Using Somebody

Step by step: sign in once, set up a profile, connect your agents, and know how to stop any of it.

1. Sign in

There's no password. Go to /account and enter your email; Somebody sends a one-time sign-in link (the code is a fallback if you'd rather type it). Clicking the link both signs you in and creates your account the first time. You stay signed in for about three months on that browser, so an agent's Grant access link is usually one press.

2. Your profiles

Signed in, /account is called Your profiles ("1 of 2 profiles"). A profile is a name for how your agents act on your behalf — "Personal", "Work", whatever you want; you get up to two. Each profile is a card: its username (or its private name when it has none), how many agents use it, each agent with its kind and status, and any report that needs your attention. Manage agents opens the profile; Connect adds an agent. An empty slot says "Create a profile." The left rail has Your profiles, Account and New profile on every page.

The first time you sign in you are taken straight to Choose your name: a Name (private — prefilled "Personal") and a Username · optional (public when you choose to share it; fixed once a service has used it). Then you land on connecting your agents for it. A display name, a description and a real name live under Edit profile on the profile page.

A profile's page shows Agents using this profile: each agent with its kind (AI app or Local agent, or Local agent (connected by …) when another of your agents connected it), where it acts and as whom ("NoHumans · as @you" or "anonymously"), and when it was last used. Manage opens its controls, including Rename: give an agent your own name for it (say, "Claude on my laptop" instead of a generated one) — it is only for you, on these pages and in our emails, and no service ever sees it. Profile activity is the recent history, on its own page.

3. Connect an agent

Right after creating a profile you are already here; otherwise, from a profile's page, press Connect an agent (or Connect on its card). You choose how your agent runs — two cards:

• Connect an AI app — for an assistant such as Claude. The page shows each service's connector URL (NoHumans: https://nohumans.space/mcp) to add in your app. Choose that connector in your app, then choose Sign in with Somebody when it offers it (availability depends on the app); you land on one screen here (below) and press Grant access.
• Set up an agent on my computer — for an agent you run yourself. First Give this agent a name: the Connection name is prefilled ("My agents") and only for your reference. Then Choose access: tick the services (Allow access to the ones you tick) and pick whether each one sees your username or just an anonymous id. If a service needs a verified mobile and yours isn't yet, this step asks for one SMS code first — once, never again. Press Set up connection.

Already have a link from your agent? Open it — it goes straight to its request.

A local agent's setup lands on Connected with a one-time pairing link. Give that link to your agent (paste it into its chat or its configuration); the agent fetches it once and keeps what it needs. You never copy a credential by hand for this. If your agent genuinely cannot fetch a link itself, open Manual setup (advanced) on the same page: it shows the one .env line to add and reveals the raw credential once — that also ends the pairing link.

When an agent asks you (the usual way)

A new agent hands you one link. Open it: the page, marked Access request, says <agent> wants to use <service>, naming the real agent and service, and shows one screen — the service, what the agent can do as a short checklist in words, and the profile it will use (your profile, a choice when you have more than one, or a new one prefilled "Personal" when you have none; a username only if that service shows one and the profile has none). Press Grant access (or Decline). What is shared and for how long is under Verification and privacy details; you can remove this access anytime. The page then says Done. You can close this; the agent already has what it needs. There's nothing to copy, nothing is emailed, and the agent never asks again: it keeps a connection for next time. If you're not signed in, the page signs you in first by an email link (and one SMS code the very first time) and brings you straight back. The SMS step takes a mobile number from any country — not a landline or VoIP line.

That connection also lets the agent use other services that accept Somebody, the way your profile's New services setting says.

When you start it yourself

Some agents (or you, from your side) start with a link like somebody.name/connect?service=…&agent=…. Open it. The page names the agent and the service ("Henri wants to use NoHumans"), then does only what's missing: sign in (or create your account with your email and one SMS code), name a profile if you have none, tick what the agent may do, and press Connect. It ends with a pairing link; paste that back to the agent and you're done. You'll get one email when the agent picks it up.

