Vibe Coding with Vant

The fastest way in: one command, and your project has memory. Every command on this page works with zero configuration - no tokens, no accounts, no YAML. When you want more, the deep docs are linked at the bottom.

On this page

• One minute: give your project a memory
• Five minutes: notes that stick
• Ask your notes
• When you want sync
• A tiny agent habit
• Where the deep docs live

One minute: give your project a memory

In your project folder:

vant memory learn todos "ship the login page, then the settings page"

That’s it. The note is in your project’s brain, on disk, in plain markdown you can open and read.

Five minutes: notes that stick

The same verb does everything - the first word after learn is just a label you invent:

vant memory learn decisions "chose Postgres over Mongo - team already knows SQL"
vant memory learn gotchas "the API rate-limits at 60 req/min - back off on 429s"
vant memory learn vibe "this project uses tabs, not spaces. deal with it."

Short-term scratchpad values (they expire, unlike learned notes):

vant memory state session-id abc123
vant memory recall session-id

Read everything the brain holds:

vant memory list

Ask your notes

Search across everything you taught it:

vant search "why did we choose postgres"
vant search "rate limits" --compact

Search is hybrid (text plus semantic), so paraphrases work - you do not need to remember the exact words you stored.

When you want sync

Everything above works offline with zero accounts. When you want the brain to travel with the repo, add two lines to .env:

VANT_GITHUB_REPO=your-username/your-brain-repo
GITHUB_TOKEN=ghp_xxxxxxxxxxxx

Then push and pull the brain like code:

vant sync --push
vant sync --pull

This is the original premise: ship the memory with the repo. Plain markdown files in a git repo you own - never locked into a vendor.

A tiny agent habit

If you code with an AI agent (Copilot, Claude, Cursor, anything with a shell), teach it two sentences:

Before we start, run: vant memory query decisions
When we decide something, run: vant memory learn decisions "<what and why>"

Now every session starts with the context and ends with the lesson. That loop - remember, decide, record - is the whole methodology at hobbyist scale. When your project grows into a crew of agents, it is the same loop with more structure:

• The brain files an agent actually reads: The Brain
• The wake/work/sleep loop: Agent Onboarding
• One install, many agents: Multi-agent

Where the deep docs live

One honest note: vant health will mention setup and GitHub sync as “not initialized” when you skipped them. That is fine - the memory commands on this page work without any of it. The health check is guarding the full enterprise setup, not the one-liners.