You can set up your Inrō AI agent from outside the dashboard: through Claude or ChatGPT (via the MCP Server) or through the public REST API. That means you can manage your agent's knowledge base, intent-triggered actions, contact property tracking, and core settings from your own code or a simple conversation.
For what the agent does and how to tune its behaviour, see the AI Agent articles linked at the end. This article covers the programmatic surface.
What you can manage
Four parts of your AI agent are available here:
Agent settings: the trigger mode (when the agent replies), folder rules, role, message length, and reply-handling scenarios.
Knowledge base: text snippets and URLs the agent uses to answer questions.
Intent-triggered actions: actions that fire when the agent detects an intent. For example, when someone says they want to buy, add them to a folder, run a scenario, send a conversion link, or send a survey.
Contact property tracking: which properties the agent detects and saves during a chat (the built-in
emailandphonefields, or your custom properties).
⚠️ The agent's AI model and reasoning level show up in the config returned by GET /api/v1/agent, but they aren't fields you can change with PATCH /api/v1/agent. Set those in the dashboard (see AI Agent Best Practices).
Via Claude or ChatGPT
If you've already connected Inrō to Claude or ChatGPT via the MCP Server, these tools are available automatically, with no extra setup. Ask in plain language:
"Add a knowledge item to my agent: our delivery takes 3 to 5 business days"
"Create an action: when someone wants to buy, run my 'Purchase Flow' scenario"
"Track the email and phone number properties"
"Switch my agent to reply to all messages"
"Show me all my current agent actions"
Not connected yet? See Connect Your AI Agent to Inrō with the MCP Server.
Via the public API
The same capabilities live under /api/v1/agent. This is the path to use when you want to build an integration, sync agent setup from another source, or configure new accounts automatically.
Agent settings:
GET /api/v1/agentto read,PATCH /api/v1/agentto update.Knowledge items: full create, read, update, and delete at
/api/v1/agent/items.Actions: full create, read, update, and delete at
/api/v1/agent/actions.Contact property tracking: list, add, and remove at
/api/v1/agent/contact_properties.
When you add an action, set its trigger (the intent to match), a target_type (one of ConversionLink, Folder, Scenario, or Survey), and the target_id. See The Inrō Private API: Getting Started for auth and the full parameter reference.
🐾 Netsuke's Tips
Knowledge items accept URLs, so you can point the agent at a page on your site and it will use that content to answer questions. Over MCP, you can also set
crawl_source_urlto pull in the linked page's sub-pages, not only the one URL.Intent-triggered actions stack: several actions can fire for the same intent, each doing something different (for example, add to a folder and send a survey).
When adding knowledge, be specific. Paste the actual text you want the agent to know rather than describing it. The more precise the item, the more reliably the agent uses it.