Marc Maurer

building

[Jan @j_ebbinghaus]

europe is cooking. just in sf

[Damian Zaker @San_Damianoz]

Sundays are for bringing European founders & friends together to 🍜

@j_ebbinghaus and @maurermarc are cooking and becoming regulars here

@otmanesab from Zauber catching some SF air

@sebastien_hess1 crafting his hone

@MaxThoelen finally moved over

@Marco_Henri_ about to move over

@YamamotoPhD revolutionizing bio

Join us next time!

[Marc Maurer @maurermarc]

@DavidPawlan Would love to join!

[Marc Maurer @maurermarc]

@emankhoubian for sure something to be mindful of when yapping - took a minute to get myself to not do this accidentally

[Marc Maurer @maurermarc]

Built a little tool for myself and figured some of you might find it useful.

I love Wispr Flow, but when building UI I kept describing edits, taking screenshots and trying to explain what I was pointing at. Not a great experience...

[Marc Maurer @maurermarc]

@luisascharff yep any agent - simply works via MCP

[Marc Maurer @maurermarc]

@topphamilia Exactly, felt the same way when using voice only!

[Marc Maurer @maurermarc]

Runs fully on your Mac, no account. Made it free and open source in case it helps anyone else.

https://t.co/dRZOZpfqYL

[Marc Maurer @maurermarc]

Everything goes straight into my coding agent via MCP, including a screenshot of what I pointed at, the element under the cursor and what I said. Claude Code, Codex or Cursor then fix it and mark it done.

[Marc Maurer @maurermarc]

@karolzdeb Aren't both TT and IG using device motion signals, to detect phone farms? Then those will get flagged and shadow banned in no time, no?

[Marc Maurer @maurermarc]

@HosseinGorji021 Mix of all, really still no solution that just simply 'works' without manually tuning it all the time

[Marc Maurer @maurermarc]

@VictorTaelin Really dope! How would you suggest to coordinate across multiple agents? Like swapping between Claude, GPT, etc.

[Marc Maurer @maurermarc]

@tseglevskiy Biggest unlock will be once the model just works with a central continuously updating memory from every part of your life. Across all agents and all tools...

[Marc Maurer @maurermarc]

@morganlinton Is this then staying synced or is it a one way import?

[Marc Maurer @maurermarc]

@aka_ssy Agree, also many of the modern memory infra providers overcomplicate their architecture to lean into having a technical edge. In 90% of cases this infra is not even required

[baba yaga @babayagatwt]

OpenAI should add a switch: let Codex use my ChatGPT memory

you can plan a project in ChatGPT with all the decisions and constraints, then open Codex and explain everything again because that context does not carry over.
(OpenAI docs confirm this separation)

this is not a new request either. people have been asking for ChatGPT and Codex memory to work together for nearly a year.

Not asking for Codex to ingest every chat. That's a mess and ChatGPT's idea of the code goes stale. Asking for an option, off by default:

- choose what Codex can access (saved memories only or one project)
- see what context Codex is using
- let users exclude anything or turn it off

If the product line is "one account, ChatGPT and Codex" the memory should be able to cross when you say so. Right now the user is the sync layer.

[Marc Maurer @maurermarc]

@babayagatwt Would be best if this would even extend across models. So that the GPT memory is available across Codex but also CC, Grok, etc.

[Jan @j_ebbinghaus]

I love how Muse is essentially distributing free, entry-level (virtual) laptops around the world. Pretty unprecedented. @alexandr_wang