Quickstart — Make

You need: a SynapsAI account on GROWTH or above and a Make account. The SynapsAI app is available by invitation while it is in review — ask support for the invite link.

1. Key and API channel

Do steps 1 and 2 of the n8n Cloud quickstart: mint an sk_live_ key (scopes messages:write, agents:read, hooks:write, plus what your scenario reads — leads:read for Lead captured, contacts:read for a trigger’s contact) and turn on Workflow API (n8n · Make) on the agent’s Extensions page.

2. Install the app and connect

1. Open the SynapsAI invite link and Install the app in your organisation.
2. In a scenario add any SynapsAI module → Create a connection → paste the API key → Save. Make checks it with GET /v1/me and shows the key name and plan (below GROWTH it refuses with PLAN_FEATURE_UNAVAILABLE).

3. Scenario B — receive the reply

Create it first, so it is listening when the first answer comes back.

1. New scenario → SynapsAI → Agent replied → pick the agent → Save. The trigger subscribes itself to SynapsAI (an API badge appears on the agent’s hooks page).
2. Add what to do with the answer — e.g. Slack → Create a message with {{1.text}}.
3. Turn the scenario on (Immediately as data arrives).

Agent replied also fires when a turn fails (status = FAILED, with errorCode, e.g. INSUFFICIENT_CREDITS): add a filter on status when the two need different handling.

4. Scenario A — ask the agent

1. New scenario → SynapsAI → Ask agent.
   • Agent: yours. External user ID: demo-1. Text: Hi! What are your opening hours?
   • Optional Metadata (e.g. dealId) travels with the reply to n8n and your own endpoints; Make’s trigger does not pass it on (see below) — keep the messageId instead.
2. Run once. The module returns {messageId, status: "QUEUED"}; within seconds scenario B runs with the reply.

Correlate the two on messageId (the one Ask agent returned) or conversationId.

Prefer one scenario?

Use Ask agent → Tools › Sleep (10 s) → SynapsAI › Get a message, repeating until status is REPLIED or FAILED. Two scenarios are faster and cost fewer operations.

Every event is re-read with your key

Make sees the parsed webhook body, not its raw bytes, so it cannot check SynapsAI’s signature. The app does not trust the delivery at all: it is a notification. Each trigger re-reads the event by id with your connection’s API key and outputs only what it read back — from the posted body it keeps just the event type and delivery id.

Trigger	Re-reads	Key scope
Agent replied	GET /v1/messages/{messageId}	messages:write (+ conversations:read to subscribe)
Operator replied	GET /v1/conversations/{chatId} — operatorText is its last operator message	conversations:read
Conversation ended, Human requested	GET /v1/conversations/{chatId}	conversations:read
Lead captured	GET /v1/leads (newest 100 of the agent)	leads:read

An id from anywhere else does not resolve with your key (404), so a forged POST to the webhook URL cannot inject data: that run stops at the trigger with EVENT_NOT_VERIFIED and nothing reaches the next module. Add an Ignore error handler on the trigger to keep such runs quiet. For the person behind a lead or a conversation, add Get a contact (contacts:read).

Next

• Templates — import a ready blueprint.
• API reference.