My working toolkit
This is how I keep ideas, reading and unfinished work within reach. Each tool has a job; the useful part is how they fit into an ordinary day.
Readwise catches what I want to keep. Obsidian holds the record. AI helps me find, question and make something from it. I keep the judgement and decide what gets shared.
Read the full story: How I Use AI Day to Day ↗The foundation
Catch it before it disappears
A forgiving place to save reading and highlights while they still have my attention. The useful material can then connect to notes and projects in my vault.
Save an article now; come back to the question it raised when I have room to think.
Keep the record
Daily notes, project decisions and handoffs give me something reliable to return to. The context lives here, so changing AI tools does not mean starting the story again.
Reopen a project after a few days away and find what happened, what I decided and the next step.
The assistants
Talk an idea into shape
A place to think aloud, question an idea and begin a draft. A conversation often helps me find the point before I know how to write it.
Catch an idea from a conversation before I forget it, then work out whether there is an article in it.
Work with the knowledge I have kept
Helps me work through notes, sources and writing in the vault, using a defined scope and the evidence already there.
Bring a post seed, experiment record and daily notes together so a draft starts with what actually happened.
Build, check and prepare
Works with me on website code, practical projects and defined tasks in the vault. It also helps prepare inbox triage and reports that make the next decision easier.
Turn a website idea into a local preview, check it, and let me review it before it goes live.
Carry useful work between sessions
The memory, tools and saved procedures around the model, rather than another model itself. I’m exploring how it carries preferences and proven steps into a fresh session, including when I change providers. Remembering a boundary is not the same as enforcing it; a desktop app does not make every connected model local.
A fresh session recovered my operating preferences and reused a saved health check. An existing knowledge workflow also ran after separate approval. Promising continuity, with longer-term usefulness still being tested.
Start a conversation already briefed
Another thinking partner, given selected context from my system. That briefing makes it useful without assuming it remembers my other conversations.
Pick up a question with the relevant background already supplied, instead of explaining the whole project again.
Find the conversation
I increasingly use Grok for quick searches, particularly around Reddit and X. I also use Grok bot to keep up with Substack Notes that interest me.
Find a discussion worth reading or a Note worth returning to. I still check the source and decide what deserves a response.
Used less often · Still exploring
I have a working local setup, but I am reaching for Gemma less often. Privacy and control still make local AI interesting to test. It has not become the centre of my daily work.
See the desktop test and its limits →If you are starting out
Start with a daily note: what happened, what you decided and what comes next. Add a tool when there is a specific job you want help with. You do not need my entire setup to make that useful.
See how it works through a day ↗