Reference¶
Anychat has two public API surfaces. They run on different hosts, carry different version numbers, and authenticate differently, so check which one your task belongs to before you write any code.
| Surface | Host and version | What it does | Authentication |
|---|---|---|---|
| Agent API | https://api.anychat.ai/v1 |
The management plane: create and configure agents, people, channels, outreach, and tests. | An org-scoped API key (sk-any-…) or a console user's session token, carrying scopes. |
| Messaging API | https://gateway.anychat.ai/v4 |
The runtime plane: send and receive messages for one agent. | A per-agent access token. The token is the agent, and there are no scopes or roles. |
Use the Agent API to build and operate agents, including reading their conversation history. Use the Messaging API when your own code produces the messages, rather than an Anychat agent.
- API quickstart: mint a key, make your first authenticated call, and receive your first webhook.
- Limits and quotas: rate limits, sending limits, channel limits, and retention.
- Glossary: one-line definitions of every term used in this documentation, each linking to the concept page that defines it.
- Agent API: create, configure, and inspect agents: identity, playbooks, capabilities, templates, people, channels, outreach, tests, export and import. Everything the console does, scriptable.
- Messaging API (AMF/1): the message format and the send and receive surface. One rich-message envelope of intents and parts, degrading across every channel.
- AML: Anychat Markup Language. Teach any LLM agent to produce rich messaging, including cards, carousels, and suggested replies, in one page.
To drive Anychat from your own AI agent instead, see Drive Anychat from Claude Code, Claude Desktop, or Cursor.