Rexei (@iamrexei) header
Rexei (@iamrexei) profile picture

building AI stacks in public | open-source tools, automations & experiments https://teapaw.org/t.co/pemovF0kqz - 195 selected open-source repositories

Joined October 2012
Photo by Rexei (@iamrexei)Photo by Rexei (@iamrexei)Photo by Rexei (@iamrexei)Photo by Rexei (@iamrexei)Photo by Rexei (@iamrexei)Photo by Rexei (@iamrexei)Photo by Rexei (@iamrexei)Photo by Rexei (@iamrexei)Photo by Rexei (@iamrexei)Photo by Rexei (@iamrexei)
4-HOUR STREAM. 3 FINISHED CLIPS. 13 MINUTES. that's the run this creator shows with his Claude Code skill it finds the moments, ranks them, then builds vertical clips with facecam, hooks and word-by-word captions he also walks through a free local setup using Ollama. save this before you spend another evening scrubbing through VODs
2
74
THIS GUY TURNED CLAUDE INTO A FUCKING ANIMATION STUDIO a 90-second Roblox-style story with characters, voiceover, music, sound effects and captions the setup: Claude Code + a custom skill + Flashloop + Blender. He picks the character design. Claude handles the production workflow 8:28 showing the setup and finished results save this for your next faceless channel experiment.
1
1
17
1,404
HOLY S**T THESE ANIMATED SHORTS START WITH ONE LINE OF TEXT this creator shows an engine where AI writes the code, then renders the animation with voiceover DeepSeek, GLM and Gemini get tested against Claude. Actual videos shown side by side 4:22 covering the workflow, results and setup save this for your next faceless channel experiment
3
653
THIS CLAUDE PLUGIN MAKES STICKMAN VIDEOS aiTrends shows the full workflow: pick a story, set up the voiceover, and turn it into an animated Short Hypothetical: 12 client videos at $350 each = $4,200 in revenue Save this for your next faceless content idea
1
5
177
THIS CLAUDE WORKFLOW MAKES STICKMAN ANIMATIONS Jacksons AI shows how to turn a script into an animated video using Claude and a plugin Hypothetical: 4 client videos at $300 each = $1,200 in revenue Save the 13-minute walkthrough for your next content service idea
1
2
142
OPENAI IS GIVING CHATGPT PRO AND BUSINESS SUBSCRIBERS 100% MATCHED API CREDITS openai quietly rolled out bundled API credits for paid ChatGPT Pro and Business tiers, effectively turning your web subscription into developer allowance ▸ Pro subscribers get an exact 1:1 match in monthly API credits: $100, $200, or $500 depending on the tier ▸ Business plans receive $20/mo per Standard seat and $100/mo per Premium seat, pooled up to $500/mo per workspace (annual accounts get the full year upfront) ▸ credits cover chat, Agents API, and developer tools, but expire at the end of each billing cycle with no rollover ▸ to claim, link your account manually under Billing settings in ChatGPT or the API platform (subscription must be active for at least 7 days) → if you're already paying for Pro or Business, you now have zero financial excuse not to run standalone agents or move repetitive tasks directly to the API link in first reply ↓
1
2
157
GOOGLE DROPPED GENIE 4 AND REAL-TIME WORLD MODELS ARE GETTING SCARY google just released Genie 4, and it shifts world models from passive video generation to fully interactive, promptable simulations ▸ real-time generation with native audio: ambient sounds, adaptive music, and character dialogues generate simultaneously with video and react to user input ▸ live prompt steering: direct scene changes on the fly—swap weather conditions, drop in objects, or alter character behavior mid-run ▸ multi-model orchestration: Gemini Omni drives visual fidelity and physics, while Gemini 3.8 Flash runs as an in-the-loop agent that plans tasks and re-evaluates goals ▸ robotics testbed: built to evaluate embodied agents, demoed with Apptronik's Apollo humanoid playing checkers inside a generated environment → world models are no longer just demo videos—they're becoming controllable synthetic training grounds for physical AI link in first reply ↓
1
1
130
8,300 STARS FOR A MARKDOWN FILE THAT BANS THE WORD "SEAMLESS" elayadesign published ai-design-skills — a repo of design skills for coding agents. seven commits, 8.3k stars, 595 forks, MIT, no attribution needed ▸ landing-page-design runs the whole page: intake questions in one batch, layout choice, hero, benefits, FAQ, SEO, then a locked visual system ▸ the visual part is all negative rules — no Inter, Roboto, Arial or Open Sans, no italics, no 900 weight, no background gradients, no Material icons ▸ spacing only from a fixed token table, every font size snapped down to a Tailwind step, nested radius = outer − gap ▸ content bans are the good part: no Lorem Ipsum, no "Acme Corp", no 99.99%, no "Elevate", "Unleash", "Seamless" ▸ free bonus skill design-to-motion turns a static page into a looping MP4 plus an 800×500 GIF for X install: clone skills/landing-page-design into .claude/skills, or paste SKILL.md into AGENTS.md for Codex → the artifact isn't the generated landing. it's a designer writing down 400 lines of "no". taste is mostly subtraction, and that's exactly what a model won't infer from a screenshot link in first reply ↓
2
4
278
