totebag

Portable, vendor-neutral project knowledge any AI Agent can restore over a CLI.

Built by StratoNext.

Your context shouldn't live inside one vendor's chat history. totebag keeps it in a portable store you address with a single root URL — any agent that can run a shell command restores a project with one call and writes back what it learns. No server, no API keys, no lock-in.

Install

uv tool install totebag

Quickstart

Create a project with a description, add work to it, then dump its context:

totebag init
PID=$(totebag project create --name "Payments API" --description "Billing service")
totebag project use $PID

totebag task add --title "Wire webhooks" --description "handle Stripe retries"
totebag tool add --name reconcile --description "ledger-reconciliation skill" --type skill

totebag project context
totebag workspace get --brief --recursive

That's the whole loop: describe a project, add tasks/skills to it, and project context is the restore blob an agent reads. Point your agent at the bundled guide so it knows the rest of the commands:

totebag --skill   

What a project can hold

Plus the project's own description and instructions. Everything you store carries a description — so project context stays high-signal.

Store location

totebag defaults to ~/.totebag. Point it anywhere with TOTEBAG_ROOT (or --root) — the URL scheme picks the backend:

export TOTEBAG_ROOT=s3://my-bucket/totebag

S3 and GCS need an extra: totebag[s3] / totebag[gcs].

Contributing

Prerequisites and the local development workflow are in CONTRIBUTING.md.

License

MIT

Credits

Icon: Totebag icons created by Khoirul Huda - Flaticon