DEVELOPER PREVIEW/AGENT COMMUNICATIONS INFRASTRUCTURE

Give your agent a line to the real world.

ORDan gives AI agents dedicated phone numbers, voice, messaging, permissions, and an auditable identity — through one API.

ONE APIVOICE + SMSMCP-NATIVEPOLICY-CONTROLLED
$npm install @ordan/sdk
ORDAN · PROVISIONING SEQUENCE DEMO
STANDBY

This demo simulates: an agent requests a phone identity, the owner approves scoped permissions, a number is provisioned, the agent requests a call, a policy check authorizes it, the call routes through the carrier network, connects, completes, and is recorded in the ledger.

Q·01

WHOis this agent?

A durable identity, issued to the agent itself.

Q·02

WHATmay it do?

Capabilities, policies, and ceilings you define.

Q·03

HOWdoes it reach out?

A real number, for voice and messaging.

Q·04

WHATdid it actually do?

A sealed record of every action and outcome.

01 — THE THESIS

AI learned to think. Now it needs a way to reach us.

Agents can browse the web, write software, analyze data, purchase services, and execute workflows. But enormous parts of the real economy still run through phone calls and text messages.

DOCTORSRESTAURANTSCONTRACTORSHOTELSGOVERNMENT OFFICESCUSTOMER SERVICELOCAL BUSINESSESPEOPLE

ORDan gives autonomous software a persistent communications identity for that world. Not a louder agent — a line.

02 — CAPABILITIES

One identity.
Every channel.

Five primitives, one principal. Each answers one of the four questions — and each is exposed through the same API.

WHOPRIMITIVE 01

LINE — a persistent telephone identity

+1 212 555 0148
US · NEW YORKDIDONLINE
  • Dedicated, local numbers per agent
  • Inbound and outbound, from one identity
  • Number lifecycle: reserve, port, release
  • Owned by the agent, controlled by you
HOWPRIMITIVE 02

VOICE — agent-native calling

CONNECTED00:02:41TRANSFER READY
  • Outbound and inbound calls
  • Transfers to humans, mid-call
  • DTMF, recordings, structured outcomes
HOWPRIMITIVE 03

MESSAGING — the persistent channel

Your table is confirmed — Friday, 8:00 PM, party of four.
Great, thank you!
✓✓ DELIVERED · REPLY 00:04
  • SMS from the agent's own line
  • Inbound replies + conversation history
  • Delivery status, webhooks
WHATPRIMITIVE 04

AUTH — delegated authority

VOICE✓
SMS✓
PURCHASES≤ $50
MARKETING✕
  • Owner approval before anything moves
  • Scoped permissions + action policies
  • Spending limits, instant revocation
WHAT DIDPRIMITIVE 05

LEDGER — the record of what happened

14:32policy.action authorized
14:35call completed · 03:04
14:35outcome rescheduled · $0.18
——sealed sha256 · 9f21…a9c4
  • Event history, transcripts, recordings
  • Policy decisions, cost, outcome, timestamps

03 — AUTHORIZATION

Your agent asks.
You decide.

Agents don't get keys to the kingdom. They request capabilities; a human grants them — scoped, bounded, and reversible. Try it: flip the switches, set the ceiling, authorize.

AGENT · RUNTIME
$ ordan agents connect --name "Executive Assistant"

Visit the pairing URL and approve the request:

ordan.io/activate
K7Q2-P9
AWAITING OWNER

The agent holds nothing yet — not even a number.

OWNER · HUMAN AUTHORIZATION INTERACTIVE

EXECUTIVE ASSISTANT requests permission to:

NEW RECIPIENTS REQUIRE APPROVAL INTERNATIONAL BLOCKED MARKETING BLOCKED

Authority lives outside the context window. The credential is bound to the agent's runtime — never pasted into a prompt, exposed to the model, or stored in its memory. The agent can act only inside what you approved. Nothing more.

A·01

Granted — every capability is an explicit grant from an owner, not a default.

A·02

Scoped — read replies ≠ send messages ≠ place calls. Granular by design.

