
Deep in Agents, LLMs and automation · Open source at ♥ · Sharing AI gems daily to help you work smarter · Ex-data scientist @Snowflake | @Streamlit ·
📩 DM for Collabs
Joined January 2009
- Tweets36.5K
- Following49.7K
- Followers184K
- Likes112K
Pinned Tweet
people don't realize what just happened..
one of @deepseek_ai's earliest members just open-sourced a personal agent that runs on your PC.
12 local models. up to 5x faster inference. 0 token fees for local use.
@HeyLingerAI is your local GrokBot.
on your machine.
open-source.
In the 1960s, computing was centralized until the PC put power into everyone's hands. Now Al is trained in datacenters and rented back to you token by token.
Today we open source Linger (@HeyLingerAl), your personal agent living on your own computer.
We believe intelligence should be distributed. Your context, your data, and the intelligence that powers them should be owned entirely by you.
Github: github.com/simplexai-labs/li… Homepage: heylinger.com
Grok @Bot users after reading this:
I think a lot of people aren't using Grok Bot to its full potential, because knowing how to automate things is actually quite a difficult skill to learn. I wrote this post to break down how I approach automating anything with Grok Bot!
1
2
384
Charly Wargnier retweeted
Claude = coding. ($20/mo)
GitHub = version control. (Free)
Supabase = backend. (Free)
Clerk = auth. (Free)
Resend = emails. (Free)
Vercel = deploying. (Free)
Cloudflare = DNS. (Free)
Upstash = Redis. (Free)
Pinecone = vector DB. (Free)
PostHog = analytics. (Free)
Sentry = error tracking. (Free)
Stripe = payments. (2.9%/transaction)
Namecheap = domain. ($12/yr)
Total monthly cost to kick-off your startup: ~$20
There has never been a cheaper time to build
Charly Wargnier retweeted
people don't realize what just happened..
one of @deepseek_ai's earliest members just open-sourced a personal agent that runs on your PC.
12 local models. up to 5x faster inference. 0 token fees for local use.
@HeyLingerAI is your local GrokBot.
on your machine.
open-source.
In the 1960s, computing was centralized until the PC put power into everyone's hands. Now Al is trained in datacenters and rented back to you token by token.
Today we open source Linger (@HeyLingerAl), your personal agent living on your own computer.
We believe intelligence should be distributed. Your context, your data, and the intelligence that powers them should be owned entirely by you.
Github: github.com/simplexai-labs/li… Homepage: heylinger.com
Charly Wargnier retweeted
RAG IS COOKED FOR GOOD
Researchers just proved an LLM can work with a billion-triple knowledge graph using less than 20GB of GPU VRAM.
no separate vector stack. Pinecone, Weaviate, Qdrant: you can sit this one out.
it’s called AtlasKV, and it tackles the whole “give AI more knowledge problem differently.
❌ Skip the external retriever
❌ Skip cramming retrieved docs into a massive prompt
✅ Just feed structured knowledge directly into the model’s attention instead
2 pieces can make this possible:
→ KG2KV converts graph facts into key-value representations.
→ HiKVP progressively filters the knowledge hierarchy, keeping the GPU workload manageable.
and the numbers are just INSANE:
> less than 20GB of GPU VRAM at billion-triple scale
> add new knowledge without another round of training after initial setup
> no external retrieval module
> no huge block of retrieved text eating your context window
the clever part?
most knowledge stays in CPU memory.
selected pieces move onto the GPU as needed, though those transfers still add latency.
we’ve spent years tuning chunk sizes, retrievers and prompts.
seeing researchers rethink that entire workflow is pretty damn exciting!
Pretty cool!
I gave IQuest-Q1, the open-source model from @IQuest_research, one task:
"Simulate two spiral galaxies colliding, interactively. One HTML file. No libraries."
Round 1 (~7 min): it wrote everything from scratch.
