Get Started

Quickstart

Set up Braingent in about ten minutes — clone, copy, point your agent, capture your first task.

This is the fastest path from zero to your agent searches Braingent before planning and writes to it after work. Four steps. Around ten minutes.

By the end you’ll have:

A private memory repo at ~/Documents/repos/braingent (rename freely).
Personalized entrypoints for Claude, Codex, ChatGPT, and Gemini CLI.
One captured task record committed in Git.
Search and validation working locally.

Tip: You don’t need to install anything fancy on day one. Markdown works without a CLI. The optional braingent CLI just makes things faster.

Prerequisites

The day-zero install is genuinely small:

Git — to clone and track your memory repo.
A text editor — anything that opens Markdown.
At least one AI agent — Claude Code, Codex, ChatGPT, or Gemini CLI.

That’s it. You can complete this entire Quickstart with only those.

Optional, in roughly the order most users add them:

rg (ripgrep) + jq + sqlite3 — fast search and index inspection.
gh (GitHub CLI) — pull PRs and issues into memory.
Python ≥ 3.14 + uv — runs the helper scripts (doctor, find, recall, validate, reindex, synthesize, task-*).
Bun ≥ 1.3 — runs the Local Dashboard and the website. Not used by the helper scripts.

See Installation for the full breakdown of what each one buys you.

Step 1 — Clone the manifesto

Get the public starter pack. It’s MIT-licensed and contains everything you’ll need to bootstrap.

git clone https://github.com/thedoublejay/braingent-manifesto
cd braingent-manifesto

This repo is your reference. You will not commit work here. Your real memory will live in a separate private repo.

Step 2 — Copy the starter pack into your private memory repo

Create your memory repo wherever you keep code. The convention is ~/Documents/repos/braingent, but anywhere works.

mkdir -p ~/Documents/repos/braingent
cp -R starter-pack/* ~/Documents/repos/braingent/
cd ~/Documents/repos/braingent
git init
git add .
git commit -m "feat: bootstrap from braingent starter pack"

You now have a working memory repo with:

CLAUDE.md, AGENTS.md, CHATGPT_PROJECT_BRIEF.md, GEMINI.md — agent entrypoints.
preferences/ — your standing rules and policies.
templates/ — frontmatter-stamped templates for every record kind.
workflows/ — named procedures agents follow on trigger phrases.
Empty trees for tasks/, decisions/, reviews/, learnings/, repos/, projects/, topics/, and tools/.

Step 3 — Personalize the placeholders

Open the entrypoint files and replace the obvious placeholders:

<YOUR_NAME> → your name or handle.
<YOUR_ROLE> → “senior backend engineer”, “founder”, whatever fits.
<MEMORY_PATH> → the absolute path to this repo.
Any company-specific tags or repo names.

Keep it small. You’re not building a profile — you’re seeding signal that the agent will reference before planning.

Tip: If you also use Braingent’s optional CLI, you can run braingent init to walk through these prompts interactively and stamp the placeholders for you. See Installation.

Step 4 — Wire up your agents

Tell each AI tool to read from this memory repo before planning. The starter pack includes ready-to-paste entrypoints; the only thing you do is drop them in the right place.

Claude Code — already auto-loads CLAUDE.md from the repo root. Done.

Codex / Codex CLI — the file is AGENTS.md. Codex picks it up automatically.

ChatGPT (custom GPT or project) — paste the contents of CHATGPT_PROJECT_BRIEF.md into the Project’s “Instructions” field.

Gemini CLI — Gemini reads GEMINI.md from the working directory.

For details, see Wire Up Your Agents and the Integrations pages.

Step 5 — Capture your first task

Open a regular working repo (not the memory repo) and finish any small piece of real work — a bug fix, a refactor, a config change. Then say it out loud to your agent:

capture this — short summary of what we just did

The agent will:

Resolve the right memory path (~/Documents/repos/braingent).
Pick the correct record kind (task, decision, review, or learning).
Generate frontmatter with id, title, status, tags, repos, date.
Write the file under the correct directory.
Commit it with a message like capture: <slug>.

Open the file in your memory repo. Read it. This is the smallest possible unit of memory that survives across sessions, agents, and tools.

You’re done

That’s the loop. From here it’s just repetition:

Tomorrow, your agent searches ~/Documents/repos/braingent before planning a related task.
Next week, a different agent reads the same record and inherits the context.
Three months from now, you git log your memory repo and watch the decisions you forgot you made.