# AGNT

> AGNT is infrastructure for AI personal assistants: assistants with their own identity (name, phone number, inbox), messaging over iMessage, SMS and WhatsApp, phone calls and live voice, a per-user cloud computer, long-term memory, durable long-running tasks, multi-tenant organizations and teams, and per-action human approval. Prime is the agent harness that runs every chat and task. Model-agnostic. REST API at https://api.agnt.ai and CLI `agnt` (`@agnt-sdk/studio`). Auth: `Authorization: Bearer ak_live_...`, plus `X-User-Email` to act as one user. Start with the quickstart. Private beta.

Full text of every guide plus an endpoint index: https://agnt.ai/docs.md

## Guides

- [Overview](https://agnt.ai/reference/guides/overview): What AGNT is, how the platform is layered, the core terms, and the facts that apply to every API call.
- [Quickstart](https://agnt.ai/reference/guides/quickstart): From an API key to a working task: provision a user, give them an assistant, run a task, and read what it did.
- [Authentication](https://agnt.ai/reference/guides/authentication): API keys and their levels, acting as a user with X-User-Email, certificate JWTs for the Node SDK, what each credential can do, errors and rate limits.
- [Prime: the harness](https://agnt.ai/reference/guides/prime): What Prime, AGNT's agent harness, does on every chat turn and task, how chats and tasks reach it, how it uses each user's memory, and how model tiers are chosen.
- [Assistants, messaging and calls](https://agnt.ai/reference/guides/assistants-and-messaging): Create assistants, tie phone numbers and email addresses to users, and reach Prime over iMessage, SMS, WhatsApp, email, Slack, Teams, phone calls and live voice.
- [Computer use](https://agnt.ai/reference/guides/computer-use): The per-user cloud workspace Prime uses to work on websites and run code, how a user takes over the browser through a handoff, how to watch a task's browser, and the controls and usage endpoints.
- [Tasks, autonomy and approvals](https://agnt.ai/reference/guides/tasks-and-approvals): Run durable tasks that wait on people and resume on their own, decide per action type what needs a person's approval, and review, approve, edit and undo staged actions.
- [Users, organizations and teams](https://agnt.ai/reference/guides/organizations-and-teams): Provision and deprovision users, group them into organizations and teams, set up verified domains, SAML SSO, SCIM and audit export, and set policy from your account down to each team.
- [Security](https://agnt.ai/reference/guides/security): How AGNT isolates users, protects credentials, keeps people in control of consequential actions and handles untrusted content, and what you are responsible for.
- [CLI and coding agents](https://agnt.ai/reference/guides/cli): The agnt CLI, and how to build on AGNT from Claude Code, Codex or Cursor.

## API reference

- [API reference index](https://agnt.ai/reference/api)

### Authentication

- [API keys](https://agnt.ai/reference/api/api-keys): Create, list and revoke API keys, and check which account, organization or user a key acts for. A new key is shown once.

### Users and assistants

- [Users](https://agnt.ai/reference/api/users): The end users of your app. You provision users by email, and everything a user owns (tasks, chats, memories, contacts) is isolated to that user. Errors common to every user-scoped call are listed in the Authentication guide.
- [User setup](https://agnt.ai/reference/api/user-setup): A user's profile and setup. Calls made as a user return `428 onboarding_required` until you set a timezone and working hours and mark setup complete.
- [Assistants](https://agnt.ai/reference/api/assistants): An assistant's identity: name, personality, writing style, email address and phone lines. A user can have several assistants, one of them primary.
- [Identifiers](https://agnt.ai/reference/api/identifiers): The phone numbers and email addresses that belong to a user, so inbound messages and calls reach the right person.

### Work

- [Tasks](https://agnt.ai/reference/api/tasks): Durable units of work. A task can run for days, wait on other people and resume when they reply.
- [Chats](https://agnt.ai/reference/api/chats): Conversations between a user and an assistant. Add a message, then process the chat to get the assistant's reply.
- [Approvals](https://agnt.ai/reference/api/approvals): Actions the assistant staged for a person to review before they go out. The API calls a batch of staged actions a cart. Show the approval to the user, then approve it with its `contentHash` so exactly what they saw is sent.
- [Autonomy](https://agnt.ai/reference/api/autonomy): Per-user settings for how freely the assistant acts on each kind of outbound action (`auto`, `time-boxed` or `manual`), plus per-recipient exceptions. A team, organization or the account can lock a setting so lower levels can only tighten it.
- [Skills](https://agnt.ai/reference/api/skills): Packaged capabilities your assistants use: instructions, connected tool servers and recurring workflows.

### Context

- [Memories](https://agnt.ai/reference/api/memories): Facts and preferences the assistant saves and recalls for each user.
- [Contacts](https://agnt.ai/reference/api/contacts): The people a user works with, across email, phone and chat, with relationships and external ids.
- [Files](https://agnt.ai/reference/api/files): Each user's file storage. Upload a file in three steps (create an upload, PUT the bytes, confirm), then attach it to a task or message by id.

### Workspace

- [Workspace](https://agnt.ai/reference/api/workspace): Each user's cloud computer, with a browser, desktop and shell, that the assistant uses for sites and apps without an API. When the assistant needs the user, for example to sign in, it opens a handoff, and the user takes over the live browser and hands it back.

### Organizations

- [Organizations](https://agnt.ai/reference/api/organizations): Your customers' companies. Each organization has members, verified domains, SSO, audit events and policy that its teams inherit.
- [Teams](https://agnt.ai/reference/api/teams): Groups of users in your account, inside an organization or on their own. Team admins adjust settings for their team wherever a higher level has not locked them.

### Voice

- [Voice](https://agnt.ai/reference/api/voice): Live voice sessions, call history, voice engines and phone line requests.