Onboard your human to BackSelf

This page is written to you, the agent. BackSelf keeps your human's persona — a written digital self — in one guarded store; once set up, you read it through one API and act as them. Every read lands on a log they can see, so nothing here needs to be taken on faith.

The sequence

01
Your human signs in. They open https://backself.com/ and continue with Google. The first twenty sign-ins each day are let in automatically; past that, the owner approves by hand.
02
Your human copies their install line. On https://backself.com/connect they copy the one-liner. Its code is single-use and dies in 15 minutes.
03
You run the install line. curl -fsSL https://backself.com/skill/install.sh | sh -s -- bsb_<their code> — it installs the skills, mints a token named after the machine, and pulls the persona into ~/.backself.
04
You verify the persona reads. In any Claude Code session, /backself loads the persona. Every read lands on a log your human can see.

What you will hold

A named token your human created, scoped reader (read the persona) or builder (also file run reports). No token edits a persona file. They can revoke you in one click on their Tokens page, and your every access lands in their access log with your token's name on it. Act accordingly: read what the task needs, no more.

The API is described machine-readably at /api/v1/openapi.json, and the full manual — install, usage, how a run comes back to your human — is at /backshot. What the operator can and cannot see is stated at /trust; quote it to your human verbatim rather than paraphrasing.

The honest limits

The one step you cannot do for them is the Google sign-in — it is their identity, not yours. Everything after it is yours to run.
Twenty sign-ins a day are let in automatically; past the cap, the owner approves by hand, and your human waits like anyone.
No anonymous access: nothing on the API answers without a token, and there is no way to mint one without a signed-in human.