QuickRead

01

✨ Try It Tonight ⬡ GitHub trending HEAT 77

nexu-io/html-video — Programmatic video for coding agents — HTML to video on your laptop. Turn HTML, CSS & data into real MP4s with pluggable render engines, 21 templates, AI soundtrack. Apache-2.0, no per-render fees. An official project by the Open Design team.

Open-source tool lets coding agents generate MP4s from HTML/CSS locally—no per-render fees, 21 templates included.

👉 The take Ditch expensive video tools; have your agent whip up explainer vids on your own laptop tonight.

02

🌍 Big Picture ◆ Financial Times 18h ago HEAT 68

OpenAI plots biggest ChatGPT overhaul since launch - Financial Times

OpenAI is reportedly working on its biggest ChatGPT redesign since launch, hinting at a super-app shift.

👉 The take If you live in ChatGPT, brace for a UI that might actually make sense—or break your muscle memory.

source 1 source 2 source 3 source 4

03

💸 Save Money ▲ Hacker News 1h ago HEAT 66

Show HN: Claude Code on Slack/Discord/Telegram for flat $20/mo – no API bills

Claude Code now on Slack, Discord, and Telegram for a flat $20/month—no surprise API bills.

👉 The take Ditch per-token anxiety; pay one flat fee and let Claude loose in your team chat.

join the discussion ↗

04

🤖 Agent Watch ⬡ GitHub trending HEAT 64

withkynam/vibecode-pro-max-kit — Your AI forgets. This remembers. Spec-driven coding harness for vibecoders, product owners, CEOs and real builders — self-improving context memory, 12 agents, 32 skills. Kills context rot, ships features, not spaghetti. Claude Code & Codex. Any stack. 30 seconds

A spec-driven coding harness that ends AI context amnesia with 12 agents and self-improving memory.

👉 The take Stop your AI forgetting what it built yesterday—ship real features instead of debugging spaghetti.

05

✨ Try It Tonight ⬡ GitHub trending HEAT 64

UditAkhourii/adhd — ADHD — a skill for coding agents. Tree-of-thought with pruning, built on the Claude & Codex Agent SDK. Fans out parallel divergent thoughts under different cognitive frames, scores, prunes traps, deepens the survivors. The no-brainer skill for creative and interdisciplinary work.

Coding skill that fans out parallel thoughts, scores them, prunes dead ends, and deepens survivors.

👉 The take Give your coding agent a creative ADHD boost—great for brainstorming and untangling messy problems.

06

💸 Save Money ⬡ GitHub trending HEAT 63

ClaudioDrews/memory-os — A 7-layer memory operating system for Hermes Agent — persistent memory with Qdrant, structured facts, fabric recall, auto-curated wiki, and surgical context injection. Runs locally, any LLM provider.

Seven-layer persistent memory system for Hermes Agent using Qdrant—runs locally with any LLM.

👉 The take Make your agent remember everything like a second brain—runs on your own hardware, no cloud costs.

07

✨ Try It Tonight ▲ Hacker News 13h ago HEAT 62

Show HN: Best setup local LLM found for a 5090 (llama.cpp fork + turboquant)

A llama.cpp fork with turbo quantizations claims the best local LLM setup for an RTX 5090.

👉 The take Got a 5090? Turbocharge local inference without paying per token—tweak and try tonight.

join the discussion ↗

08

💸 Save Money ▲ Hacker News 5h ago HEAT 60

AbTARS – Self-hosted AI agent with persistent memory and 5-layer self-healing

Self-hosted AI agent with persistent memory and a five-layer self-healing system to recover from failures.

👉 The take Run an agent that fixes itself when it breaks—no cloud bills, just resilience on your own machine.

join the discussion ↗

09

🤖 Agent Watch 🤗 Hugging Face trending HEAT 60

nvidia/LocateAnything-3B (image-text-to-text) — new on Hugging Face

NVIDIA dropped a 3B-parameter model on Hugging Face that helps locate anything in images via text prompts.