→ 2 galaxies, 2,500 stars each, one disk tilted, one prograde
→ an in-browser leapfrog gravity integrator
→ a live HUD: time, separation, speed, tidal-tail %
No errors: 5,000 stars at ~53 FPS.
But the galaxies merged in 10 seconds. Nothing else happened.
We sent it the test numbers.
Round 2 (~13 min):
It checked its physics, used Python scripts to scan friction and orbit parameters, traced the best candidates, and reworked the encounter.
Now:
> first close pass at ~16 s
> tidal tails flung out, cores ~39 kpc apart
> the galaxies fall back and merge at ~59 s
> ~1.2 billion years simulated at ~50 FPS
> afterward: drag to rotate, scroll to zoom, change speed
It's a simplified N-body model.
The useful part is the write → test → feedback → fix loop.
we supplied the numbers, and it handled the fix 🔥
Explore IQuest-Q1 and its setup instructions on GitHub:
→ github.com/IQuestLab/IQuest-…
If you find it useful, give the repo a star :)
Charly Wargnier retweeted
NVIDIA and AMD after reading @positron_ai's Titan does "4x better cost per token than next-gen systems"
🤝 Paid partnership
Positron's Titan is joining the Parasail fleet. 5MW. Trillions of tokens a day of cost-optimized compute by late 2027. @parasail_io @positron_ai
@parasail_io is the first multi-silicon inference cloud that routes your prompts to the best chip for the job, serving trillions of tokens a day to hundreds of customers. Positron fills a vital part of compute spectrum on the cost optimized side: 4X better cost-per token than next gen NVIDIA/AMD systems which is perfect for large-scale agent swarms, background agents, and physical AI.
Positron has done incredible work building a chip and system that orchestrates across DRAM and SRAM allowing for massively scalable memory pools with commodity memory giving the best of both worlds for memory capacity and bandwidth.
We've been excited to partner with them in the early days through the launch of their Atlas chip and now big capacity plans with Titan.
@positron_ai
NVIDIA and AMD after reading @positron_ai's Titan does "4x better cost per token than next-gen systems"
🤝 Paid partnership
Positron's Titan is joining the Parasail fleet. 5MW. Trillions of tokens a day of cost-optimized compute by late 2027. @parasail_io @positron_ai
@parasail_io is the first multi-silicon inference cloud that routes your prompts to the best chip for the job, serving trillions of tokens a day to hundreds of customers. Positron fills a vital part of compute spectrum on the cost optimized side: 4X better cost-per token than next gen NVIDIA/AMD systems which is perfect for large-scale agent swarms, background agents, and physical AI.
Positron has done incredible work building a chip and system that orchestrates across DRAM and SRAM allowing for massively scalable memory pools with commodity memory giving the best of both worlds for memory capacity and bandwidth.
We've been excited to partner with them in the early days through the launch of their Atlas chip and now big capacity plans with Titan.
@positron_ai
Claude = coding. ($20/mo)
GitHub = version control. (Free)
Supabase = backend. (Free)
Clerk = auth. (Free)
Resend = emails. (Free)
Vercel = deploying. (Free)
Cloudflare = DNS. (Free)
Upstash = Redis. (Free)
Pinecone = vector DB. (Free)
PostHog = analytics. (Free)
Sentry = error tracking. (Free)
Stripe = payments. (2.9%/transaction)
Namecheap = domain. ($12/yr)
Total monthly cost to kick-off your startup: ~$20
There has never been a cheaper time to build
Charly Wargnier retweeted
This is wild.
This guy turned his unslopping process into a free Claude skill 🤯
Landing pages, dashboards, mobile apps, you name it.
Built something useful, but the UI still screams "AI made this"?
That's what @heydamir's unslop is designed to fix.
It:
> spots generic styling
> proposes a direction based on your product
> then refines the screen one pass at a time
React, Vue, Svelte, Tailwind or plain HTML/CSS.
Free and open-source.
Repo in 🧵↓









