greythinkinglab is an applied AI lab building a personal AI assistant that remembers, follows through, and thinks before it answers.

We want AI that thinks in the grey — that remembers what matters, weighs it, and reconsiders. We're building it as three connected pieces: memory, follow-through and reasoning. Each ships on its own. Together they are the assistant.

  1. 01 — MemoryMurmurCarries your context across every AI tool, local-first. The assistant's memory.macOS · launching Q1 2027
  2. 02 — Follow-throughGrittoAn assistant that helps you keep the commitments you set for yourself, starting with fitness. The assistant's first job.iPhone + Apple Watch · launching late 2026
  3. 03 — ReasoningDeliberationHow an AI holds a problem, critiques its own draft and revises. The assistant's ability to think.Research · open

01—Thesis

Brilliant, and forgetful.

Language models can write, code and explain almost anything. But every conversation starts from zero, and every answer is a first draft. The model doesn't know what you decided last week, and it rarely stops to ask whether it's wrong.

We think the next step isn't only bigger models. It's two missing abilities: memory, so context follows you, and deliberation, so a model can hold a problem, revisit it and change its mind.

An assistant with both could do something today's tools can't: know what you've committed to and help you follow through. That's what we're building — one layer at a time, starting with memory and with personal fitness as the first area it serves.

Grey thinking is the space between a quick yes and a quick no — where you weigh, remember and reconsider. That's the kind of thinking we're trying to give machines.

02—Research agenda

Two open questions.

01Memory

Applied · shipping as Murmur

What should an AI know about you, across every tool you use, without you having to repeat yourself — and without giving up your privacy?

  • — Retrieving the few moments from past conversations that matter for what you're asking now, not the whole transcript.
  • — Memory that works across Claude, ChatGPT, Cursor and anything else, instead of inside one vendor's app.
  • — Local-first storage and on-device speech, so personal context stays on your machine.
Applied in Murmur →

02Deliberation

Open problem · early

How do you give an AI the ability to think — not answer in one pass, but hold a problem, revisit it, and notice when it's wrong?

  • — Loops where a model can pause, critique its own draft, and continue with memory of what failed.
  • — A working scratchpad that persists across sessions, not just a chat transcript.
  • — Whether "thinking" is better as a product surface people feel, or as infrastructure under other tools.

Exploratory. No ship date — this one stays open until we understand it.

03—Products

macOS · launching Q1 2027

Murmur

The memory layer. The first product from our memory research.

Murmur is a macOS app. Hold the fn key, speak, and the right moments from your past AI conversations come with you — into Claude, ChatGPT, Cursor, anywhere you type.

Building Murmur is how we test our ideas about memory against real work: what's worth keeping, what to bring back, and what should never be captured at all.

Launching Q1 2027. The early access list is open now.

claude.ai · murmurcontext attached

Should I use Stripe Tax or handle VAT manually?

  • TueB2B SaaS for European customers, €10–50/mo
  • TuePicked Stripe over Lemon Squeezy for EU VAT
  • Tue~100 customers expected year 1

3 memories attached · ready to send

How Murmur works

  1. 01

    Capture

    Hold fn and speak. Audio is held in memory only, and text is read only from apps you've allowed.

  2. 02

    Transcribe

    whisper.cpp runs on your Mac. Audio is discarded as soon as it's transcribed.

  3. 03

    Organise

    Transcripts and AI sessions go into a local database, grouped by project.

  4. 04

    Retrieve

    For what you're asking now, Murmur finds the few past moments that matter — decisions, constraints, numbers.

  5. 05

    Attach

    Those moments are added to your prompt before you send it, in whichever AI tool you're using.

On-device speech. Local-first memory. Capture off by default.

iPhone + Apple Watch · launching late 2026

Gritto

The follow-through layer. An assistant for the commitments you make to yourself, starting with fitness.

gritto.app ↗More at launch.

05—About

greythinkinglab is an applied AI lab. We're building a personal AI assistant that remembers what matters to you, helps you follow through on what you set out to do, and reasons before it answers. We build it as three connected pieces, each shipped on its own: Murmur, the memory layer; Gritto, the follow-through layer, starting with fitness; and Deliberation, our research into how an AI can hold a problem and reconsider. We publish what we learn and ship what works.

Founder

Nik G

Senior iOS engineer with eleven years of shipping apps, including medical-device software. I lead the research and build the products, and make videos on tech, life and whatever I'm thinking about.

Also made here: SlotOwl, appointment alerts for US visa, Mexico INM and Schengen portals.