SynapsAI in n8n and Make

Your workflow stays the orchestrator; SynapsAI becomes one step in it. A form, a CRM, a spreadsheet or your messaging provider sends a message to an agent, the agent answers with your knowledge base, catalog and tools, and the reply comes back into the same workflow.

You want to…	Use
Ask an agent from a workflow and use the answer	Ask agent (n8n node / Make module), with Wait for Reply on n8n
React when the agent replies, an operator replies, a chat ends, a person is requested or a lead is captured	SynapsAI Trigger (n8n) / the Make instant triggers
Read conversations (filter by CSAT), contacts and leads	the list and get operations
Create or update a contact from your own data	Create or update contact

Start here:

How it works

 Your workflow                         connect.synapsai.app                   SynapsAI agent
     │ POST /v1/agents/{agent}/messages ──►  202 {messageId}  ──► queued turn (same engine as WhatsApp / web chat)
     │                                                                  │
     │ ◄── signed POST message.replied  (to your callbackUrl or your "Agent replied" subscription)
     │ ◄── signed POST message.operator (a console operator answered)
     │ ◄── signed POST message.failed   (e.g. INSUFFICIENT_CREDITS)
     │      GET /v1/messages/{messageId}  — read the status and reply at any time

Your workflow posts the message. SynapsAI answers 202 at once with a messageId: the agent works asynchronously, exactly as on every other channel.
When the agent answers, SynapsAI posts a signed message.replied to the callbackUrl you sent — on n8n, the node’s own resume URL — or, when you sent none, to every Agent replied subscription of that agent. Never both.
Failed deliveries are retried (30 s, 1, 2, 4, 8 min). The reply always stays readable with GET /v1/messages/{messageId}.

The same externalUserId continues the same conversation, so a customer keeps their history and memory. SynapsAI never merges that contact with a WhatsApp or web-chat contact on its own — even when the phone matches. To unify them, call Create or update contact or send the existing contactId.

Plans, keys and limits

Plans: GROWTH, PRO and ENTERPRISE. Each API message is priced like a message on any other channel.
API keys: mint one in the console, Account → API keys. Pick only the scopes the workflow needs:

Scope	Allows
`messages:write`	send messages (Ask agent), read their status
`conversations:read`	list and read conversations; subscribe to `message.replied` / `message.operator`; the conversation sections of subscriptions (intel, lead, handoff, operator, messages, transcript, memory)
`contacts:read`	list and read contacts; the `contact` and `operatorEmail` sections of subscriptions
`contacts:write`	create or update contacts; send `contactId` with a message
`leads:read`	list leads; subscribe to `lead.captured`
`hooks:write`	subscribe and unsubscribe triggers
`agents:read`	the agent picker

Agent API channel: turn it on per agent on the agent’s Extensions page, card Workflow API (n8n · Make) → Turn on. Turning it on creates the agent’s signing secret (shown once; Rotate secret there). Ask agent with Wait for Reply uses it to check each reply.
Rate limits: 60 requests per minute per key and 600 messages per minute per account, plus edge limits per key and per IP (all 429 RATE_LIMITED; wait Retry-After). At most 20 messages per minute for one externalUserId (a loop guard) — answered 429 LOOP_GUARD for that user only.
Subscriptions: up to 20 API subscriptions (triggers) per agent. They appear on the agent’s hooks page with an API badge; you can delete them there.
Callbacks: HTTPS on port 443, a public host name — never an IP address, a private network or a redirect.
Text only in this version: attachments are not accepted.