Buy A Modem

Show HN: Real-time local TTS (31M params, 5.6x CPU, voice cloning, ONNX)

Hi guys and gals, I made a TTS model based on my highly upgraded VITS base, conditioned on external speaker embeddings (Resemble AI's Resemblyzer).The model, with ~31M parameters (ONNX), is tuned for latency and local inference, and comes already exported. I was trying to push the limits of what I could do with small, fast models. Runs 5.6x realtime on a server CPUIt supports voice cloning, voice blending (mix two or more speakers to make a new voice), the license is Apache 2.0 and it uses

Show HN: HF-agents, CLI extension to find the best model/quant for your hardware

We've been building out CLI extensions for the Hugging Face hub, and hf-agents is a fun one to share.It uses llmfit under the hood to profile your hardware and automatically select the best-fit model and quantization — no manual GGUF hunting. It then launches a Pi Agent on top of it. One command, local, fully open.If you've been using Claude Code or Codex CLI and want something that runs entirely on your own hardware/models, this is a nice lightweight alternative to try.Happy to a

Show HN: ClawMem – Open-source agent memory with SOTA local GPU retrieval

So I've been building ClawMem, an open-source context engine that gives AI coding agents persistent memory across sessions. It works with Claude Code (hooks + MCP) and OpenClaw (ContextEngine plugin + REST API), and both can share the same SQLite vault, so your CLI agent and your voice/chat agent build on the same memory without syncing anything.The retrieval architecture is a Frankenstein, which is pretty much always my process. I pulled the best parts from recent projects and researc

Launch HN: Canary (YC W26) – AI QA that understands your code

Hey HN! We're Aakash and Viswesh, and we're building Canary (https://www.runcanary.ai). We build AI agents that read your codebase, figure out what a pull request actually changed, and generate and execute tests for every affected user workflow.Aakash and I previously built AI coding tools at Windsurf, Cognition, and Google. AI tools were making every team faster at shipping, but nobody was testing real user behavior before merge. PRs got bigger, reviews still happened in fil

Show HN: How I built a resume editor using AI with zero web dev experience

Hi,I have recently been applying for summer internships and got frustrated when tailoring my resumes in Word. I started learning Python last autumn, but had absolutely zero experience with web development or deploying something to the front/backend. I wanted to experiment with the new coding agents to build a resume editor that would make my application process less painful.Here it is: www.tailortojob.appHow I built it: A friend helped me set up the initial infrastructure because I struggle

Show HN: Oo – compress output for coding agents (cargo test → "47 passed, 2.1s")

I've been running coding agents heavily for the past year or so using frontier model APIs, open weight model APIs and, most recently, local models (Qwen family models on a Strix Halo).Starting to run local inference has highlighted something I've been aware for longer: just running tests output shedloads of text into the context window that is there for good until compaction or starting afresh. For example, a single `cargo test` dumping 8KB into the agent's context just to communi

Where do we stand with Claude 20x Max vs. Codex Pro after Opus 1M context window

Has anyone tried the latest and greatest models of both camps, with the highest thinking level and maximum possible context window setting, and compared performances and observed patterns / specific behaviors which make you choose one over the other? [Of course, everyone's mileage varies, but still want to gather insights from folks who have the privilege to be able to use both extensively]I'm talking about $200 versions of both.I couldn't find any such detail over the web fo

Show HN: AI Roundtable – Let 200 models debate your question

Hey HN! After the Car Wash Test post got quite a big discussion going (400+ comments, https://news.ycombinator.com/item?id=47128138), I spent the past few weeks building a tool so anyone can run these kinds of questions and get structured results. No signup and free to use.You type a question, define answer options, pick up to 50 models at a time from a pool of 200+, and they all answer independently under identical conditions. No system prompt, structured output, same setup for e

Show HN: Sup AI, a confidence-weighted ensemble (52.15% on Humanity's Last Exam)

Hi HN. I'm Ken, a 20-year-old Stanford CS student. I built Sup AI.I started working on this because no single AI model is right all the time, but their errors don’t strongly correlate. In other words, models often make unique mistakes relative to other models. So I run multiple models in parallel and synthesize the outputs by weighting segments based on confidence. Low entropy in the output token probability distributions correlates with accuracy. High entropy is often where hallucinations

Using Catastrophic Forgetting as a Knowledge Topology Probe

I'm an undergrad with no research affiliation. I've been thinking about why LLM training is so expensive and why continuous learning remains unsolved. This post is where that thinking led — a concrete architecture proposal with a cheap falsifiable experiment at its core.The Core Idea (30 seconds) Catastrophic forgetting — when fine-tuning a model on new knowledge destroys old knowledge — is universally treated as a problem to minimize. I think it's a measurement instrument. The fo

Best Mesh Wi-Fi Routers for 2026

Find the best mesh Wi-Fi router system for your needs. Our team of experts evaluated mesh routers on a variety of criteria to determine the best option for any use case.

