Every repo from my series, newest first. I keep this same link updated — save it.
Follow me on Instagram for the next one: @angelotrifanoff.ai
https://github.com/MadsLorentzen/ai-job-search (MIT, ★36K+)
Fork it, open it in Claude Code, and it scrapes job postings, tailors your resume and cover letter for each posting, and preps you for the interview. Built by a laid-off geophysicist who says he went 69 applications → 20 interviews → 1 offer with it.
git clone https://github.com/MadsLorentzen/ai-job-search.git
cd ai-job-search
Then open Claude Code in the folder and run /setup.
Heads up, and this part matters. You need Claude Code plus a LaTeX install (MacTeX or TeX Live) since it compiles your resume and cover letter as PDFs. Your CV or LinkedIn export goes in a documents folder and /setup reads it. The interview numbers are his own, not mine.
https://github.com/different-ai/openwork (★23K+)
A desktop app where an AI agent does real work on your computer, not just chats. Free, and it runs on Mac, Windows and Linux. Bring your own model, Claude included.
Easiest install, paste this one line into Claude Code and it does the rest:
Install OpenWork on my computer, set up my first workspace, and open it ready to use. Follow the steps in https://openworklabs.com/start.md?v=hero
Heads up, and this part matters. It is powered by opencode, so it is not Claude itself, it is an agent that can run on Claude or any other model you plug in. You still need your own API key or plan for whatever model you pick.
Over 1,000 apps, one question each: can AI build you a replacement today, or is it still worth paying for? Type any app you pay for and it hands you a verdict, the exact prompt to build the replacement, and free alternatives (1,800+ across the list). Completely free to use — no signup, no checkout, it runs on sponsor links.
Heads up: entries are user submitted, so expect some rough edges — verdicts are opinions with receipts, not gospel. The site itself is open source (MIT): https://github.com/robj3d3/canivibecodeit
https://github.com/OpenCut-app/OpenCut
Use it right now in your browser: https://opencut.app
No account, no subscription, and nothing locks up the second you go to export. Timeline, cuts, text, masks, a graph editor for curves, volume and speed controls, canvas backgrounds and stickers. It runs in the browser and exports from your machine. MIT licensed, 82k stars.
Heads up, and this part matters. The repo itself is being rewritten from the ground up, so don't clone it expecting a finished editor. The old codebase (opencut-classic) is now archived and read only. The thing that actually works today is the live app at opencut.app.
What's coming is the reason to watch it. The rewrite ships an MCP server, so an AI agent can drive the editor for you, plus headless mode for batch rendering and a scripting tab inside the app. Preview of the rewrite: https://new.opencut.app
https://github.com/harry0703/MoneyPrinterTurbo
Type a topic, it writes the script, finds free stock footage, voices it, burns captions, adds music, renders HD 9:16 or 16:9. MIT licensed, 116k stars.
docker compose -f docker-compose.release.yml up
Then open the WebUI. Or paste the README's "AI Agent" prompt into Claude Code and it sets it up for you (Mac/Windows).
Heads up, and this part matters. You need your own LLM API key (Claude, OpenAI, Gemini, DeepSeek all work). Voice (Edge TTS) and footage (Pexels/Pixabay) are free. Output is stock-footage style, so it's best for b-roll, faceless and explainer videos.
https://github.com/calesthio/OpenMontage (AGPL, ★51K+)
Describe the video in plain English from Claude Code and it scripts, pulls free footage, narrates, edits, self-reviews and stops for your approval. 11 pipelines (explainers, documentaries, talking head, podcast clips, dubbing). Send it a short you like and it makes one about your topic.
git clone https://github.com/calesthio/OpenMontage.git
cd OpenMontage
make setup
Then open Claude Code in the folder and ask for a video.
Heads up, and this part matters. The base is free (offline narration, open archives, local render). AI-generated clips, images, music and premium voices need paid keys (cents to a few dollars a video). Local LLM support is coming, not here yet.
https://github.com/docusealco/docuseal
Try it right now in your browser: https://demo.docuseal.tech
Upload a PDF, drop the fields on it (signature, date, checkbox, file upload, 12 types), send it to one or more people, and they sign from a link. Automated emails, audit trail, the same flow as DocuSign. Proposals, onboarding forms, contracts. AGPL-3.0, 18k stars.