763B DEEPSEEK ON A 12GB GPU? HOLY SH*T researchers at IST Austria released a heavily quantized DeepSeek-V4.1-Flash. Strata can run it on a single 12GB GPU by keeping most of the model in RAM and reading the rest from an SSD ▸ RTX 5070 + 64GB RAM: about 5 output tokens/s ▸ RTX 3060 + 46GB RAM: about 8–10 output tokens/s in a separate test ▸ you still need 239GB of disk for the smallest package, or 257GB for the tested 2-bit version → “runs on 12GB VRAM” does not mean “fits in 12GB.” this is an impressive local run, but i wouldn't call it a fast setup for coding agents yet links in first reply ↓
2
5
402
OVER 1,300 FREE 3D ASSETS IN ONE HUGGING FACE COLLECTION Scenario published 1,342 AI-generated 3D examples. You can browse previews and download the original files instead of generating every asset from scratch ▸ 1,500 output files, mostly GLB ▸ also includes Blender scenes, splats, PLY, VOX and FBX ▸ licensed under CC BY 4.0 ❬credit Scenario when you use them❭ the full collection is almost 39 GB, so i'd browse for what i need before downloading the whole thing save this for your next 3D project link in first reply ↓
1
5
277
HOLY SH*T, MINIMAX JUST OPENED THE WEIGHTS OF M3.1 FLASH a 428B MoE model built for coding and agents. around 23B parameters activate per token ▸ up to 1M tokens of context ▸ MiniMax reports 73.7% on DeepSWE v1.1 and 61.7 on OSWorld V2 ▸ the weights are available for self-hosting and independent tests i've already tried it in OpenCode. on my tasks, it responded fast and got the work done quickly. that's my first impression, not a measured speed test → 23B active doesn't mean you only need to store 23B. the full model is still huge link in first reply ↓
1
3
117
YOUR AI TEAM SHARES ONE COMPUTER. HOLY SH*T, THAT’S THE PART I’D FIX FIRST In my Grok Bot vs Dots article, multiple Grok Bots under one user share a machine. If I built my own agent team, I’d want separate mailboxes, separate environments, and a real stop before Send 6 repos for those pieces: Open Agent Email ▸ one mailbox per agent, with an optional approval hold on outgoing mail ▸ github.com/Anil-matcha/open-… Kata Containers ▸ runs workloads in lightweight VMs instead of sharing the host kernel ▸ github.com/kata-containers/k… Agent Sandbox ▸ manages agent environments in Kubernetes; can use Kata or gVisor for isolation ▸ github.com/kubernetes-sigs/a… Presidio ▸ detects and redacts sensitive data in text and images ▸ github.com/data-privacy-stac… AwaitHumans ▸ pauses your workflow until a person returns an explicit decision ▸ github.com/awaithumans/await… detect-secrets ▸ scans code for keys before they end up in a commit ▸ github.com/Yelp/detect-secre… → none of these are plug-ins for hosted Grok Bot or Dots. And Presidio won’t tell you whether an invoice amount is correct. Before approving Send, I still want the original invoice, the recipient, and the exact draft on one screen Save this if you’re giving agents access to real customer email
1
4
160
YOUR AGENT IS BURNING TOKENS BEFORE IT DOES A SINGLE F*CKING THING In my Grok Bot article, the expensive part isn’t just the answer. It’s every browser step and every routine that wakes up to find nothing new 5 repos I’d use to cut that waste in an agent stack I control: MCP Servers ▸ reference Fetch, Git and Filesystem tools that show how to give an agent direct access without browser clicks. This isn’t a ready-made Gmail connector. ▸ github.com/modelcontextproto… MCP Compressor ▸ shows the agent a small tool list, then loads the full schema only for the tool it picks ▸ github.com/atlassian-labs/mc… Token Optimizer MCP ▸ measures and reduces context returned to local coding agents ▸ github.com/ooples/token-opti… Langfuse ▸ traces calls, tokens and cost so you can find the expensive step ▸ github.com/langfuse/langfuse Trigger dev ▸ runs scheduled jobs with logs and retry controls, outside an endless agent chat ▸ github.com/triggerdotdev/tri… → first remove work the agent never needed to do. A routine checking every 15 minutes still wakes up 96 times a day, even if its reply is one line These tools won’t change Grok Bot’s quota for you. Save them for workflows where you control the calls.
My Grok Bot limits ran out in two days, and I got tired of it So I found a way out of this situation and compiled tips in a single article
1
5
189
HOLY SH*T, AN APPROVE BUTTON IS USELESS IF THE AGENT MADE UP THE INVOICE AMOUNT My Grok Bot vs Dots article asks one question: would you let an agent find an overdue invoice and press Send? For a self-hosted version, these 4 repos address different parts of that decision: Invariant ▸ can block a get_inbox → send_email tool-call sequence when the recipient is outside your company ▸ github.com/invariantlabs-ai/… Deliberate ▸ pauses a LangGraph agent and shows the approver the amount, reasoning and evidence you pass in ▸ github.com/beomwookang/delib… Paperless-ngx ▸ gives the agent a searchable invoice archive to cite instead of guessing from memory ▸ github.com/paperless-ngx/pap… Bubblewrap ▸ limits which files a local Linux agent process can see ▸ github.com/containers/bubble… → these aren’t add-ons for Grok Bot or Dots. They show what I’d want to control in my own agent. And a tool-call rule won’t catch a browser click unless that action passes through the same guard Before I approve Send, show me the invoice, the amount and the recipient. Save this for that test
2
3
127