Now, Claude runs as fast as you do.
One trusted place for your decisions. Stop re-briefing every session. Claude picks up exactly where you left off.
No code. No new app. If you can use Google Drive, you can run this.
One folder. One hour. By your second session, Claude already knows where you left off. Or delete the folder, and you've lost nothing.
12projects
158sessions
49+hours tracked
I'm Jeriel. These numbers come from my own live system, pulled fresh, never typed by hand. As of 12 Jul 2026.
AI without structure is just conversation.
Talk to Claude enough and you'll have real insight buried in a dozen threads, with no way to find it, build on it, or act on it twice.
So you re-explain. You track things by hand. You scroll old chats asking "where was I?" That's admin work, and it eats the attention (and the tokens) you'd rather spend on the decisions only you can make.
Small habit. Big multiplier.
It's three small moves, repeated. Do them, and every session builds on the last instead of starting from zero.
- Open. Claude already knows where you left off: the decisions, the why, what's next.
- Work. Do the work. Every decision gets caught and structured as you go.
- Save. Close up, and the next session starts one step ahead, not from zero.
Do it enough times, and you're steering, not scrambling.
This isn't a demo. It's my actual system, today.
Here's a real line from the log, the day this page shipped (11 Jul 2026), on the project that built it:
Fable /review of the Build Spec passed (verdict: ship) … metric recounted **12 projects · 148 sessions** … Spec marked **BUILD-READY**.
And here's the part that's harder to fake: Claude doesn't just remember what I did. It tracks how I work, dated and factual, and proposes promoting a pattern into my permanent profile once it's shown up three times.
Two parts:
A curated Baseline, and a living mirror: dated observations appended from real sessions, approved by me, including the unflattering ones. It reflects, it doesn't direct. When a pattern repeats 3+ times, it gets proposed for promotion into the Baseline.
# PART 2 — OBSERVED PATTERNS (the living mirror)
> Appended by `/offload`, newest first. Rules: **observed, dated, factual — reflect, don't direct.** Flattering and unflattering carry equal weight — a cherry-picked mirror is a portrait.
**[10 Jul 2026 — SB Website, Sess.2]** … drove crisp product direction himself: reframed the site's objective from download-first to comprehension + experience, and chose the defining metric unprompted (total projects + sessions, "no information loss"), rejecting "10x" as undefinable — a further instance of the choose-the-honest-metric-over-an-invented-number pattern. *(observed)*
**[11 Jul 2026 — Education, Sess.11]** Three separate voice-dictation mishearings in one session went uncaught by him at the moment of sending, each only caught by Claude reading the instruction against the actual file state … the pattern held across the whole session rather than being a one-off slip. *(observed)*
Real entries from the live file, trimmed (…). Nothing added, nothing invented. These two are shown because they contain no one's name but mine; yes, the second one is Claude keeping receipts on me.
The four traits that earned promotion:
- Decomposition beats motivation. The gap between knowing and doing closes on its own when work is broken into small steps with visible progress, not by trying harder.promoted from the mirror, 12 Jun 2026 · observed 4+ times
- Guardrail before agency. Power gets granted to a tool, a system, or an AI only after a guardrail exists to make the worst case safe. Never trust-first.promoted from the mirror, 28 Jun 2026 · observed 3×
- Strategy-as-motion, self-corrected. When an exciting new idea threatens to displace an unfinished harder task, it now gets caught and routed back, increasingly before being told.promoted from the mirror, 29 Jun 2026 · observed 3×
- Verify-before-trust. No claim, build, or grade counts as settled on one verification layer, including the system's own.promoted from the mirror, 06 Jul 2026 · observed 10+ times
What you get
- Works with what you already have. A paid Claude plan (Pro, Max, or Team) and your Google Drive. No new subscriptions, no new accounts.
- Free, for now. Personal-use license: yours to keep, free to pass to a friend, not for resale. Get it while that's true.
- What's in the box: the engine (same for everyone, you never touch it), your own private profile (built with you, stays yours), and the handful of commands that run it.
Why I built this
I run a few small businesses, and I'm not a developer. I built this because I was tired of re-briefing Claude every session: the decisions, the reasons, where we left off. It worked well enough that I started installing it for friends, in person, one at a time. The kit is the exact system running all my projects today, minus my personal profile. That stays mine, the way yours stays yours. And I run it under my own red line: the moment it costs more effort than it returns, it has failed.
FAQ
Is my data private?
Yes. Your profile (your working style, your decisions, your patterns) lives in your own Google Drive and is read by your own Claude account. It's never sent to me, never uploaded anywhere else. I can't see anything you build.
What do I need?
A paid Claude plan (Pro, Max, or Team) and a Google account with Drive installed. About an hour to set up.
Why not just use Claude by itself?
Claude's built-in memory is real, but it's vague. It recalls the gist of you, not the exact state of your work. The kit keeps a structured record instead: each project gets one trusted file for its decisions and next steps, in plain files you can open and read, in your own Drive. Claude starts every session from that record, not from a fuzzy recollection. That's the difference between "it sort of remembers me" and "it knows exactly where we left off."
Is it hard to set up?
No. It's a guided walkthrough, one step at a time. See the install guide.
What if I set something up wrong?
The system checks itself. Every session starts with Claude confirming it can actually see and write to your folders before any work begins. Forget a step, like connecting a folder, and you get a simple prompt to fix it in seconds, not a broken system. The install guide covers the common snags too.
What does it cost?
Free, for now. I'm still building this out, and while that's true the kit costs nothing. Personal-use license, yours to keep, free to pass to a friend. It won't stay free forever, so this is a good time. Star the repo on GitHub if you want to hear when things change.
What can it do for me now?
- One trusted place per project
- Starts every session already briefed
- Takes the tracking and "where was I?" off you
- Breaks big things down into the next step
- Works across your machines
- Learns alongside you, at whatever depth you ask
- Demo mode: try before you commit
- Notes that travel between projects automatically
- Sends feedback straight to me