Run it yourself in one line:
docker run --name docuseal -p 3000:3000 -v.:/data docuseal/docuseal
There are one-click deploy buttons in the README too (Railway, Render, DigitalOcean, Heroku), or hand the repo to Claude Code and have it put it on a server for you. Self-hosted is free with no document limit, you just plug in your own email (SMTP) for the sending.
Heads up, and this part matters. Their hosted cloud version has a free tier, but it's capped at 10 signature requests a month, and the unlimited cloud plan is $20 per user per month. The unlimited-for-nothing version is the one you host yourself. For comparison, DocuSign's cheapest plan is $11 a month for 5 documents, and the normal plan is $30 per user per month capped at 100 a year.
https://github.com/google/skills
claude plugin marketplace add google/skills
Then claude plugin install <plugin>@google-plugins to pick what you want. Same thing on Codex: codex plugin marketplace add google/skills, then install from the /plugins browser. Or npx skills add google/skills works anywhere.
110 skills written by Google themselves, so Claude stops guessing at their APIs. Apache-2.0, 18k stars.
Heads up, and this part matters: 96 of the 110 are Google Cloud infrastructure (GKE, BigQuery, that whole world) — great if you build on Google Cloud, skip them if you don't. The ones most people actually want: the Google Ads skills (there's a Google Ads MCP server, so you can point Claude at your ad account and ask what's broken, plus an account performance diagnostics skill), the two Google Analytics skills, and Firebase / Cloud Run basics if you're building.
Bonus: Google links their other official skills repos in the README — Android, Flutter, Dart, Firebase/Firestore, Genkit — all installable the same way.
https://github.com/CapSoftware/Cap
Download the app: https://cap.so/download
Record your screen, get a shareable link — the same flow as Loom, except your recordings stay on your machine and you can host the whole thing yourself. Real app for Mac and Windows, set up and recording in under a minute. 20k+ stars.
Heads up, and this part matters. The free tier is "free forever" but recordings cap at 5 minutes and it's personal use only. Commercial use of the desktop app is a $29/year license, and their cloud hosting (Cap Pro) is $12 per user per month. Self-hosting is free — or download the source, build it with your AI agent, and put it up on Vercel or Cloudflare yourself. License is mixed: MIT for the camera/capture crates, AGPLv3 for the rest.
https://github.com/macro-inc/macro
Use it right now: https://macro.com/app (iOS app too, Android coming)
Email, channels, Linear-style tasks, docs, a canvas, calls, file storage, pull requests and a CRM, all in one workspace and @-linked together. The part that matters: every surface shares one team-level memory, so an agent can read your calls, tasks and emails and go do the thing instead of you briefing it each time. Open source, AGPL-3.0, ~4k stars and climbing fast.
Hook it to Claude Code:
claude mcp add --transport http macro https://mcp-server.macro.com/mcp
Then see their MCP setup and agent recipes in the docs: https://docs.macro.com/AI/mcp/overview
Heads up, and this part matters. Free forever for personal use (one seat, with a "Sent with Macro" email signature and storage/AI limits). Teams are $40 per seat per month, or self-host it yourself. And it doesn't replace Gmail — it's the email client on top of your Gmail or Google Workspace, it connects to your existing account.
https://github.com/cloudflare/cloudflare-os
git clone https://github.com/cloudflare/cloudflare-os.git
cd cloudflare-os
pnpm run-local
Then open http://localhost:8787 — the whole stack runs on your machine.
The agent workspace thousands of Cloudflare employees (not just engineers) use daily — agent chat preloaded with company context, sandboxed "gadget" app building, and a security layer (Gatekeepers) so non-technical people can't break anything. Apache 2.0, open-sourced during Agent Week (Aug 2026).
- Starter/deploy guide: https://github.com/cloudflare/cloudflare-os-starter
- Announcement: https://blog.cloudflare.com/cloudflare-os/
https://github.com/cloudflare/computer
Your agent writes some files, the session ends, and they're gone. This gives it a filesystem that actually survives — the files live in a database, so the agent can go to sleep and everything's still sitting there when it wakes up.
