dmthepm/mainbranch

Give your AI the business brain it was missing.
One folder for your offers, proof, research, decisions, launches, ads, pages, bets, and lessons, readable by Claude Code, Codex, and you.
Set up with AI ·
Quickstart ·
What changes ·
What it does ·
Workflows ·
Works with ·
Beginner setup ·
Documentation guide
## Why this exists
Your business context is probably scattered: a Notion page for the offer, Looms
for voice, Google Docs for research, a few chats with good ideas, a launch plan
somewhere, and decisions no one can find when the next AI session starts.
Main Branch turns one folder on your computer into business memory your agent
reads before it answers, then saves approved decisions, launches, bets, pages,
ads, research, and lessons back into files you own.
It is built for revenue-producing work: sharpening offers, building proof,
planning launches, writing ads, creating organic content, checking pages,
tracking bets, and turning what you learn into the next decision.
Open the folder. In Claude Code, run /mb-start. In Codex, use the global
mb-start skill. Tell the agent what you want help with. Main Branch checks
what changed, what matters, and what to do next.
| Step | What you do | What happens |
| 01 | Open the business folder | Your offer, audience, voice, proof, research, bets, pushes, logs, and documents live there. |
| 02 | Start Main Branch | Claude Code uses /mb-start; Codex uses the global mb-start skill. The agent checks current Main Branch facts before giving advice: status, MoneyPath, recent work, connected tools, drift, checkpoints, and next actions. |
| 03 | Pick the next move | The agent routes the work into an offer, bet, decision, push, playbook, research note, outcome, or checkpoint. |
| 04 | Approve the work | Writing files, publishing, spending, account changes, and customer contact stay your call. |
| 05 | Keep the lesson | Approved work becomes durable memory for the next session. |
If you want AI business memory that stays yours, star this repo so more
business owners find it.
## Set up with AI (recommended)
Open your strongest model — Claude Fable with the 1M context window works
great; on Codex, pick the strongest model with extra reasoning — and paste
this:
text
I want Main Branch: AI business memory I own as files in one folder
(https://github.com/dmthepm/mainbranch). First read the entire README (fetch the raw file if your tools summarize),
then explore the repo enough to understand the architecture (the mb CLI is
the deterministic control plane; skills are the judgment layer; my business
lives in its own folder, never in the engine; bets, research, decisions,
pushes, playbooks, outcomes, and checkpoints are the primitives; writing
files, publishing, spending, and customer contact stay my call). Then
interview me briefly about my business, my offer, and my audience. Then:
install the engine, run mb onboard to create my business folder,
then run mb doctor inside that folder and fix anything it flags, seed my
core files from the interview using the templates the folder ships with,
save the first checkpoint, and teach me the daily loop — /mb-start to