PXB CLI

Give your agents a business to run

Everything a company runs on at Pixelbase sits behind one API. The PXB CLI puts it a single command away — so an AI agent can read the order book, answer a customer, move a ticket and ship new business functionality without a browser in the loop.

Node 18.17+. One executable, no dependencies, no package manager.

A morning of work. In eight commands.

An agent authenticates once, pins the company and location it's working in, and from there every call is one line. The scoping id every endpoint needs is filled in automatically — it never has to thread a location through your prompt.

Point it at a second company and the same script runs again, isolated: separate credentials, separate context, separate token.

agent@pixelbase
$ pxb company use acme
Now using Acme Corp (profile "acme")
$ pxb location use headquarters
Now using location Headquarters
$ pxb order list --json -q status=PAYMENT_PENDING
using locationId=Headquarters { "content": { "orders": [ … ] } }
$ pxb support-case list --json -q status=OPEN
{ "content": { "supportCases": [ … ] } }
$ pxb post support-cases/case_123/messages -d '{"text":"On it."}'
200 POST support-cases/case_123/messages
least privilege
$ pxb whoami --json | jq -r '.content.token.scopes[]'
orders:read support:cases:read support:cases:write
$ pxb product list
403 GET products
Access token is missing required scope(s): products:read

Autonomy you can bound.

An agent is only as dangerous as the scopes you hand it. A triage agent that reads orders and replies to support cases gets exactly three scopes — it cannot reprice your catalogue or delete a customer, no matter what ends up in its context.

Every agent gets its own client, so you can revoke one without disturbing the rest. Secrets live in the operating system keychain and are passed over stdin, never in a command line where another process could read them.

How authentication works

Install it in one line.

Then hand an agent a scoped client and let it get to work.