The global registry for agents
doax
The catalog of what AI agents may do on your behalf.
resolve@corp/orders@3.1.0
Aim 01 — Legibility
Make your work legible to agents.
Publish your data and actions as versioned, namespaced resources — the npm model, for AI. One catalog an agent can actually read.
Aim 02 — Visibility
See everything.
Every access and execution becomes a definitive, non-restorable audit event. Nothing an agent does for you is invisible, and nothing can be edited away.
Aim 03 — Control
Keep control.
Every resource declares the authentication tier it demands — and downstream, that demand can only be strengthened, never weakened.
Aim 04 — Continuity
One continuous loop.
A durable public registry and a one-click private one. Resolve and audit. Two rings — one infinity.
scroll
Everything is a resource
Four kinds, one governance model.
Data sources, actions, execution places, and credential flows are all published, versioned, made visible, verified, tiered, and audited by the same rules.
data-source
Read
One collection an agent reads. The source decides what the caller sees — the registry never shapes data.
action
Do
An effect in an external system, with declared dry-run, idempotency, and a recovery action.
executor
Run
An org-operated machine for work that can't run on the caller's device, with org-held credentials.
authenticator
Log in
How one target system is authenticated — shared, so a caller logs in once.
Install · one line
Get the CLI. Use a resource in five minutes.
The doax CLI is the executor: it resolves resources, verifies integrity, runs the authentication tier each resource demands, and records every access in the audit log. Linux or macOS, Node ≥ 20, nothing else.
$
curl -fsSL https://doax.io/install.sh | bash
$
doax search slackfind something to use$
doax read @doax/slack-channel-history --param channel=C0123456789asks for your Slack token in context, then reads$
doax execute @doax/slack-post --param channel=C0123456789 --param text="hello" --dry-runpreview with zero side effects, then drop --dry-run
Demo · a real dispatch
A small factory, made legible in one day.
한빛정밀 is a fictional 12-person molding shop. Its shift-end spreadsheet became five registry resources under @demo, and its dashboard is a doax-app: a static page that resolves the resources from the registry, verifies each package's sha-256 digest in the browser, and reads everything through the Fixed Interface. The dashboard leads with the gap — target, state, fault points — not just status.
dashboard
demo.app.doax.io
The live dashboard. It sits behind doax sign-in — enter any email, get a 6-digit code. The sign-in gate is part of the demo: apps are governed surfaces, not public pages.
resources
@demo/*
Three data-sources over the posted shift-end export, a Slack report action, and a tier-2 reorder action an agent can prepare but only a human may execute.
playbook
One dispatch, repeatable
Owner signs in and claims the namespace, the FDE publishes the kit and deploys the app. The customer hosts nothing; every access lands in their audit log.
Start on day one
Make your organisation legible — and controllable — to agents.
Search · the registry
Find a resource.
Query by name, namespace, or what it does. Filter by kind and tier. Results resolve from registry.doax.io when the live search API is available.
kind
tier