> For the complete documentation index, see [llms.txt](https://docs.360dialog.com/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.360dialog.com/docs/mba/meta-business-agent/3.-configure-your-agent.md). # 3. Configure your Agent Configuration is where your agent becomes *yours*. You set three things: what it **knows**, how it **behaves**, and what it can **do**. You only need the parts your use case calls for — start small and add more as you iterate. {% hint style="info" %} All configuration endpoints live under `agent_config` (and `agent_connectors` for actions), called through 360dialog with your `D360-API-KEY`. For exact routes and every field, see the **Meta Business Agent API reference**. One quirk to know up front: the list endpoints return **bare JSON arrays**, not a `{ "data": [ ... ] }` wrapper — and `settings` comes back as a single-element array. Endpoints can also fail intermittently with a 4XX or 500 for no clear reason while Meta debugs an ongoing instability — retry after a few hours before assuming your payload is wrong. See Known limitations. {% endhint %} *** ## Knowledge ### Business info Core facts about your business, stored as a **fixed set of six fields**. This is where the agent gets hours, policies, and contact details. {% hint style="warning" %} **Business info is a closed object — you can't add custom keys.** If a fact doesn't fit one of the six fields, put it in an [FAQ](#faqs) or a [file](#files) instead. {% endhint %} ```bash curl -s -X POST https://waba-v2.360dialog.io/agent_config/business_info \ -H "D360-API-KEY: " \ -H "Content-Type: application/json" \ -d '{ "business_description": "Independent bicycle shop selling city, road, and mountain bikes, parts, and repairs.", "return_policy": "Unused items can be returned within 7 days for a full refund.", "delivery_and_shipping": "Free local delivery on orders over R$300; nationwide via Correios, 3–10 business days.", "payment_method": "Pix, credit and debit cards, and cash in store.", "purchase_info": "Buy in store or on WhatsApp. Reserve a bike with a 20% deposit." }' ``` [Create or replace business information](/docs/mba/api-ref/knowledge-business-info/create-or-replace-business-information.md#put-agent_config-business_info) ### FAQs Question-and-answer pairs the agent draws on. Each FAQ needs a `question` and an `answer`. {% hint style="info" %} **Keep FAQs focused, and don't overload.** One specific topic per entry, phrased the way a customer would actually ask ("What is your return policy?" beats "Returns, exchanges, refunds"). Keep answers self-contained — the agent retrieves each entry independently, so don't reference other FAQs. Adding too many (beyond a few hundred) makes it harder for the agent to pick the right one. Prioritize your most common questions. {% endhint %} ```bash curl -s -X POST https://waba-v2.360dialog.io/agent_config/faq \ -H "D360-API-KEY: " \ -H "Content-Type: application/json" \ -d '{ "question": "What are your opening hours?", "answer": "Monday to Friday 9:00–19:00, Saturday 9:00–14:00. Closed Sundays and public holidays." }' ``` [Create an FAQ](/docs/mba/api-ref/knowledge-faq/create-an-faq.md#post-agent_config-faq) ### Files Upload documents (PDFs, images, etc.) as knowledge sources. Uploads are `multipart/form-data`, up to **100 MB** per file. {% hint style="warning" %} **File handling has real gaps today — read before relying on it.** * **Uploads fail when a file name is included.** Upload without setting a filename in the multipart part as a workaround. * **CSV files, and tables inside PDFs, error out** when the agent tries to read them. Avoid tabular data in files for now; put critical structured facts into FAQs or business info instead. * **The agent can't reliably send files or images back out** to customers. Don't design a flow that depends on the agent sharing a file. We've raised these with Meta. Until they're fixed, keep files to plain-text-style PDFs and treat outbound file sharing as unavailable. {% endhint %} ```bash curl -s -X POST https://waba-v2.360dialog.io/agent_config/files \ -H "D360-API-KEY: " \ -F "file=@./return-policy.pdf" ``` [Upload a knowledge file](/docs/mba/api-ref/knowledge-files/upload-a-knowledge-file.md) **Supported types:** `.pdf`, `.doc`, `.docx`, `.png`, `.jpg`, `.jpeg` , `.csv` and `.xlsx` **Maximum size:** 100,000,000 bytes (100 MB). ### Websites Give the agent website URLs to crawl and reference. Add a `url` and the agent crawls it; the entry reports a `crawl_status` (e.g. `pending`, `in_progress`, `completed`, `failed`) and how many pages were crawled. ```bash curl -s -X POST https://waba-v2.360dialog.io/agent_config/websites \ -H "D360-API-KEY: " \ -H "Content-Type: application/json" \ -d '{ "url": "https://www.example.com" }' ``` [Add a knowledge website](/docs/mba/api-ref/knowledge-websites/add-a-knowledge-website.md) *** ## Behavior ### Skills Skills are the agent's behavioral instructions — its tone, priorities, and brand voice. They're also how you **teach the agent your processes**: encode an existing business workflow (how you qualify a lead, take a booking, triage a support request) as ordered steps in a skill, and the agent follows it. This lets you reuse the processes you already run rather than inventing new ones. {% hint style="info" %} **Write skills as clear directives, and avoid conflicting priorities.