled by companies through engineering and market pressure, rather than through new regulation. In my opinion, the power of consumers and the public is sometimes simply not strong enough to counterbalance large technology companies. That s why I believe governments sometimes need to step in and establish rules and regulations. Do you think AI should be regulated by governments ... ... move money, send emails, touch production systems) than rules scoped to the model itself. That also happens to be the layer where a small team can actually build good defaults - permission scopes, human approval steps, audit logs - versus

... deal is done. NVIDIA has officially agreed to acquire Hugging Face, the company known for its GitHub-like repository of open-source AI models, for $12.9 billion. What it means: NVIDIA

... agent that lives in your terminal instead of a browser tab. The idea started from being tired of switching between chat windows and my editor. So loomcode plugs into whatever model you already pay for Anthropic, OpenAI, Gemini, OpenRouter, NVIDIA NIM (free tier), or local Ollama and you can switch models mid-session with /models. No lock-in, your own keys. The part I'm most unsure about (and would love feedback on) is Loom Computer , a beta feature where ...

... between retrieval/evidence and the final answer and checks whether the available evidence actually supports the claims being produced. If support is insufficient, it can fail closed. I recently ran the same frozen black-box protocol against: AletheionAGI Guardrails AI NVIDIA NeMo Guardrails Patronus Lynx Final observed results: AletheionAGI 9 PASS 0 FAIL 1 SKIPPED 0 ERROR Guardrails AI 4 PASS 2 FAIL 4 SKIPPED 0 ERROR NVIDIA

... Moonshot : Kimi K3 2.8T params, largest open weights ever (Jul 16) ByteDance : Seedance 2.5 (Aug 3) MiniMax : H3 multimodal (Aug 3) Huawei : openPangu-2.0-Pro 505B total / 18B active, open source (Aug 1) Meituan : LongCat 2.0 (Jul 20) NVIDIA : Nemotron TwoTower (Jul 1) poolside : Laguna S 2.1 (Jul 21) thinkingmachines : Inkling (Jul 17), Inkling Small (Jul 30) Ant Group : Ling-3.0-flash (Jul 23) Kuaishou : KAT-Coder-Air V2.5 / Pro V2.5 (Jul 10) SenseTime : SenseNova U1.5 (Aug 3) That ...

... genuinely excited to share this one. Intrascope is now a member of the @NVIDIA Inception program. We started building Intrascope because our own team needed a better way to use generative AI across the company without losing control over access, context, models, usage, and costs. What began as an internal need has grown into a platform that helps teams connect leading AI providers, manage permissions, define shared company and project context, and understand how AI is being used across ... ... organization. Joining the NVIDIA

... pledged with Nvidia-backed Firebird to build "Data Center Valley" 100,000 GPUs, live by 2027, part of Kazakhstan's push to become a regional AI hub. Thoughts anyone? I'm personally a little confused about why they chose Kazakhstan as the new data center capital ...

... Wired or Integrating: FastAPI, React, Three.js, WebGL, Tailwind, Socket.io, Docker, Kubernetes, Terraform, Nginx, Redis, PostgreSQL, SQLite, Figma, Notion, Linear, Jira, Slack. COMPANIES & LABS WE WANT AT THE TABLE: OpenAI, Anthropic, Google DeepMind, Microsoft, Meta AI, xAI, Moonshot AI, NVIDIA, Mistral AI, Cohere, Stability AI, Midjourney, Runway, ElevenLabs, Replicate, Hugging Face, LangChain, Vercel, Cloudflare, Stripe, Shopify, Notion, Linear, Figma. FOR CONTRIBUTORS: Verified dev badge. Tiered system access. Fair credit/revenue distribution. Owners, devs, superusers, archival users everyone gets a seat. Distributed ...

... this problem. After testing similar apps for Windows, I realized that the goal of my app is maximum ease of use (modelled after Google s visual translation on Pixel devices). The app works as an overlay (similar to Nvidia Overlay) that sits in the system tray and can be launched with a hotkey. The architecture is designed so that you can change most settings in real time with minimal effort. Currently, the following core features are available: Overlay mode Text ...

... companies are not catching up with creating sustainable revenue, they rather burn VC money in most cases. So it is... a huge risk here. Comment from Sebastian(@sebxdev): I've been investing for a year now, I've google, nvidia and more but my favorite so far is: - nebius group Comment from Sunny K. Singh(@sunny_k_singh): Nvidia

... planning, from idea to celebration. Plan your occasion, discover vendors, manage your budget, coordinate tasks, and bring everything together in one place. www.tendr.co.in Comment from Richie Varghese(@ric_v36): What you're building: Building NexQL, a PostgreSQL extension for VS Code database explorer, SQL notebooks, real-time dashboard, and an AI assistant, all without leaving the editor. Who it's for: developers and small teams running Postgres who want AI help writing/optimizing queries but don't want an agent with ... ... here is the specific thing' isnt something people have practice writing. we're prototyping prompts that lower the barrier without turning it into a form. Comment from Mustafa Arian(@thenameisarian): What you're building TAM Network. the alternative

... annual budget. Their COO admitted they can t justify the expense. Microsoft has revoked AI licenses from its own developers to control costs. Starbucks scrapped its AI inventory system after 9 months because it performed worse than human employees. NVIDIA s VP recently said that AI is costing more than human workers. So the question is (are): Will companies eventually find a way to significantly reduce the cost of AI? Or will we see a partial return to old-school ... ... reduce the cost and go old school as well. A balance that will often tip either way. Comment from Nika(@busmark_w_nika): @jiaqi_lu  Do you have any tracking system that estimates what is cheaper? (human vs

... deploy. Paste a model ID, it scans your GPU, VRAM, OS, and serving libraries, gives you a Hardware Ready verdict and full memory breakdown (weights + KV cache + serving overhead). No surprises at runtime. Supports Apple Silicon (M1 to M5), NVIDIA (CUDA), AMD (ROCm), and serving libraries including Ollama, llama.cpp, SGLang, TensorRT-LLM, PyTorch and many more coming. It tells you why something won't work, not just that it won't. CPU support is something we're actively working through ... ... oneInfer-edge is part of the broader oneinfer.ai inference control plane, a platform built for teams shipping multimodal AI products at scale. oneInfer-edge brings that same infrastructure intelligence to your local machine so self-hosting is a genuine alternative

... deploy. Paste a model ID, it scans your GPU, VRAM, OS, and serving libraries, gives you a Hardware Ready verdict and full memory breakdown (weights + KV cache + serving overhead). No surprises at runtime. Supports Apple Silicon (M1 to M5), NVIDIA (CUDA), AMD (ROCm), and serving libraries including Ollama, llama.cpp, SGLang, TensorRT-LLM, PyTorch and many more coming. It tells you why something won't work, not just that it won't. CPU support is something we're actively working through ... ... oneinfer-edge is part of the broader oneinfer.ai inference control plane, a platform built for teams shipping multimodal AI products at scale. oneinfer-edge brings that same infrastructure intelligence to your local machine so self-hosting is a genuine alternative

... mostly about making local coding-agent setups less fragile. What changed: - Reworked the Launch / Profile flow so picking agent + profile + workspace + session is faster - Better bridge routing and model metadata for Codex / Claude-style launches - New provider profiles: xAI, NVIDIA NIM, and Mimo - Pi profile launches and direct Pi session support - Agent enable controls, bridge proxy controls, and an in-app release update indicator - Web Chat now surfaces context-compaction notices more clearly If you use multiple coding agents ...