Felix

Replacing voicemail hell for HVAC crews. CEO @priovoxai building the AI voice receptionist that actually books jobs 24/7.

[Felix @lastonetocode]

One of those simple yet awesome ideas you see and think... why didn't I think of that

[Andrew McCalip @andrewmccalip]

Get paid to wait

The Claude Code spinner might be the most watched line on Earth.

So I turned it into an ad marketplace.

Advertisers bid on it. You keep 50% of the money.

Install the extension → get cash from ads.

Introducing Kickbacks

[Felix @lastonetocode]

@enzo_gte Where would you say the sweet spot is in terms of effort level for fable?

[Felix @lastonetocode]

Picked plain Opus 4.8 in the Claude Code desktop app. Not the 1M context version.

It hands me 1M anyway — 66k / 1.0M staring back at me.

So the context-window choice just… doesn't stick? Anyone else hitting this, or is there a fix I'm missing?

[Felix @lastonetocode]

spent all day building an internal panel for running demos with prospects. register a demo agent, flip it on, dial the lead, route the
call summary to email. clicked through the whole thing, every button green. "sent", "bound", done. felt great.

then i actually checked the backend. it'd been lying to me the entire time lol.

one config field had a doubled prefix so its env var never got read, which meant the post-call webhook was silently hitting the prod URL
even on dev. and that "email sent" popup? the send was a straight no-op because the key wasn't set. just a cheerful green checkmark
sitting on top of nothing.

only caught both by reading the logs and the stored email record, not the screen. the frontend will smile at you while quietly doing
nothing. check what's actually running.

[Felix @lastonetocode]

Nothing was on fire. That was the problem.

Our voice AI opened every call with last week's script for 2.5 hours yesterday. No crash. No error. The dashboard said the new version
was live. It wasn't.

Turns out the first sentence an agent speaks lives in a completely different field than its "brain." We updated the brain, shipped,
moved on — and every call kept opening with the exact wall-of-text intro we were trying to kill.

The two people we actually reached both bailed mid-sentence. Which was the bug the update was supposed to fix.

"Looks deployed" and "is running" are two different facts. Learned that one the loud way.

[Felix @lastonetocode]

@ben_sdl Currently in the cold calling trenches and that can be straining, this method is even more so I'm sure. Keep at it!

[Felix @lastonetocode]

@stspanho This is genuinely super cool. How easy is it to vibe code your own app for Meta Ray-Ban display?

[Felix @lastonetocode]

@perplexity_ai requiring a minimum of $50 to top up my API credits is not only stupid but also annoying. I already have my pro subscription, which no longer gives me any API credits. Now if I want to give Claude Code or Codex access to perplexity search, which is really the only way I still use Perplexity, I have to top it up with at least $50. Why did you guys make this limit so high? This sucks!

[Felix @lastonetocode]

This is why I can never leave this app. Need to create a demo video for my product, need to learn how to use remotion, and bam this legend has a full explainer from literally yesterday.

[Maciej Dziuba @mm_editing]

Remotion just 10x'd AI motion graphics

This new update is crazy

Here are some use cases:

[Felix @lastonetocode]

Like everyone I have been growing somewhat frustrated with Claude Code's patchy performance. Sometimes its great, sometimes its like talking to the village idiot. I want to try Codex but have everything set up in CC. Wondering if anyone has any tools to help migrate?

[Felix @lastonetocode]

@k0k1eth Würde mich über ne Einladung freuen

[Felix @lastonetocode]

@_lhermann I could really use something like this, what is the tool? Would love to try

[Felix @lastonetocode]

@Jason I'm in

[Felix @lastonetocode]

One of the biggest wins with Claude Code in terms of designing is to have a prototype sub folder. Any new page you are working on, first have it generate a visual HTML prototype, you can iterate quickly and easily before wiring up the backend

[Kingsley Ken @ui_kingsley]

Okay I owe you guys an apology.

I've been using AI wrong this entire time. literally burning hours in figma trying to build sections when Claude Design does the same thing in minutes.

And somehow it's responsive??.

Take my money Claude, all of it.

I'll build out more sections, and share progress with you fine people. 🎯

Yeah, whole stack is getting redone.

#claude

[Felix @lastonetocode]

Absolutely epic

[Banana @banana_baeee]

Guys look Claude helped me - a random guy in his basement - build a wetlab and do vibe genomics!

I sequenced my whole genome despite zero lab experience, without my DNA leaving home!

I put together my notes and a step by step guide here:

https://t.co/T5x6PKkwjW

It was a lot easier than I was expecting!

Ultimately I hit ~16x coverage and compared my results against my 600k raw 23andme SNPs, and it held up!

[Felix @lastonetocode]

@CoachDanGo Psyllium husk helped my digestion a ton, so been a fan for a while, but never new about needing water. I take it in the morning and usually drink more water after lunch. Will take this into account

[Felix @lastonetocode]

@irabukht @resend This looks amazing! Can't wait to try it myself, thanks for sharing the skill

[Felix @lastonetocode]

@theonejvo Is that because 0days were just so expensive or required so much effort it didn't make sense for general pen testing? Because then if you could just prompt a 0day wouldn't that make it trivial enough that even script kiddies are going to be using 0days?

[Moshe Schwartz @YWNReporter]

🚨 DEVELOPING: Iranian media is circulating images of ID cards and documents allegedly left behind by U.S. forces during the dramatic rescue of a downed American pilot.

The documents are being linked online to “Amanda M. Ryder” — but early analysis raises major doubts about that claim.

Officials have been referring to the rescued pilot as a “Colonel,” while the ID shown does not match that rank. In addition, the individual appears to be a maintenance technician — a role considered highly unlikely to be part of a high-risk rescue mission.

Sources reviewing the materials say the items were likely recovered from U.S. aircraft destroyed during the operation — reportedly 2 C-130 planes and 3 helicopters — rather than belonging to the rescued airman.

Adding to the mystery, one of the documents includes an Israeli B2 entry permit, typically issued for tourism or business purposes.

Bottom line: The viral claim about the pilot’s identity appears inconsistent with the known details — and the story behind these documents is only getting more complicated.

[Felix @lastonetocode]

Open source and local is accelerating at a crazy pace

[Ben Sigman @bensig]

My friend Milla Jovovich and I spent months creating an AI memory system with Claude. It just posted a perfect score on the standard benchmark - beating every product in the space, free or paid.

It's called MemPalace, and it works nothing like anything else out there.

Instead of sending your data to a background agent in the cloud, it mines your conversations locally and organizes them into a palace - a structured architecture with wings, halls, and rooms that mirrors how human memory actually works.

Here is what that gets you:

→ Your AI knows who you are before you type a single word - family, projects, preferences, loaded in ~120 tokens
→ Palace architecture organizes memories by domain and type - not a flat list of facts, a navigable structure
→ Semantic search across months of conversations finds the answer in position 1 or 2
→ AAAK compression fits your entire life context into 120 tokens - 30x lossless compression any LLM reads natively
→ Contradiction detection catches wrong names, wrong pronouns, wrong ages before you ever see them

The benchmarks:

100% recall on LongMemEval — first perfect score ever recorded. 500/500 questions. Every question type at 100%.

92.9% on ConvoMem — more than 2x Mem0's score.

100% on LoCoMo — every multi-hop reasoning category, including temporal inference which stumps most systems.

No API key. No cloud. No subscription. One dependency. Runs on your machine. Your memories never leave.

MIT License. 100% Open Source.

https://t.co/KggwTqijmD