👉 The take Point your agent at any image and ask "find the thing"—great for visual QA without expensive APIs.

10

🤖 Agent Watch ▲ Hacker News 11h ago HEAT 60

Show HN: Version Control for AI Agents

Version control for AI agents—track, rollback, and manage agent behavior like you do code.

👉 The take Stop fear-testing new agent configs; commit, rollback, and ship changes with confidence.

join the discussion ↗

11

✨ Try It Tonight ▲ Hacker News 2h ago HEAT 58

Show HN: Nightwatch, The open-source, read-only AI SRE

An open-source, read-only AI that monitors your systems like an SRE—without making changes.

👉 The take Let an AI watch your logs and alert you for free—no write access, no accidental chaos.

join the discussion ↗

12

🤖 Agent Watch ▲ Hacker News 16h ago HEAT 58

Obsidian-agent-bridge – let AI agents read, write, and deepen Obsidian vaults

Bridge lets AI agents read, write, and expand your Obsidian vault—turning notes into an agent brain.

👉 The take Give your agent direct access to your second brain—query and update your notes programmatically.

join the discussion ↗

13

✨ Try It Tonight ▲ Hacker News 5h ago HEAT 57

Show HN: A 178K Neural Net that beats Pokémon Roguelike

A tiny 178K-parameter neural net that beats the Pokémon Roguelike game—small enough to run anywhere.

👉 The take Train your own pocket-sized game-beating AI for fun or to learn—fits on a microcontroller.

join the discussion ↗

14

🤖 Agent Watch ▲ Hacker News 18h ago HEAT 57

Async hierarchical memory middleware for LLM agents

Async hierarchical memory middleware for LLM agents—lets them manage context across long-running tasks.

👉 The take Help your agent handle multi-hour workflows without losing its mind or hitting context limits.

join the discussion ↗

15

✨ Try It Tonight ▲ Hacker News 11h ago HEAT 55

Show HN: Lathe – Use LLMs to learn a new domain, not skip past it

Tool that uses LLMs to help you actually learn a new domain—not just skip to the answer.

👉 The take Stop AI from doing your homework; use it as a tutor that forces you to think.

source 1 source 2 source 3 source 4 source 5 source 6

join the discussion ↗

16

🌍 Big Picture ▲ Hacker News 21h ago HEAT 55

Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments

New benchmark Gaia2 tests LLM agents on dynamic, asynchronous environments that change in real time.

👉 The take See how your favorite agent handles chaos—benchmarks matter if you trust it with real-world tasks.

join the discussion ↗

17

✨ Try It Tonight ▲ Hacker News 3h ago HEAT 55

Gemma 4 E4B as a primary local LLM (replaced Qwen)

Gemma 4 E4B now preferred as a primary local LLM by some users, replacing Qwen on local setups.

👉 The take Swap your local model to see if Gemma 4 E4B runs faster or smarter on your hardware—no cost.

join the discussion ↗

18

🌍 Big Picture ▲ Hacker News 17h ago HEAT 52

I design with Claude more than Figma now

A designer reports using Claude more than Figma now—AI is eating the design toolchain.

👉 The take Let Claude draft your next UI layout—you might iterate faster than opening a design tool.

join the discussion ↗

19

✨ Try It Tonight ▲ Hacker News 8h ago HEAT 52

Helping Claude Code to find undocumented DuckDB APIs from the code

Workaround to help Claude Code discover undocumented DuckDB APIs by scanning the source code.

👉 The take Unlock hidden database powers—let Claude dig through DuckDB's internals for you.

join the discussion ↗

20

🌍 Big Picture ■ TechCrunch AI 6h ago HEAT 46

OpenAI is still working on that ‘super app’

OpenAI is still quietly building a 'super app'—a senior employee says 'chat is dead.'