What to Look for in a Home Internet Plan in 2026

Choosing from the best home internet plans in 2026 is no longer just about picking the fastest option available. As technology evolves and households become more connected than ever, selecting the right home internet plan requires a deeper understanding of performance, reliability, pricing, and future readiness. With smart homes, remote work, streaming, and gaming all competing for bandwidth, your home internet needs to keep up without interruptions or hidden costs. This guide breaks down exactly what to look for so you can confidently choose a plan that fits your lifestyle and budget.

What is 1 gigabit internet? Is it worth the cost?

Gigabit internet (1 GB) has extremely fast download speeds of 1,000 Mbps. Although having the fastest internet possible might sound like a good idea, you don't want to overpay for speeds you don't ...

Valentines Day Love GIF by Calmlings

Show HN: We built AI agents that reduce mortgage processing from 18 days to 3–5

Most mortgage processing delays aren’t due to risk — they’re due to manual workflows.We’ve been working on SimplAI, an AI-driven system designed for banking and financial services, starting with mortgage operations.The problem we kept seeing:15–22 day processing timelinesHeavy manual document handling (500+ pages per loan)Repetitive data entry + verification loopsUnderwriters spending hours on non-decision workSo we built a set of AI agents that handle the operational layer:Document AI (IDP) → c

Show HN: We built a way to try 50 AI models from one API

Hey,We have been working on Qubrid AI — a platform to try and use 50+ AI models (text, vision, audio) from a single API.While building AI apps, we kept running into the same issue: switching between providers, APIs, and formats just to test different models. This slowed down iteration a lot, so we built a simpler way to experiment and compare.You can:- Run different models from one place - Compare outputs side-by-side - Use a unified API instead of multiple integrationsYou can try it here (playg

Show HN: Precise AI Motion Control for Kling 3.0

Hi HN,Like many of you, I've been experimenting with generative video, but the "camera lottery" was driving me crazy. Even with the best models, getting a consistent 360-degree orbit or a smooth crane shot felt more like luck than engineering.I built AI Motion Control to bring more determinism to the workflow. It's a specialized layer for AI motion control, specifically optimized for the Kling 3.0 architecture.Why this matters: The latest Kling AI motion control update (v3.0)

Show HN: OpenGranola – meeting copilot that searches your notes in real time

link: https://github.com/yazinsai/OpenGranolahey HN, I built OpenGranola — a macOS app that sits next to your calls, transcribes both sides of the conversation locally, and surfaces talking points from your own notes in real time.The idea came from having too many calls where I knew I had the perfect data point or quote somewhere in my notes, but couldn't find it fast enough. I wanted something that would do the retrieval for me, while the conversation is still happening

Show HN: Xybrid – run LLM and speech locally in your app (no back end, Rust)

Hi HN,We built Xybrid, a Rust library for running LLM + speech pipelines directly inside your app, no server, no daemon, just one binary.We started building it while working on a privacy-focused LLM app with Tauri and realized there wasn’t a straightforward way to embed models directly into shipped applications without relying on a separate server process.Xybrid links into your process like any other library. It supports GGUF / ONNX / CoreML and integrates with Flutter, Swift, Kotlin,

Ask HN: Are MiniMax Models Scams?

I kept trying to use their M2.5 model and now they released M2.7, but they are TERRIBLE.See this comparison I made:https://aibenchy.com/compare/minimax-minimax-m2-7-medium/minimax-minimax-m2-5-medium/z-ai-glm-5-medium/google-gemini-3-1-flash-lite-preview-medium/Not only that, but M2.5 is #1 on OpenRouter, which is crazy: https://openrouter.ai/rankingsI think the only reason why it is #1 is because it is a scam. In the comparison you can see