DEDUPEDIdeas from people who actually tried the repos we cover: demos, workflows, results and quick hacks posted on X. Each card credits its author and links to the original post. Open the repo page for step-by-step use cases and a copy-paste agent prompt.
Latest
-
Naaackers@Naaackers
Using OpenClaw
I finally got around to trying "agent video editing", via OpenClaw + Claude Code. Here is what I did, how it's setup, what it cost, and my thoughts. My setup is different than most of the tests I've seen. TL;dr? Shheeeeeiiiiiiiiiitttt The initial goal Pop the SD card into my…
-
GitTested@GitTested
Using Headroom
we tested headroom, a tool that shrinks what your coding agent sends to the model. json: 8.2% fewer tokens. logs: 57.4% more, because claude kept asking for the cut lines back.
-
Jonny Campbell@jonnycampbell
Using Hyperframes
Spent the weekend finishing my personal site and finally pushed it live. Still not perfect but I'm calling it done for now. It has a power button. Press it and it boots into one of 14 movie interfaces, from HAL to Arrival. I made a short video using Opus and Hyperframes. It…
-
Mika Createsⓧ@arifcreationai
Using ComfyUI
For anyone curious how this was made: I made Mika and Kuro with WAI-illustrious-SDXL, then turned them into a video with MiniMax H3 (it handled the voice and lip-sync too). All of it runs in ComfyUI. Everything here is AI-generated.
-
Rey Brando@ReyBrando1
Using Hyperframes
I made this AI fighting game intro with Claude Code. No After Effects, no timeline, no editing app. The skills I used: 🎬 HyperFrames: turns HTML into video (every slam, flash and health bar) 🔊 media-use: pulled in the sound effects 🛠️ hyperframes-cli: checked every frame and…
-
Martin Cleary@cleary79
Using OpenClaw
I built a voice plugin for @omarchy + @openclaw 🦞 OpenClaw on a VPS, Omarchy on your desktop. Wake word, spoken replies and on-screen answers. Looking for testers! Requires OpenClaw main for now.
-
MEGA@Megaprompting
Using Graphify
Did my own benchmarks with Graphify it's useless in my domain it does not improve search retrieval times unless the query is extremely specific. Maybe I'm using it wrong but please Benchmark it with your own use case before spinning up an entire structure that needs none.
-
0xCS@FT_IOxCS
Using SGLang
Finally got DeepSeek-V4.1-Flash-1017 running properly across my 4× NVIDIA DGX Spark cluster. Repo: Model: deepseek-ai/DeepSeek-V4.1-Flash-1017 revision: 1bae0e1088993b9b82de08f59de5be5a34e42170 My setup: • 4× DGX Spark / GB10 • TP4 / EP1 • SGLang canary • 2× 200GbE RoCE rails…
-
kiter@kiter_js
Using Herdr
I've got a server with a bridge that listens to inbound from different sources first one was Whatsapp, now I'm adding a PWA built chat. The prompts arrive to a front desk agent who routes the prompt into a claude session in herdr so i can continue it or just directly handles…
-
Slobodan Margetic@WeeyugaFounder
Using llama.cpp
My setup for getting work done with AI on a budget: two 8 year old gaming laptops and a few weak computers joined in a private mesh, running Ollama and llama.cpp. Codex is the harness. When subscriptions run out, a local qwen 27b keeps things moving. Not fast, but it doesn't…
-
Squirrel Master@squirrelsmastrs
Using ComfyUI
i am using still images and using a 20gb 3080 with comfyui and the trellis and pixal3d models with mutli angle refrence shots and then run it through those models no 3d scanner
-
Thibaud Zamora@thibaudz
Using ComfyUI
i addedd some skills to give them access to my comfyui on my local ai server and they use it to generate video, music, sounds... all of that + some "cron" task. that's the latest things i use (today). but now, it's - roughly - the only app i m using.
-
Wendel@WendelAndrady
Using Supabase
I tried to break my own SaaS starter kit. Found 4 real bugs. Fixed all 4. Added regression tests. Your SaaS might have RLS enabled and STILL have security flaws. Building with Next.js + Supabase? 👇
-
furiyuri@StevenLHess
Using Ollama
I set up ollama on my @openSUSE Linux Box with WebUI running in a Podman container. Doing local LLM interactions. Would be nice if it had @grok intagration.