Three ways to run code against those same files: a full Linux container when it needs real programs, a shell that runs with no container at all, or plain JavaScript. Cloudflare says a container is only needed for under 10% of the work.
Heads up: this is preview only, not for production. The APIs are unstable and Cloudflare says so themselves. Read it to see where agents are going, don't build on it yet.
Blog: https://blog.cloudflare.com/cloudflare-computer/
https://github.com/tornikegomareli/Talkify
Download the app: https://github.com/tornikegomareli/Talkify/releases/latest
Hold one key, talk, let go, and the text lands wherever your cursor is. Any app. It runs on the on-device speech engine Apple shipped in macOS 26, so nothing you say is ever uploaded anywhere. It also reads selected text back to you. MIT licensed, 407 stars, and the repo went up on August 5th.
Grab the .dmg from the releases page, that's the easy path. There's a Homebrew tap in the README too, but it needs an extra brew trust step that looks scarier than it is.
Heads up, and this part matters. You need macOS 26 (Tahoe) on an Apple Silicon Mac. Both are hard requirements, so Intel Macs are out and so is anyone who hasn't updated yet.
It's also pre-1.0 and moving fast. Two things to know: it inserts text by pasting, so some apps that handle paste in their own way can swallow it — terminals and virtual machines are the known ones (open issue #38, fix in progress). And version 1 deliberately types exactly what you said, so there's no filler removal or punctuation cleanup yet.
https://github.com/cathrynlavery/diagram-design
/plugin marketplace add cathrynlavery/diagram-design
/plugin install diagram-design@diagram-design
Same two commands on Codex, just codex plugin instead of /plugin.
27 visual types your agent can draw as self-contained HTML + SVG — no build step, no Mermaid. Architecture, flowcharts, sequence, ER, timelines, swimlanes, org charts, Venn, funnels, the consultant 2x2, radar, Gantt, bar/line/scatter, process, data flow. MIT, 18k stars, and it hit number one trending on GitHub this week.
The part people miss: this is not just an engineering tool. The README says it straight out — editorial-quality diagrams for general professional communication, not exclusively software engineering. Org charts, funnels and Gantt charts are in there next to the architecture stuff.
The good part is the onboarding. It reads your website once, pulls your real brand colors and typography, maps them to semantic roles, checks them for WCAG AA contrast, and writes it all to references/style-guide.md. Every diagram after that comes out already looking like your brand instead of looking like a default template.
Heads up, and this part matters: PNG export is not included out of the box. The HTML works immediately, but rasterising needs a one-time pip install playwright && playwright install chromium. The README also has an explicit "don't use this for" list — quick unicode diagrams, plain lists, before/after comparisons, single shapes. Their rule of thumb is worth stealing: before drawing, ask whether a reader would learn more from this than from a well-written paragraph. If no, don't draw it.
https://github.com/digimata/quill
git clone https://github.com/digimata/quill.git
cd quill
swift build -c release
sudo cp .build/release/quill /usr/local/bin/quill
Click the feather in your menu bar to record, click stop and it transcribes right on your Mac. It records your mic and the system audio as two separate tracks, so the transcript already knows who said what — you vs them — with no speaker-detection model. No account, no upload, no bot joining your call. Everything lands in ~/Recordings/ — both audio tracks plus a transcript you can read or hand straight to Claude. Same developer as Parrot below.
Heads up: you build it from source (no installer yet), it needs macOS 15+ (Apple Silicon recommended), and the first transcription downloads the models (~600 MB) once. If the menu bar icon disappears it's still recording.
https://github.com/digimata/parrot
curl -fsSL https://digimata.github.io/parrot/install.sh | sh
parrot setup
Hold the fn key, talk, let go — the text types itself in at your cursor. Runs small AI models locally on the Neural Engine, so your voice never leaves your Mac. No cloud, no account, no subscription. Menu bar app, single Swift binary, macOS 14+ on Apple Silicon.
Heads up: builds are unsigned right now, and if the menu bar icon disappears it's still running.