Four steps: install, authenticate, run, inspect. Five minutes if your wifi cooperates.
[ 01 ]Install
Install
One package gives you the CLI and the local toolchain. Node 20+ is the only prerequisite.
$ npm install -g @mynd/cli
✦ @mynd/cli@0.4.2 — 14 packages, 3.1s
$ mynd --version
0.4.2
[ 02 ]Authenticate
Authenticate
Login opens a browser handshake and stores a scoped key locally. Keys are project-scoped — one key cannot read another project's context.
$ mynd auth login
→ opening browser for handshake…
✓ authenticated as you@example.com
✦ key stored — scope: project/default
[ 03 ]First run
First run
A run is one unit of work: a prompt, optional context, a step ceiling. This one attaches your calendar and asks the runtime to prepare your week.
$ mynd runs create --model y0-fast --context calendar \
--max-steps 4 "Prepare my Thursday."
✓ run created — run_4af2c19e
✓ result ready — 612ms, 2/4 steps used
[ 04 ]Inspect the trace
Inspect the trace
Every run produces a replayable trace: the plan, each context fetch with the scope it used, each step, the result. Debugging is reading, not guessing.
$ mynd runs trace run_4af2c19e
✓ plan composed — 2 steps, 96ms
✓ context fetched — calendar (3 items, scope calendar:read)
✓ step 1/2 — deadlines collected
✓ step 2/2 — brief drafted
friction: removed_
[ 05 ]Where next
You have a run and a trace. Now learn what the runtime is actually doing.