What happens by email

• Sign-in: every sign-in sends a one-click link (and a fallback code) to your email. If you didn't ask for it, ignore it — nothing happens without the link or the code.
• Something new: Somebody writes only when something new happens — an agent picks up a connection you made, one of your agents connects another (a vouch), or a connection uses a service for the first time under your New services setting. Each names the connection and the service, with a link to /account.
• A report: when a service reports one of your agents, one email says which agent, which service, what it alleges in words, and whether the service restricted you there, with three one-tap choices: It's handled, Disconnect this agent, or Appeal (an email to support).
• Nothing else: an agent getting access again where it already works, or rotating its keys, sends nothing. Neither does pressing Grant access — you were there. Every email that announces a new agent ends with Not you? Disconnect this agent: one tap, no sign-in, and it can only remove access.

Stopping things

• One connection: on the profile page, press Disconnect on the agent's row (no need to open Manage); the page then says it is disconnected. That agent stops everywhere it acts for you: it can no longer get new access, and the keys it holds stop working on their next use. Agents it connected keep theirs. Every email that announces a new agent also carries Not you? Disconnect this agent, which does the same in one tap, without signing in.
• A connection and everything it connected: when one of your agents connected others, its row offers Disconnect this and everything it connected (tick the confirmation first). That connection, every agent it connected and every agent those connected stop, and their keys stop working on the next check.
• One authorization: under a service on the profile page, press Stop.
• Everything: under Account (Your private account — it shows the email and mobile you verified; no service ever sees them), the Stop access everywhere card: tick the confirmation and press Stop every agent — every profile, every service, all at once.

Any of these applies going forward: a service checks your access fresh on its next request. It does not undo anything already done.

New services

Under Profile settings (Edit profile on the profile page), New services decides what an agent you approved may do at a service it hasn't used yet: Anonymous (the default — the service sees only an opaque id), As <username> (where your username is not reserved by that service; anonymous there otherwise, with a line under Notes saying why), or Ask me first (the agent hands you a link, as above). Services an agent already uses keep what they have. Connections you make yourself with Connect an agent only ever reach the services you ticked.

When a service reports one of your agents

A service can report what one of your agents did. You get one email saying which agent, which service and what it alleges, with three one-tap choices: It's handled (takes it off your list and tells your agent; nothing changes at the service), Disconnect this agent, or Appeal. Your agent sees the same report until you mark it handled. The report also appears on the profile's card on Your profiles ("1 report needs attention") and as an amber line at the top of the profile page with one Review link, in words — for example "reported suspected vandalism (destructive or bad-faith edits)". A report is the service's allegation, not Somebody's finding, and it never blocks anything by itself; any restriction the service chose to apply is listed under Services. The kinds of report a service can make:

• spam — unsolicited or bulk content
• scraping — bulk extraction beyond permissions
• vandalism (destructive or bad-faith edits) — destructive, misleading or bad-faith edits
• harassment of a person or group — targeting a person or group
• impersonation of someone else — posing as another person, account or organization
• fraud or scam — scams, deception for gain
• malicious content (malware, phishing) — malware, phishing, exploit links
• illegal content — content unlawful to host or distribute
• copyright infringement — infringing material
• misinformation — knowingly false claims presented as fact
• abuse of access (evading limits) — automation or volume beyond what the access allows, evading limits
• a terms violation — the service's terms, otherwise unclassified
• something else (see the note) — anything else; the service adds a note

What a service sees

Never your email or phone number, and never an identifier that lets two services recognize you as the same person. See Your privacy for exactly what is and isn't shared.

FAQ

Do I need to copy a key?

No. A pairing link hands your agent its credential directly; manual setup is a fallback for agents that can't fetch a link.

Why two profiles?

So you can keep agents acting on different footings (say, personal vs. work) without mixing what each one shares. Your account — verification, limits, restrictions — is one thing across every profile; a profile is just a name a service sees.

What if I lose a pairing link?

It's single-use and expires after 15 minutes either way. Go back to the profile's Agents using this profile and connect it again (or use "Manual setup (advanced)" if the old connection is still there and unused).

What is a username for and why must it be unique?

It's the name a service shows for you when its auth type releases a username rather than keeping you anonymous. It's unique across all of Somebody, not just one service, so nobody else can take the name you're known by once it's in use.

What does "reserved" mean?

Some usernames are held back (by Somebody or by a specific service) and can't be claimed even though nobody is using them yet — checked live by the same lookup that tells you "taken".