Report by The Next New Thing.
Scroll for the report ↓
Turns your AI coding assistant (Claude Code, Cursor, Copilot, Windsurf, Codex) into a full video-production studio. Created less than three months ago and pushed same-day; the README badges #1 on GitHub Trending and momentum is commit- and marketing-driven. Heavily self-promoted via its own YouTube and X.
View on GitHub →Works with any AI coding assistant; README claims 12 pipelines, 52 tools, 500+ skills.
Cuts actual stock/archive footage into a timeline, not just animating stills.
Start from a reference video → 2–3 concepts + cost estimates before producing.
Disclosed demo costs: $1.33 Pixar-style short, $0.69 one-key ad, $0.15 FLUX anime.
Optional FAL/Pexels/Suno/ElevenLabs keys; ffprobe/frame/audio checks on output.
A non-engineer can produce a narrated, captioned, music-backed video for cents to a couple dollars by telling their coding agent what they want.
Indexes your repo into a searchable knowledge graph so your agent answers "what calls what" instantly instead of re-reading files. Created Feb 2026, pushed same-day, latest release v0.8.1. Strong perf claims (Linux kernel in 3 min, sub-1ms queries), an arXiv preprint, and plug-and-play setup across 11 coding agents.
View on GitHub →Pure C, zero dependencies/API keys; one-line install; 100% local.
tree-sitter parsing plus Hybrid LSP type resolution for 11 languages.
README claims 120× fewer tokens; the paper reports 10× fewer tokens, 2.1× fewer tool calls.
One install wires up Claude Code, Codex, Gemini CLI, Zed, OpenCode, Aider, VS Code.
Semantic + BM25 search, Cypher queries, call graphs, dead-code detection, git-diff impact.
Makes an AI coding agent smarter and cheaper on big codebases — it stops burning tokens re-reading files and answers architecture questions in one query.
We aim to bring this man in for questioning — on camera.
If you know him, tag him. If you are him… let's talk.
A free, pretrained time-series foundation model from Google Research. The most-established repo in the slate (live since 2024). TimesFM 2.5 is the latest model; recent updates include PyPI 2.0.0, a LoRA fine-tuning example, and integration into BigQuery ML, Sheets, and Vertex.
View on GitHub →200M params (down from 500M), up to 16k context length, drops the frequency indicator.
Optional 30M quantile head gives continuous quantile forecasts up to a 1k horizon.
pip install timesfm[torch]/[flax]; pretrained checkpoints on Hugging Face.
Backed by an ICML 2024 paper (arXiv 2310.10688).
Reachable via BigQuery ML (SQL), Google Sheets, and Vertex Model Garden.
Founders can get solid demand/sales/traffic forecasts with uncertainty ranges without a data scientist or custom training — load the model, pass your numbers, get predictions.
The TimesFM 2.5 launch thread on Hacker News drew 327 points and 109 comments — practitioners debating zero-shot forecasting quality, the parameter cut from 500M to 200M, and where a foundation model beats classical methods.
I've used Zapier across multiple businesses for over 10 years. They've been doing automation reliably long before this moment — and the MCP service brings that same reliability to any AI agent you're building with.
Try Zapier MCP →Exposes 8,000+ tools as MCP resources any agent can call.
Claude Code, OpenCode, Cursor — anything that speaks MCP.
Enable the tools you want in Zapier, grab the MCP endpoint, paste it into your agent.
Zapier handles auth, rate limits, and reliability for you.
peerd is brand new — created June 22, 2026 — and already hit the Hacker News front page with a “Show HN” post at 68 points and 25 comments. It pitches itself as “the first AI agent harness native to the browser,” running the agent loop client-side with a security model where the key-holding agent never directly reads raw page content. It's an early 0.x developer preview, so it's one to watch rather than rely on.
View on GitHub →Runs the full agent loop in Chrome/Firefox — no servers, no external processes.
Spins up JS notebooks, WASM Linux VMs, and client-side apps inside your tabs.
Shares what the agent builds between peers over WebRTC.
Bring your own key (Anthropic, OpenRouter, local Ollama); data stays encrypted locally.
If you want to experiment with autonomous browser agents that use your existing logged-in sessions without shipping your data to a vendor's cloud, peerd is the bleeding-edge place to start.