-
Qasim Ali@QasimAliSidhu
Using Crawl4AI
A lead score should come with receipts. I built a Python lead finder with support-signal breakdowns, public contact sources and CSV export. Demo in repo uses fictional data. Enrichment: @unclecode / Crawl4AI. #Python
-
Vatsal Mishra@watt_cell
Using OpenClaw
Openclaw. Been using it since launch and it works great! I also use Hermes but I prefer OpenClaw as a personal assistant/agent.
-
Albert@TheGreenCedar
Using OpenClaw
How do you guarantee that the saved instructions over time aren’t slop? I tried that with openclaw and it was never a great experience
-
Ryan Dunn@dunnry
Using Opencode
I tried the exact same prompt from Deepseek API and Opencode Go and got 15min 40s @ 131.7 tok/s with 137K tokens used vs 3 hr 20 min @ 21.3 tok/s and 284K tokens used. Very similar quality result however. Deepseek API is MUCH faster.
-
Vaibhav Rana@vib_z
Using Herdr
Same pain, so I made a Grok Bot that walks you through it: Tailscale, Herdr and the stampede plugin on your Mac, then a test job to prove it works. Bot: Plugin, MIT
-
Hyperspek ⌐◨-◨@hyperspek
Using Ollama
Direct calls to Ollama works best imo. I tried routing via Hermes. Very inefficient, large context needed and 5-10x slower.
-
Serhiy Krytskyi@kritzkidj
Using Voice Studio
I’ve been using ElevenLabs for a while and also installed VoiceStudio to test it. For most everyday cases it actually works well, but from my experience it still falls short of ElevenLabs in overall quality (especially on longer content and more nuanced voices). This is just my…
-
Vantekai@Vantekai_
Using Hyperframes
I get pretty amazing videos using hyperframes without the studio so what does it offer over and above the regular skills and package claude.and grokbot etc use?
-
SunsetSyntax@sunsetsyntax
Using Think Term
I kept losing agent runs every time I closed a terminal window by mistake. Mid-refactor, gone. ThinkTerm fixed that for me. It's a Rust terminal built on WezTerm where a mux server owns the sessions: detach, reattach from the TUI or browser, and the agent is still going.
-
Paul Villanueva@paulvilloff
Using OpenClaw
Hermes has been the one for me too lol I use OpenClaw still but only so I can learn and adapt with both harnesses. I also built a lot of my own tools so both can use them too, makes it easy if I ever want to create my own harness in the future.
-
Sai Krishna Bolla@saikrishnabolla
Using Skills
I adapted pstack and Matt Pocock skills, with rules for which machine does the work. I still check what actually got tested. Heavy builds overloaded Oracle, so I move them where there’s capacity. 7/8
-
M A N D O@MandoMendoz
Using ComfyUI
ComfyUi Day 2. I have an RTX 5080 with 16GB of VRAM. When running large AI models, formats like GGUF and tools like Nunchaku can help reduce VRAM usage. Understanding optimization is also part of learning how to work with generative AI. Here i´m using the z-image model fp8 r128
-
Scarab@ScarabOfficial
Using ComfyUI
I've been rendering a whole bunch of #Wan21 SCAIL2 videos in #ComfyUI the past few hours, and have just realised I am not using the fastest model I have.
-
Felix Lopez@felixlopmsc
Using Supabase
At an insurance agency the small things are the renewal reminder and the follow-up call. Automating both on a custom Supabase CRM is the most boring and most useful thing I've built. What's the smallest habit that compounded for you?
-
Gean@geankarloo
Using PhotoCraft
Is anyone having a bad workflow with PhotoCraft? I canceled my Adobe subscription thinking it would perform well, haha, my mistake. I used an older PSD file and the workflow is a nightmare.
