Siddharth

should I fire my marketing intern?

@p4cs_pedro @steipete control works for all displays, ultrabright only for Pro Display XDR and MacBook Pro displays

@sids7 @steipete does it only work on macbook display, or third party too?

Codex's ability to reverse engineer private macOS frameworks is uncanny. Just added an HDR toggle to DisplayBuddy - so many users asked for this!

@adityadaniel honestly just ask it directly. I asked it to "find a way to enable/disable HDR per display on macOS" and it did! previously ChatGPT used to refuse touching private frameworks, but codex just goes brrrr

@sids7 Do you mind sharing the prompt or how to do this?

@BMP46511 gpt-5.4 high or medium

@sids7 What model are you using?

@p4cs_pedro @steipete Yep it sure does! It's called UltraBright

@sids7 @steipete does displaybuddy do that thing where you can go over the monitor’s brightness limit?

@oskargroth Can the hopper MCP + Claude even reverse engineer other apps? Like what private APIs other apps are using?

Wasn't joking about this one btw You can reverse-engineer pretty much any part of Apple platform internals in seconds using Claude or Codex with Hopper MCP

@oskargroth Wow, gotta try that! Thanks!

@sids7 Connect to Hopper MCP and it’s basically as good as having access to all of Apple’s internal source code

@DavidOndrej1 I felt the same codex-5.3-spark. It’s 4x faster than Claude’s fast mode

using FAST MODE with Opus 4.6 inside of Claude Code is a total gamechanger the inference feels at least 3x faster (not double) it's a huge unlock for productivity. anything feels possible now.

@andrewxroas Any advice on scaling Mac apps? Apps that cannot be on the store due to sandbox restrictions

I've scaled my app to 600k MRR I want to give you some sauce on scaling apps, distribution, running ads, etc. Drop your questions below, I'll try to answer as many as possible

@rudrank You should! I’d love to come for that talk!

@sids7 I want to apply as a speaker haha 😬

Anyone going to AIE Singapore? Looks interesting ai.engineer/singapore

TLDR: big companies are providing one click OpenClaw installs because they want their cloud inference capacity to be used

@averycode I’d love to join!

Our indie hackers community is having a meetup this thursday - come have coffee and cowork with us! 🇸🇬

What a day 🚀 I asked my OpenClaw to update DisplayBuddy release notes, clone the repo, set up `gh`, open a PR, and I merged it from my phone 📱✅ Then it drafted + posted the release tweet too.

@k2xl @OpenAIDevs It uses 2x of your limits

@OpenAIDevs So if it is the same why isnt it always enabled

Codex got more speed. With /fast mode, GPT-5.4 runs 1.5x faster with the same intelligence and reasoning. Move through coding tasks, iteration, and debugging while staying in flow.

@hieudinh_ Love it!

My weekly SEO workflow is now a single Claude Code command that does: 1. Pull GSC data 2. Detect keywords ranking 4-15 3. Expand with Keywords Everywhere API 4. Auto-write & publish blog posts Follow and reply 👋 if you want to steal it.

This exactly echoes my thoughts when I first tried spark at the OpenAI hackathon last week. Once you try spark, there’s no going back. It makes the $200 pro plan seem cheap. It’s so fast that I sat down to do one thing, and ended up burning 30% of my todos for @DisplayBuddy

@krunkosaurus Are you coming to this? luma.com/8b1sz5k8?tk=50…

I will say it: if you don't want to manage your own OpenClaw instance just sign up for a Manus AI plan for $20/mo. - I told it to use the Sogni-gen plugin from Clawhub and now it can do image and video generation in my chats with it. - I told it to use the custom voice of Jarvis from Sogni Voice and it's now sending me voice note summaries of entire blog posts I send it. I still use OpenClaw but for normies it's really nice, simple, and just works. Manus AI still dominates on web design, slide presentation design, etc over ChatGPT or Claude, too.

I’ll have to quit my job if I have to read all the 469,283 ways to make money with OpenClaw articles that I’ve bookmarked

@krunkosaurus Fantastic. You might already know this, but try llmfit, will tell you exactly which models run fine on each Mac github.com/AlexsJones/llm…

Did the math on running Qwen3.5-122B-A10B (second biggest model) locally on the M5 MacBook Pro Max (128GB). 122B params @ Q4 = ~61GB weights. MoE activates ~10B params per token, so you’re only reading ~6.5GB of weights per forward pass. At ~614 GB/s memory bandwidth, even with KV cache overhead at 64K context (maybe even 128k!) you’re looking at roughly 26-40 tok/s. Not frontier, but covers 95% of real work — code, writing, reasoning, image/OCR, tool use. No 3rd party API, no latency, no per-token costs. Ordering in 12 hours. Ships next week.