A free, open-source macOS dictation app that turns speech into text on-device. It's free and open source (GPLv3) and the homepage reports 25,000+ downloads. A power dictation user publicly cancelled his paid Wispr Flow plan for it in a tweet that hit ~179k views. Latest release v1.6.0 shipped June 22, 2026, with multiple speech models (Parakeet, Cohere, Whisper, Apple Speech) and on-device AI correction.
View on GitHub →On-device transcription with optional AI cleanup — no audio leaves your Mac, no API key needed.
Global hotkey inserts text into any app — Notes, Slack, Gmail, Cursor, code editors.
Choose Parakeet, Cohere, Whisper, Nemotron, or Apple Speech; up to ~99 languages.
Completely free forever, installable via Homebrew (macOS 15+, Apple Silicon).
A founder can replace a paid dictation subscription with a private, on-device tool and dictate roughly 3.7× faster than typing into any app.
A local-first Twitter/X workspace that stores all your tweets, DMs, likes and bookmarks in a single SQLite database. Built by Peter Steinberger (@steipete), a well-known developer — a big part of its pull. His own tweet calling it “my favorite way to read Twitter” did ~603k views, and the repo has crossed 1.3k stars since launching March 2026. Bundles archive import, cached live reads, full-text search, and AI inbox ranking into one local app + CLI (latest release 0.8.5, June 21, 2026).
View on GitHub →Stores tweets, DMs, likes, bookmarks, and follow graph locally under ~/.birdclaw.
Home timeline, mentions queue, DMs workspace, and JSON-piping CLI in one tool.
FTS5 search across all tweets and DMs, plus follow-graph analysis.
Structured data “claw-able” by agents, with AI-powered inbox ranking and digests.
A creator can archive and actually search their entire Twitter history offline — and point AI agents at it — instead of being locked inside X's app.
A free, real-time global situation room in your browser — a self-hostable, Palantir-style OSINT dashboard. (OSINT = open-source intelligence: insight built only from publicly available data — news, markets, sensors, social feeds.) For a founder that's three money angles in one: anticipate shifts in the business you're already in, get early reads on the stock market, and spot mispriced Polymarket bets. Hundreds of news feeds plus geopolitical, financial, and infrastructure data sit in one map-based dashboard. Highest star count in the slate (~59.8k) despite being created only Jan 2026; pushed same-day.
View on GitHub →15 categories AI-synthesized into briefs; aggregates 65+ providers/APIs.
3D globe (globe.gl) + WebGL flat map (deck.gl) with 56 layer types; signal correlation.
Country Instability Index for 31 Tier-1 countries; 29 exchanges, commodities, crypto.
Ollama with no API keys; also Groq/OpenRouter and browser Transformers.js.
6 variants from one codebase + Tauri 2 app (Win/macOS/Linux), 24 languages.
If you trade, run a business, or bet on Polymarket, you can self-host a serious OSINT dashboard for free (or use worldmonitor.app) and try to see the move before everyone else — instead of paying for an enterprise tool.
It's a free, self-hostable Figma alternative — and every design comes out as real code. Designers and developers work on the same files in real time, with full ownership and no vendor lock-in. A mature 53.7k-star project pushed today; the README now pushes its new MCP server for AI design-to-code plus native Design Tokens — momentum is sustained development plus the AI/MCP angle.
View on GitHub →Docker/Kubernetes/Elestio or the hosted SaaS — full ownership, no lock-in.
Open standards (SVG/CSS/HTML/JSON); the Inspect tab gives ready-to-use code.
Enables multi-directional design↔code↔AI workflows.
Components and Variants as a single source of truth for design systems.
Kaleidos raised $20M — an $8M Series A (Sept 2022) and a $12M extension (Feb 2023), both led by Decibel Partners, with Figma's ex-COO Eric Wittman among the angels.
A real Figma-class design tool that's free and self-hostable — and because designs are code plus an MCP server, your AI tools can actually read and act on them.
A 775-point, 207-comment Hacker News thread (2025-11-27) on Penpot as the open-source Figma — a substantive debate about self-hosting, design-as-code, and whether an open tool can match Figma in practice.
A free, open-source voice studio from Jamie Pine (creator of Spacedrive), pitched as one free local replacement for two paid clouds at once — ElevenLabs (output) and WisprFlow (input). ~34k stars in roughly five months, with a Trendshift badge. Note the freshness flag — last pushed 2026-04-26.