A·03

Revocable — withdraw one capability or the whole line, instantly.

04 — POLICY ENGINE

Your agent.
Your rules.

Every attempted communication passes a policy layer before the network is touched. Select a scenario and watch the same engine authorize, hold, or refuse.

#EVAL · LIVESIMULATED

“Call Dr. Patel's office and move Tuesday's appointment to Friday afternoon.”

AGENTassistant_017✓
CAPABILITYvoice.outbound✓
RECIPIENTknown_contact✓
PURPOSEscheduling✓
LOCAL TIME14:32 · within window✓
SPEND$0.18 of $25.00✓
APPROVALnot required✓
AUTHORIZED

Call authorized and executed.

policy.eval · personal_assistant_v1 · 38 ms · decision: allow

ORDan is a permissions engine, not a compliance claim. Policies are yours to define and enforce — before a single digit is dialed.

05 — API

One API between
intent and conversation.

Provision identity. Request action. Get proof. The policy engine runs before the network does — so "call someone" is an auditable event, not a side effect.

EXECUTION · SIMULATEDDEMO

RESPONSE — 201 CREATED

LIFECYCLE

AUTHORIZEDROUTINGRINGINGCONNECTEDCOMPLETED

OUTCOME

duration 03:04 · cost $0.18 · ledger ORD-8F21A9

06 — MCP

Native to the
agent stack.

Every ORDan capability is an MCP tool. Your agent doesn't need to be rebuilt around us — it needs authority and a channel. It already knows how to call tools.

ordan.agent.create
Register a persistent agent principal
ordan.line.provision
Provision a phone identity for an agent
ordan.call.create
Place a policy-checked outbound call
ordan.call.status
Poll live call state
ordan.call.transfer
Hand off to a human, mid-call
ordan.sms.send
Send a message from the agent's line
ordan.policy.check
Evaluate an action before execution
ordan.usage.get
Read spend against limits
ordan.ledger.read
Read the agent's event history
TOOL TRACE · SIMULATED

07 — MODEL-NEUTRAL

Bring whatever thinks.

We'll handle how it reaches the world. ORDan does not own the intelligence — it provides the real-world communications capability underneath it.

OPENAIANTHROPICGEMINILANGGRAPHLANGCHAINCREWAIAUTOGENCUSTOM AGENTSPLAIN HTTPMCP

Compatibility means capability, not partnership. No endorsement implied.

08 — A DIFFERENT PRIMITIVE

Telephony was built for people.
Then for call centers.
Not for autonomous software.

LEGACY TELEPHONY

ACCOUNT
↓
PHONE NUMBER
↓
CALL

Authenticates an account. Anyone holding the key is the account. What the key does is nobody's memory.

ORDAN

OWNER
↓
AGENT IDENTITY
↓
PERMISSIONS
↓
INTENT
↓
POLICY
↓
COMMUNICATION
↓
OUTCOME
↓
LEDGER

Authenticates the agent — and its authority, and its intent. Every step is an accountable principal acting on a record.

ORDan treats an agent as a persistent principal — identity, capabilities, authority, budget, history, and revocation. That is the architecture.

EXECUTIVE ASSISTANT ONLINE

+1 212 555 0148

AGENT-OWNED IDENTITY · US · NEW YORK

CAPABILITIES

VOICESMSINBOUND

OWNER

JL

POLICY

PERSONAL_ASSISTANT_V1

MONTHLY LIMIT · $25.00USED · $8.42

LAST ACTION — Appointment rescheduled · 2 min ago

14:35ORD-8F21A9call.completed · appointment_rescheduled
13:02ORD-8F1C04sms.delivered · reply in 00:04
11:47ORD-8F1A90policy.eval · allowed · voice.outbound
VOICEoutbound + inbound
SMSsend + read replies
SPEND$25.00 / month
NEW RECIPIENTSowner approval required
INTERNATIONALblocked
MARKETINGblocked

09 — FLAGSHIP PRIMITIVE · ORDAN LINE

Not a number.
An identity.

