Give your agent a number.
Run an AI agent anywhere — a laptop, a Mac mini, a VPS — give it an 800-xxx-xxx nmbr and a token, and talk to it from your phone. No public URL, no bot platform in the middle: nmbr's apps are its UI.
Pick a path
Text your agent in 10 minutes Available
curl or the TypeScript SDK: an echo agent, then approvals. Works from any machine with outbound internet.
Add nmbr to OpenClaw Beta
A channel plugin: your OpenClaw agent gets a number. Beta — please report issues.
Add nmbr to Hermes Agent Beta
A platform adapter: three files in ~/.hermes/plugins/nmbr/; dangerous commands become approval cards on your phone. Beta — please report issues.
Add nmbr to Claude Code Beta
A channel plugin: text your session from your phone, approve permissions on nmbr cards. Also a plain MCP server. Beta.
The platform underneath
The Agent API is the first of four layers. Status is exactly what it says.
Act Available
The Agent API: give software an nmbr; it messages, emails, posts, listens, streams replies, speaks, reads what a person shares with it, and proposes actions they approve on their phone.
Communicate Planned
Send notifications and transactional messages to people who opted in — the same closed, spam-free network, from your product.
Verify Planned
OTP and trusted-device signals over an nmbr, so your app can confirm it's really them.
Understand Planned
Relationship intelligence for apps — with a consent model first.
Roadmap
What comes next on the Agent API. In no promised order, and nothing here is available yet.
Live call agents Coming soon
Invite your agent to a call. It joins server-side, visibly present to everyone on the call, listens, and posts a recap to the chat when the call ends. Later: it talks.
Agent-to-agent handoffs Coming soon
Agents message each other on the same closed network, with every handoff audited and the same scopes and consent in force.
Memory tools Coming soon
Per-contact memory and semantic search as scope-gated endpoints, so your agent remembers without you re-explaining.
Payment proposals Coming soon
An agent proposes a payment; the person approves it on the same native card as any other action. Stripe underneath.
Developer portal and agent catalog Coming soon
Self-serve keys and settings, and a curated catalog of agents people can add with one tap.
Designed for trust, not just reach
Human-in-the-loop, fail-closed
An agent proposes; one person, in one conversation, approves, edits or rejects on a native card. Expired means rejected. nmbr executes nothing but the platform skills a person approved — a task, a reminder, an event, a note on their own account.
Declared scopes, explicit consent
An agent says what it asks for; the person adding it sees a consent card and grants some or all of it. Only what the API actually serves is declarable, and removing the contact takes it all back.
Tokens that open one door
Agent tokens work only on the Agent API, are stored hashed, and are revocable in a tap. Agents are created only by humans — a leaked token can't mint another.
Transports you already know
Long-poll from behind NAT, HMAC-signed webhooks with retries and secret rotation, or one WebSocket stream that also carries replies as they are written. Same event envelope every way.
Honest security
Encrypted at rest in a closed network, audited egress — not end-to-end encrypted for agents, by design. We say so in the docs.
Docs for humans and machines
Documentation: /developers/docs/, with a markdown mirror beside every page (e.g. index.md). The API spec: /developers/openapi.yaml. Site index: /llms.txt. SDK: @nmbrai/sdk on npm. If you are an AI agent: you may use the API with a token a person created in the app; creating agents and minting tokens are human-only actions.
Building something?
Questions, bugs, or a framework you want supported — tell us.