-
MrMySQL@be1k0
Using T3 Code
connecting multiple devices via Tailscale (not t3 connector) is still a nightmare. I have to run 2 instances on each machine - one is background running t3 server, another one is built in t3code. I still couldn't make visible all background t3 servers on all devices, only mobile
Most liked
-
patch@patch_gg
Using Universal Modder
NPCs can talk back now! Introducing NPCTalk: make the NPCs of any game talk in realtime. I made every NPC in GTA V answer to me in character, in their own cloned voice, in under 1s, and they act on what you say. We packaged all the skills Claude learned into a single repo…
-
Eric Xu (e/Mettā)@xleaps
Using LangExtract
Speaking of re-writing systems implemented in other languages using Rust, I recently asked Claude to reimplement the hyped Google langextract repo in Rust. After burning about <$1 I now have a Rust version that's much faster and I have been using it. I don't know what it means…
-
Alex C.@alexcovo_eth
Using Hermes Agent
My Robot arrived yesterday afternoon, so I decided to run my own stack with it and make a Hermes-Agent out of it. She's a keeper. Robot: @M5Stack StackChan with the CoreS3 (ESP32-S3) Voice server: xiaozhi-esp32-server 0.9.6 running locally in Docker on my Mac Voice: locally…
-
Printer ⚡@Printer_Gobrrr
Using Solidcraft
Running Solidcraft. An open-source parametric 3D CAD software aiming to replace Fusion and let me tell you, Autodesk's days are numbered. I tried it for the first time and felt right at home, then created this case in 5 minutes. All the features are there, even the hotkeys are…
-
許文彥@vm3jp6u04
Using Text to Cad
Yesterday I found an open-source Text-to-CAD project: So I tried the same CAD practice model I previously used to test Adam inside Onshape. This time, the setup was different: I gave Codex only a YouTube link. No dimensions typed in. No back-and-forth. No manual CAD operations…
-
Brian Roemmele@BrianRoemmele
Using Harnessrouter
How I Use Open Source HarnessRouter + a Grok Bot For Personal Work and Corporate Clients Most people treat multi-agent systems as either “one big bot” or a pile of disconnected tools. The pattern I’ve settled on is different: keep a long-lived coordinator (Grok Bot or an…
-
Arnav Gupta@championswimmer
Using Herdr Gpui
herdr-gpui (now just called herdr-ui) has totally become my daily driver terminal now (i have multiple agents running locally, on my minipc, and cloud VMs) with today's release of Herdr UI you can directly mint a fresh new @Railway VM (no account required!) and start using it
-
Kai@hqmank
Using Magpie
Just tried a community plugin to connect ZCode to my personal AI gateway through Magpie. It works! Now I can use GLM-5.3 and GLM-5.3-Flash through an API or connect them to coding agents like Claude Code. Nice way to make use of ZCode's free 500M tokens this weekend.
-
Deimantas Norkeliunas@deimantasnork
Using Herdr Gpui
Hey captain @kunchenguid, look what I built for Firstmate on @fabienpenso’s Herdr-gpui edition 👀 Captain’s Deck: live Kanban - Charted Next → Underway → Captain’s Call → Awaiting Merge → Landed. Answer Captain’s Calls from the board.
-
Alok@analogalok
Using Kitten TTS
Stop guessing which open source TTS model to use for your local voice agents. I just ran a brutal benchmark: the new ultra lightweight KittenTTS (15M) vs Kokoro (82M). One is an absolute speed demon on a basic CPU. The other is practically indistinguishable from a real human…
-
Sabba Keynejad@sab8a
Using Open Edit
This is wild 🤯 I recreated GPT Images 2.5 launch video! With GPT-6 ASTRA on VEED OpenEdit. Test here: Comment and I'll share full prompt.
-
Jussi Tuulensuu@tuulensuu
Using PhotoCraft
Out of interest, I tried the free, vibe-coded Photoshop clone called PhotoCraft. Short verdict: Somewhat sleek looks and many of the bells and whistles, but it cannot be used for professional work, at all. Deserves to be called Photoslop. Features and usability resemble the…
-
OficineRobotica@OficineRobotica
Using Solidcraft
I tried Solidcraft @bherbruckx and got some really nice surprises as well as some early adopter pain points. Take a look here
-
Madhuri ✳︎@Iruhdam24
Using Anti Slop
I tested 67 skills and tools for killing AI slop, and built one skill that tells you which to use. It reads your repo and suggests the right anti-slop skill for your text, UI, images, motion, diagrams or charts. guide & skill below 👇
-
Gabu@gabu3d_pl
Using Project Maya
GLM5.3 Flash 65t/s on Maya-M model. Borsuk (my clanker) made this video - don't blame me in case of mistakes :) It is analysis of build I run at the moment, based on MEASURED data.
