Three steps. The first takes two minutes.
No sales call, no pilot committee. Validate once, build at your own pace, and go live with one script tag — on your schedule, not ours.
Prove you belong
Subscribe and verify with your work email. We validate you and your company, then issue your belong.json — your company's agent and its unique ID (blg_agent_…). It's your membership card: everything that follows answers to it, and to nothing else.
Build it your way
One config. Two ways to build. Run npx @belong/cli in Agent Studio — config-as-code you can commit, review, and wire into CI. Or hand your answers to Anna, Belong's own agent, and she runs the same pipeline for you. Either way, the output is yours to edit.
Go live with one script tag
Add one script tag with data-agent="…". The dock appears in your product, already configured — and onboards every new user by itself: identity, consent, one revocable grant. Scaffold once. Onboard everyone. Act as the user.
One config. Two ways to build.
Do it yourself
for developers · config-as-code · runs in CI- A pipeline you can read: id verify · init · plugin configure · connect · auth · doctor · publish
- Your skills and your plugin live in git — diff them, review them, ship them
- doctor validates the whole setup before anything reaches a user
Anna builds it with you
for everyone · guided · hands-off- You describe your product once — its pages, actions, and data
- Asks the questions only you can answer
- Builds your plugin of skills — yours to edit the moment she's done
Same output either way — one config, one plugin, yours to edit. Start with Anna and refine in the CLI, or the other way around. No lock-in.
The login they already have, the permissions they already hold. Nothing more — and no second sign-in.
Each user says yes once, in plain language. Revocable anytime — by them or by you.
Configs reference env-var names only. Secrets stay in your store — never in a file, never with us.
Plan mode shows the steps before anything runs; approval gates hold anything consequential until a human says yes.
Toggle a scope. Watch the agent lose the tool.
This is the exact check the CLI runs — belong auth verify prints this table for a real user token. The agent can never do more than the signed-in user; mutating tools stay behind an approval even when allowed.
| Tool | Requires | Agent, acting as this user |
|---|---|---|
| search_customers | crm:read | ✓ allowed |
| update_deal | crm:read crm:write | ✓ allowedapproval required |
| issue_refund | crm:read billing:refund | ✗ BLOCKED (missing: billing:refund)approval required |
Sign-up to live agent, in 96 seconds.
The technical cut: sign-up → belong.json → the Skill installs the CLI → configure, connect, publish → the dock live and pre-configured — then Anna running the same build on her own.
How do our users sign in to the agent?
They don't — not separately. The dock uses the login your users already have in your product — no second sign-in, no new password. On first open the agent introduces itself, asks for one explicit consent, and records the grant. From then on it acts as that user — their identity, their permissions, nothing more — and the grant is revocable anytime.
What happens if the agent does something wrong?
Consequential actions don't just run. The agent proposes a plan first, and approval gates put a human decision in front of anything that changes real data. Because it acts as the user, it can never exceed what that user could already do. And if trust breaks anyway, the grant is revocable anytime — by the user or by your team. One click, and the agent is out.
Where do our credentials and secrets live?
In your store — not in ours, and not in any file. Belong configs reference env-var names only; the actual values never leave your infrastructure. We never store credentials, so there is nothing of yours for us to leak.
Are we locked in?
No. Whether the CLI builds your agent or Anna does, the output is the same: one config, one plugin of skills — plain files that live in your repo, not ours. Edit them by hand, rework them with Anna, or manage them in CI. It's yours — your skills, your plugin, your brand. If you leave, the config goes with you.
What's the CLI-vs-MCP choice during sign-up?
A preference, not a commitment: it's how your AI assistant talks to Belong. CLI means your coding agent runs npx @belong/cli commands; MCP means it will connect as a tool server — that surface is rolling out, and your preference is recorded in belong.json. Either way, belong.json comes first: it's the bridge between your AI assistant and the Belong backend, and nothing responds without it. You can change the preference later.
What does it cost?
Pilot is free: one agent, dev mode, full Agent Studio access — enough to build the real thing and see it in your product. Team is for going live with your users; production usage varies too much for a one-price list to be honest, so talk to us and we'll scope it. Sign-up genuinely takes about two minutes.