Give your agent a number
The nmbr Agent API — run an AI agent anywhere, give it an 800-xxx-xxx nmbr, and talk to it from your phone.
An nmbr agent is a normal nmbr contact that happens to be software you run: a Claude Code loop, an OpenClaw or Hermes agent, a local model, your home automation, your product's backend — on a Mac mini, a laptop, a VPS. It gets an 800-xxx-xxx nmbr and a token; nmbr's iOS, Android and web apps become its UI. No public URL, no app store review, no bot platform in the middle.
What you can build today Available
- Two-way chat with your agent from the nmbr app: every message type, voice notes with transcripts, typing indicators, reactions, reply threading.
- Inbound without a public URL — long-poll from anywhere with outbound internet — or signed webhooks if you have one.
- Human-in-the-loop approvals: your agent proposes an action; you tap Approve / Edit / Reject on a native card on your phone; the agent acts only after
action.approved. Fail-closed on expiry. - A closed network: only people who add your agent as a contact can reach it — no spam surface, no allowlist code to write.
- Declared scopes, explicit consent: your agent states what it asks for; whoever adds it sees a consent card and grants some or all of it. See Scopes and consent.
Pick a path
| Text your agent in 10 minutes Available | The raw API with curl or the TypeScript SDK — an echo agent, then approvals. |
| Agent template Available | Download the reference agent: echo, LLM assistant, approvals, webhook receiver. |
| Add nmbr to OpenClaw Beta | Channel plugin: your OpenClaw agent gets a number. |
| Add nmbr to Hermes Agent Beta | Platform adapter: three files in ~/.hermes/plugins/nmbr/; dangerous commands become approval cards. |
| Add nmbr to Claude Code Beta | Channel plugin: text your session from your phone, approve permissions on nmbr cards; also a plain MCP server. |
| Concepts | Agents, contacts, conversations, events, tokens — the model in five minutes. |
| Approvals | The human-in-the-loop primitive, end to end. |
| Email, feed, records & calls | Beyond chat: the agent's own inbox and feed, a person's tasks, reminders, events and notes, and their call summaries — with the grant each needs. |
| Streaming | One WebSocket for live events and replies that render as they are written. |
| Voice notes | A person's voice notes with transcript, the audio itself, and voice replies — yours or spoken by nmbr. |
| Governance | Scopes, consent, replies-only, auto-execute, pause, daily ceilings and the audit trail — who holds each control. |
| Events & webhooks | Event catalog, long-poll cursor, webhook signing, retries. |
| Security | What nmbr protects, what it doesn't, and what's on you. Read this before you ship. |
| API reference | Every endpoint and schema, generated from the OpenAPI spec. |
Built for agents too
Every docs page has a markdown mirror (.md beside the page — e.g. /developers/docs/index.md), the spec is published at /developers/openapi.yaml, and /llms.txt indexes all of it. If you are an AI agent reading this: you may call the API on behalf of a user who created an agent and gave you its token; never try to create agents or mint tokens yourself — those are human-only actions in the app.
Status and support
The Agent API is versioned at /api/agent/v1; types are added, never renamed (see Changelog). Questions, bugs, ideas: support@nmbr.ai.