-
Badr Bellaj@BadrBellaj
Using llmfit
llmfit is an interesting ai tool It helps you check the right local LLM for your hardware. It checks which one from the 497 models can run on your machine. I have tested it on my mac mini m4 and it's excellent
-
Guitar Cat + LLM 「Will @ SGLang Summit in SF」@GenAI_is_real
Using Lithos Metal
Tried @JiaZhihao's lithos-metal immediately on the Qwen3.8-27B workload I use every day. Huge congrats to the team. On my M5 Pro, lithos is meaningfully faster than Ollama after warmup. More interestingly, this test also shows why inference speed and serving UX are two…
-
Dave Sottimano@dsottimano
Using RTK
Codex users, don't forget Rust Token Killer (rtk to drastically reduce token usage. How it works: Been using this awesome proxy on Claude Code for a while, it's fantastic. Rarely trips, usually on debugging logs, but rare.
-
odudex@odudex
Using Solidcraft
I just tried SolidCraft for an hour. I was able to do everything I tried on my first attempt, as was Opus. It sets an inflection point in CAD history.
-
Slonski@Slonski_rt
Using Claude Mem
I got tired of Claude Code wasting tokens before the real work even started new session same project same context explained again so I built a setup around it: - OmniRoute for flexible model routing - Headroom to compress noisy context - claude-mem for persistent memory -…
-
Pamela Fox@pamelafox
Using GitHub MCP Server
The GitHub MCP server added a dedicated tool for replying to inline PR comments! I updated my review-pr-comments skill to use it: I love to collaboratively triage PR comments with Copilot-I can more quickly decide to accept/reject.
-
Oracle Cloud@OracleCloud
Using PersonaPlex
What does a real-time healthcare call agent need to handle natural conversation? We built an experimental prototype combining NVIDIA PersonaPlex, LiveKit, and OCI GPU instances powered by NVIDIA A10 Tensor Core GPUs. Read the technical walkthrough
-
Duy /zuey/@goon_nguyen
Using Hyperframes
i just made this anime clip! 😁 AI video folks be like: “bro, you just prompted Seedance, big deal” 😂 plot twist: i made it with Opus 5.5! yes, seriously! Seedance is amazing, but a 15-second clip can cost $4-5, and every regeneration burns more money. so i tried something…
-
Naaackers@Naaackers
Using OpenClaw
I finally got around to trying "agent video editing", via OpenClaw + Claude Code. Here is what I did, how it's setup, what it cost, and my thoughts. My setup is different than most of the tests I've seen. TL;dr? Shheeeeeiiiiiiiiiitttt The initial goal Pop the SD card into my…
-
Aarno@TheGlobalMinima
Using Context7
The recent usage limits on antigravity do suck. But here’s a flow I’ve been using these past few weeks and it works beautifullyyy > opus 4.5 (thinking) for implementation plans and reviews > let opus set up boilerplate / start with first step > switch to gemini 3 pro for…
-
juggernaut@curlysaarthak
Using Crawl4AI
i made a @firecrawl alternative using @unclecode 's crawl4ai convert any documentation to llm context ready markdown files
-
Nidhi Singh@nidhisinghattri
Using Voice Studio
i tried voicestudio for voice cloning and it is so good def a strong competitor to elevenlabs or any other paid subscription open-source just keeps getting better
-
Andrea Intg.@andreintg
Using PhotoCraft
Still using photocraft every day, goodbye photopea, photoshop, and every other phot-like app I tried over the years 🤷
-
Jordan Hall@jordanhall
Using Text to Cad
text-to-cad + 3D printer is a ton of fun for prototyping I made an erector-like toy building set for my kids and they’re having a ton of fun with it. Still need to dial-in a few parts, but @sendcutsend next?
-
Shantanu Goel@shantanugoel
Using Spec Kit
I tried Github spec-kit with GPT 5.3 codex xhigh and Opus 4.6 (in separate experiments) today. It felt great!!! Just like kiro when it came out. But it fell flat soon. Needs too much babysitting, and generates so much stuff that one gets lost if everything is covered or not. By…
Posts are found by searching X for each repo and kept only when they show someone using it. Quotes are short excerpts with links removed; the post and its content belong to its author. Deduped is not affiliated with the authors or the repos. Want a post removed? Reply to any Deduped email.
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them free