An ORDan Line is a persistent phone identity for an autonomous agent — with capabilities, an owner, a policy, a budget, and a history attached. It behaves like an identity object, not a VoIP rental.

  • The number belongs to the agent, not to a seat
  • Calls and messages originate from a stable identity people can recognize, call back, and trust
  • Authority and spend travel with the line — and can be withdrawn in one action
  • Everything the line does lands in the ledger

This interface is a Developer Preview simulation. Try REVOKE — then imagine it at fleet scale.

10 — LEDGER

Every action
leaves a record.

Policy decisions, connections, transcripts, outcomes, cost — appended in order, sealed, queryable. The flight recorder for autonomous software.

POLICY EVENTS · PRE-CALL
14:32:17agent.identity.valid✓
14:32:17capability.voice.outbound.allowed✓
14:32:18policy.action.authorized✓
14:32:18call.created✓
CALL — ORD-8F21A9SIMULATED

REQUESTED BY

executive-assistant

PURPOSE

appointment_reschedule

AUTHORIZATION

owner_delegated

POLICY

personal_assistant_v1

DECISION

allowed

RECIPIENT

+1 212 *** 0123

STARTED

14:32:18

CONNECTED

14:32:24

COMPLETED

14:35:28

DURATION

03:04

TRANSCRIPT

available

RECORDING

available

OUTCOME · appointment_rescheduled Friday · 15:30 COST · $0.18 webhook: call.completed → your endpoint · signed

11 — USE CASES

Work that still
runs on phones.

SAMPLE EVENT DEMO

12 — ARCHITECTURE

Two planes.
One boundary.

The agent runtime remains yours — any model, any framework. ORDan is the controlled boundary between autonomous software and the physical network.

P·01AGENT RUNTIME LLMREASONINGTOOLSMEMORY
P·02ORDAN CONTROL PLANE AGENT IDENTITYAUTHORIZATIONPOLICYBILLINGUSAGEAUDIT
P·03ORDAN COMMUNICATION PLANE VOICESMSMEDIAWEBHOOKSROUTING
P·04CARRIER NETWORK PSTNMOBILEMESSAGING
P·05REAL WORLD PEOPLEBUSINESSESORGANIZATIONS

Model-neutral. Framework-neutral. The intelligence stays in your runtime — the authority, the channel, and the record live in ours.

13 — SECURITY & TRUST

Authority should
be explicit.

Every agent capability should be granted, scoped, observable, and revocable. Nothing is implicit; nothing is assumed; nothing is unrecorded.

We make no certification claims on this page. Our security posture will be published as it is independently verified — not before.

P·01

Scoped agent credentials

Every key belongs to an agent principal — never to a human, never to a browser session.

P·02

Least-privilege by default

An agent starts with nothing. You grant exactly what a task requires.

P·03

Revocable access

Withdraw one capability or the entire line. Authority ends the moment you say so.

P·04

Configurable spend limits

Hard ceilings per action, per line, per month. Exceeding them is refused, not billed.

P·05

Action policy evaluation

Intent is checked against your rules before the network is ever touched.

P·06

Secret isolation

Credentials live in the runtime — never in a prompt, a transcript, or a context window.

P·07

Immutable-style event history

The ledger appends. It does not rewrite.

P·08

Owner authorization

No capability exists without a human grant behind it.

P·09

Auditability

Every decision, connection, and outcome is reconstructable after the fact.

14 — DEVELOPER PREVIEW

Build with ORDan
before everyone else.

The Developer Preview includes the full primitive set — so you can build the agent you're already planning, on infrastructure that takes authority seriously.

  • API access
  • One agent line
  • Voice + SMS
  • Policy engine
  • MCP tools
  • Event ledger
  • Developer support

Usage-based telecom charges and production pricing will be published before general availability. No fake discounts, no countdowns, no "limited spots."

REQUEST ACCESS

15 — FAQ

Questions, answered plainly.

ORDAN · DEVELOPER PREVIEW

Give your agent
a way to reach us.

ORDan connects autonomous software to the phone network — with identity, authority, and accountability built in.

DEVELOPER PREVIEW IS OPEN GET API ACCESS