** The agent follows skills literally. If two skills both claim to go "first" on the same trigger, the agent can produce duplicate or inconsistent replies. When several actions should happen on one trigger, put them in **a single skill with explicitly ordered steps** rather than splitting them across skills. {% endhint %} Each skill has three parts: * `title` — a short name, **lowercase letters, numbers, and hyphens only** (e.g. `product-return-policy`), max 64 characters. * `description` — *when* to apply the skill (e.g. "Apply when the customer asks about returns or refunds"), max 1024 characters. The agent uses this to decide relevance. * `skill` — the actual instructions, max 20,000 characters. ```bash curl -s -X POST https://waba-v2.360dialog.io/agent_config/skills \ -H "D360-API-KEY: " \ -H "Content-Type: application/json" \ -d '{ "title": "polite-grounded-support", "description": "Apply to every customer message as the main support persona.", "skill": "You are the support assistant for our shop. Be warm, polite, and concise. Answer ONLY from the business info, FAQs, files, and tools provided. If the answer is not covered, do not guess — apologize briefly and hand off to a human. Never invent prices, stock, or policies." }' ``` [Create a skill](/docs/mba/api-ref/knowledge-skills/create-a-skill.md) ### Handoff Handoff hands the conversation from the agent to a human. It's part of the agent **settings** object — set `handoff.enabled` and an optional `message`. ```bash curl -s -X PUT https://waba-v2.360dialog.io/agent_config/settings \ -H "D360-API-KEY: " \ -H "Content-Type: application/json" \ -d '{ "handoff": { "enabled": true, "message": "Connecting you to a human agent." } }' ``` {% hint style="warning" %} **Handoff behavior to design around:** * **Once the agent hands a chat to a human, it stops answering that chat.** A working human-handover path is essential — otherwise those customers get no replies. * **The handoff notice is the fixed `message` you define** — it can't be adapted per customer language. For a multilingual audience, keep the message simple or language-neutral. * **On hand-back, the agent loses the earlier conversation context.** Don't assume it remembers what was said before the human stepped in. * **There's no endpoint to read who currently holds the thread** — infer it from the `standby` vs `messages` routing and the `messaging_handovers` events on your webhook. See [Webhooks](/docs/mba/webhooks.md) {% endhint %} [Update AI agent settings](/docs/mba/api-ref/onboarding/update-ai-agent-settings.md) ### Follow-up Follow-up sends a nudge after a period of customer inactivity. Also part of **settings** — set `followup.enabled`, a `message`, and `followup_interval_in_seconds`. The interval must be one of: `0` (disabled), `300`, `900`, `1800`, `3600`, `7200`, `28800`, or `86400` seconds. {% hint style="info" %} `settings` is a singleton (one per number) and is replaced with `PUT`. It also holds `rollout` (on/off) and `ai_audience` — the two levers you'll use in [Go live](/docs/mba/meta-business-agent/5.-go-live.md) and [Roll out to everyone](/docs/mba/meta-business-agent/6.-roll-out-to-everyone.md). {% endhint %} ```bash curl -s -X PUT https://waba-v2.360dialog.io/agent_config/settings \ -H "D360-API-KEY: " \ -H "Content-Type: application/json" \ -d '{ "followup": { "enabled": true, "followup_interval_in_seconds": 3600, "message": "Is there anything else I can help with?" } }' ``` [Update AI agent settings](/docs/mba/api-ref/onboarding/update-ai-agent-settings.md) *** ## Connectors and tools Connectors let the agent do more than answer questions — they call your own APIs to look up an order, book an appointment, or check inventory. It's a two-level setup: {% hint style="info" %} **Write descriptions like you're briefing the agent.** Give each connector and tool a clear `name` and `description`, and define the request body with explicit field types, descriptions, and required fields. Vague descriptions are the top cause of the agent calling a tool at the wrong time — or not at all. See the **Meta Business Agent API reference** for the full `request_definition` and user-auth injection schema. {% endhint %} {% stepper %} {% step %} ## Connector Defines an external API: `name`, `description`, `base_url`, and `auth_type`. Supported auth types today are `OAUTH2_CLIENT_CREDENTIALS`, `API_KEY`, and `NONE`. ```bash curl -s -X POST https://waba-v2.360dialog.io/agent_connectors \ -H "D360-API-KEY: " \ -H "Content-Type: application/json" \ -d '{ "name": "Order Management", "description": "Looks up customer orders and returns status and tracking.", "base_url": "https://api.yourshop.example/v1", "auth_type": "API_KEY" }' ``` [Create a connector](/docs/mba/api-ref/knowledge-connectors/create-a-connector.md) {% endstep %} {% step %} ## Connector tool A single operation on that connector: `name`, `description`, and a `request_definition`. The agent reads these to decide when to call the tool and how to build the request. ```bash # 2) Add a tool to it (use the connector id from the response above) curl -s -X POST https://waba-v2.360dialog.io/agent_connectors//tools \ -H "D360-API-KEY: " \ -H "Content-Type: application/json" \ -d '{ "name": "check_order_status", "description": "Use when a customer asks about an existing order. Requires an order ID. Returns status, estimated delivery, and tracking.", "user_auth_required": false, "request_definition": { "method": "GET", "path": "/orders/{order_id}" } }' ``` [Create a connector tool](/docs/mba/api-ref/knowledge-connector-tools/create-a-connector-tool.md) {% endstep %} {% endstepper %} {% hint style="warning" %} **MCP connectors are not available yet.** Support for MCP-based connectors is on Meta's roadmap (requested by many partners) but not shipped. For now, integrate via the REST connector model above. {% endhint %} --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.360dialog.com/docs/mba/meta-business-agent/3.-configure-your-agent.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.