Skip to main content
Driftless is one workspace where a team and its agents share durable state: the Knowledge you keep, the Operations you run, and the Work in flight. This quickstart starts where builders feel it first: your agent reads the relevant context before it touches an area and writes back what it learns. You don’t learn a system. Your agent does. Here’s the whole setup.
1

Install and log in

npm install -g @driftless-sh/cli
driftless login
For CI or non-interactive setup, pass a key instead: driftless login --key drift_your_api_key_here (create one at driftless.icu → Settings → API Keys).
2

Teach your agent

driftless install-skill
This writes AGENTS.md and CLAUDE.md into your repo. From now on, any coding agent that opens the repo knows the loop: pull team context before editing, persist what it learns after.
3

Go code

Work as you normally would. Your agent pulls context before it touches an area, and saves new gotchas and decisions as it learns them. Come back to your dashboard and watch your first memory show up.
That’s it. Pull before, persist after. That’s Driftless.

​
The whole loop, in one prompt

Prefer not to install the skill, or want to drop the behavior into any agent directly? This is all it needs: 
This repository uses Driftless for shared team memory.

BEFORE you edit code, pull what the team already knows about the files
you're touching:
    driftless context get --files "<files you're about to change>"
Read the gotchas, decisions and invariants before writing anything.

AFTER you learn something durable, a gotcha that cost you time, a decision
you made, a constraint that must hold, write it back:
    driftless context update <topic> --gotcha "..." --decision "..."
    # or, if no topic fits the area yet:
    driftless context add <name> --pattern "<glob>" --what "..." --how "..."

That's the whole loop: pull before, persist after. Do it as part of your work.

​
Going deeper

You don’t need any of this to start. Reach for it once you have a few topics and want more out of them:
  • The Driftless loop: the whole model, Knowledge, Operations, and Work, and the loop between them.
  • Topics: the content, anchors, and fields each one carries.
  • Drift & freshness: how a topic flags itself when its code moves underneath it.
  • The CLI: anchoring topics to files, file-matched retrieval, relations.
  • Agent setup in depth: wiring Driftless into specific coding agents.