View on GitHub →Tauri/Rust, not Electron — voice data never leaves your machine.
Qwen3-TTS, Kokoro, Chatterbox, HumeAI; zero-shot cloning or 50+ preset voices.
23 languages plus tags like [laugh] and [sigh] via Chatterbox Turbo.
Global hotkey + Whisper speech-to-text with auto-paste on macOS — a free dictation tool that stands in for WisprFlow.
One MCP call (voicebox.speak) lets Claude Code/Cursor speak back; REST + MCP server.
Replaces two paid voice subscriptions with one free download — and keeps your cloned voice and audio on your own computer instead of a vendor's cloud.
A regularly-updated archive of the hidden system prompts behind the major AI products. Actively updated (pushed today) and adds the newest models fast — the README lists Claude Design (June 23), GPT-5.5 Codex, Claude Fable 5, Gemini 3.5 Flash. Cites a real Washington Post feature driving mainstream attention.
View on GitHub →Anthropic, OpenAI, Google, xAI, plus Cursor, Copilot, Perplexity and more.
Claude Fable 5, Opus 4.8, GPT-5.5, Gemini 3.5 Flash — all dated.
See exactly what changed, e.g. Claude Opus 4.8 → Fable 5.
Web search, code, image-gen, memory — not just the chat personas.
CC0 — free to copy, study, and reuse.
For anyone writing prompts or building on these AIs, seeing how the big labs actually instruct their own models is a free masterclass — and reveals the guardrails shaping your answers.
A 627-point, 334-comment Hacker News thread links the repo’s claude.txt and dissects how large and detailed a production system prompt really is. The repo also anchors a Washington Post feature (~2026-05-11) profiling the maintainer.
One CLI that wires your agent into the platforms that normally block scrapers or charge for API access. A new project that hit ~41k stars in about four months, with a Trendshift "GitHub Trending #1 Repository of the Day" badge. The hook: paste one install sentence and minutes later it reads Twitter, Reddit, and YouTube for free.
View on GitHub →Paste a single raw-GitHub-URL sentence; the agent sets up the CLI, Node.js, gh, and MCP.
Web, YouTube transcripts+search, RSS, GitHub, full-web semantic search via Exa, V2EX.
When one method breaks (e.g. yt-dlp blocked) it auto-switches; users do nothing.
"agent-reach doctor" reports which channels work and how to fix the rest.
Open-source tools, free APIs, cookies stay local; only cost is a ~$1/mo server proxy.
If your agent comes up empty when you ask it to check Twitter or summarize a YouTube video, this is a one-paste fix that wires those sources in for free.
Runs every weekday and pushes a plain-language readout to your phone, Telegram, Slack, or email — without running a server. Created Jan 2026, already ~49.5k stars and 43k+ forks (the fork count reflects its fork-to-deploy model). README carries a Trendshift #1-Python badge and a HelloGitHub feature; buzz is China-developer-driven.
View on GitHub →A-shares, HK, US, Japan, Korea with indicators, fund flows, news, fundamentals.
Conclusion, score, trend, buy/sell levels, risk alerts, catalysts, action checklist.
Fork + add API keys; GitHub Actions auto-runs each trading day; Docker/local too.
WeChat Work, Feishu, Telegram, Discord, Slack, or email.
Gemini, OpenAI-compatible, DeepSeek, Qwen, Claude, or local Ollama.
If you follow a handful of stocks, fork it and have an AI quietly send you a daily readout on each — free, no coding, no server.
Edit without a subscription or watermarks. The description is literally "The open-source CapCut alternative" and at ~60k stars it's one of the most-starred OSS creator tools. The timeliness now is the rewrite: the README's top section announces OpenCut is being rebuilt from the ground up with a Rust core.
View on GitHub →Plugin-first architecture and an Editor API; the usable version powers opencut.app.
Desktop, mobile, and browser planned from a single Rust core.
An MCP server for agents, headless batch rendering, and an in-editor scripting tab.
First-class plugins are a stated goal of the new architecture.
Supported by sponsors including fal.ai.
A free, self-run video editor for creators who don't want to pay for CapCut — and the rewrite aims it at people who want AI agents and automation in their editing pipeline.
New hot repos every week. Subscribe so you never miss the next drop.
