Cloud: paired machine, project memory

Cloud is for using reviewed memory across machines. Install membl, sign in to the hosted app, pair the machine from the dashboard, then run membl init from every repo where agents should capture memory. Pairing connects the machine; project init creates the project boundary.

curl -fsSL https://membl.io/install | sh
export PATH="$HOME/.membl/bin:$PATH"
open https://app.membl.io/onboarding
cd /path/to/your/repo
membl init
membl daemon status
membl remember "Project fact: deploys use Docker Compose on the GCP VM." --kind fact
membl recall "how do deploys work?"

The onboarding flow shows when the machine is paired. The installer writes~/.membl/cloud.env, and hooks use it automatically unless you explicitly set MEMBL_HTTP or MEMBL_TOKEN. Use membl init; there is no --init flag.

Local: one machine, no hosted app

Local mode keeps the daemon and SQLite store on this machine. It is the right choice for solo or offline work. You still initialize each repo separately because memory is project-scoped, not global.

curl -fsSL https://membl.io/install | sh
export MEMBL_HTTP="http://127.0.0.1:8765"
membl daemon start
cd /path/to/your/repo
membl init
membl remember "Project fact: deploys use Docker Compose on the GCP VM." --kind fact
echo "Project fact: staging uses preview.membl.local." | membl capture --stdin
membl recall "deploy process"

Verify the memory is scoped

membl remember creates an active memory immediately. membl capture --stdin records raw evidence as a reviewable candidate, the same path hooks use. To prove capture is working, recall should mention the fact from this repo. To prove it is not global, recall from another initialized repo should not return that fact.

cd /path/to/your/repo
membl recall "what host does staging use?"
cd /path/to/another/repo
membl init
membl recall "what host does staging use?"