👉 The take Your chat interface might be an all-in-one platform soon—prepare for deeper integration.

source 1 source 2 source 3

21

🌍 Big Picture ▲ Hacker News 9h ago HEAT 45

Anthropic, please ship an official Claude Desktop for Linux

Community pleas for an official Claude Linux desktop client—Anthropic hasn't shipped one yet.

👉 The take Linux users: keep asking—or build your own wrapper while waiting for the real deal.

join the discussion ↗

22

💸 Save Money ▲ Hacker News 4h ago HEAT 44

Self-Hosted JA4 to combat AI bots

Use self-hosted JA4 fingerprinting to detect and block AI bots scraping your site.

👉 The take Stop AI crawlers from eating your bandwidth—implement this cheap filter on your own server.

join the discussion ↗

23

💸 Save Money ▲ Hacker News 6h ago HEAT 41

How to Save Tokens When Coding with AI Agents: Subtitle Yadda Yadda

Tips and tricks to cut token usage when coding with AI agents—saves money and speeds up responses.

👉 The take Slash your API bills by making every token count—read this before your next agent session.

join the discussion ↗

24

💸 Save Money ▲ Hacker News 8h ago HEAT 41

Price Evolution, Production Frontiers, and Market Competition in LLM Inference

Analysis of LLM inference pricing trends, showing falling costs and shifting market competition.

👉 The take Inference prices are dropping—stay informed to know when to switch providers and save cash.

join the discussion ↗

25

🌍 Big Picture ▲ Hacker News 15h ago HEAT 34

How do AI agents spend your money?

Exploration of how AI agents can spend your money—risks and safeguards for autonomous spending.

👉 The take Before giving your agent a credit card, know the risks—and how to cap the damage.

join the discussion ↗

Top 10 by Heat Score

Where it's coming from

nexu-io/html-video — Programmatic video for coding agents — HTML to video on your laptop. Turn HTML, CSS & data into real MP4s with pluggable render engines, 21 templates, AI soundtrack. Apache-2.0, no per-render fees. An official project by the Open Design team.

OpenAI plots biggest ChatGPT overhaul since launch - Financial Times

Show HN: Claude Code on Slack/Discord/Telegram for flat $20/mo – no API bills

withkynam/vibecode-pro-max-kit — Your AI forgets. This remembers. Spec-driven coding harness for vibecoders, product owners, CEOs and real builders — self-improving context memory, 12 agents, 32 skills. Kills context rot, ships features, not spaghetti. Claude Code & Codex. Any stack. 30 seconds

ClaudioDrews/memory-os — A 7-layer memory operating system for Hermes Agent — persistent memory with Qdrant, structured facts, fabric recall, auto-curated wiki, and surgical context injection. Runs locally, any LLM provider.

Show HN: Best setup local LLM found for a 5090 (llama.cpp fork + turboquant)

AbTARS – Self-hosted AI agent with persistent memory and 5-layer self-healing

nvidia/LocateAnything-3B (image-text-to-text) — new on Hugging Face

Show HN: Version Control for AI Agents

Show HN: Nightwatch, The open-source, read-only AI SRE

Obsidian-agent-bridge – let AI agents read, write, and deepen Obsidian vaults

Show HN: A 178K Neural Net that beats Pokémon Roguelike

Async hierarchical memory middleware for LLM agents

Show HN: Lathe – Use LLMs to learn a new domain, not skip past it

Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments

Gemma 4 E4B as a primary local LLM (replaced Qwen)

I design with Claude more than Figma now

Helping Claude Code to find undocumented DuckDB APIs from the code

OpenAI is still working on that ‘super app’

Anthropic, please ship an official Claude Desktop for Linux

Self-Hosted JA4 to combat AI bots

How to Save Tokens When Coding with AI Agents: Subtitle Yadda Yadda

Price Evolution, Production Frontiers, and Market Competition in LLM Inference

How do AI agents spend your money?

You decide tomorrow's lead