Flowly
Log inDownload
Use cases/Automation

n8n Workflow Orchestration

Use n8n for the integration layer, Flowly for the agent layer — the AI calls webhooks, n8n handles credentials and side effects.

Automationmedium~1h setup
Tools
exec
Plugins
n8n-bridge
Channels
telegramdesktop

A common architecture failure: giving your AI agent direct credentials to every SaaS — Slack tokens, Salesforce keys, Stripe secrets. Each one is a foot-gun. n8n is built for that layer: visual workflows, credential vault, idempotency. Flowly handles natural language and reasoning. Bridge them with webhooks: the agent calls n8n, n8n executes the actual side effects.

What it does

  • Agent calls n8n workflows via webhook tools — no SaaS credentials in agent context
  • n8n handles auth, retries, error states
  • Audit trail lives in n8n's execution history
  • Easy to add new integrations: build the n8n workflow, register a tool in Flowly, done
  • Idempotency at the n8n layer — agent can be retry-safe

What you'll need

  • n8n running somewhere (cloud, self-hosted, or in-app)
  • An n8n-bridge plugin
  • HMAC secret for webhook authentication

Setup

1. Build your first n8n workflow

In n8n, create a workflow that exposes a webhook. Example: "Send Slack message". Inputs: channel, message. The workflow uses your stored Slack credentials, calls the API, returns success/failure.

Add HMAC validation at the webhook level so only requests signed with a shared secret are accepted.

2. Build the n8n-bridge plugin

Send to Flowly
Create a plugin called "n8n-bridge". Tools: - n8n_call(workflow_id: str, payload: dict) -> str Implementation: - Read N8N_BASE_URL and N8N_HMAC_SECRET from env - For each call, compute HMAC-SHA256 of the payload using the secret - POST to {N8N_BASE_URL}/webhook/{workflow_id} with payload + signature header - Return n8n's response as a string Plus high-level convenience tools that wrap n8n_call: - slack_message(channel: str, message: str) -> str - create_jira_ticket(project: str, title: str, description: str) -> str These call n8n_call with hardcoded workflow_ids configured in env.

3. Register tools per workflow

For each n8n workflow you build, add a thin tool wrapper. Example prompt to agent:

Send to Flowly
For each meeting summary I share, call slack_message with channel "#team-updates" and the summary.

The agent calls slack_message, the plugin calls n8n_call, n8n authenticates and posts. None of those credentials touch the LLM context.

Tips

  • Use n8n for anything credentialed. Stripe, Salesforce, Twilio, Slack — all live in n8n. Keep the agent context credential-free.
  • Idempotency keys. Pass a deterministic key to every workflow so retries don't double-charge / double-post. n8n has good support for this.
  • Version workflows. When you change a workflow's behaviour, bump its ID. Old plugin versions stay pointed at old workflow versions until you upgrade.
  • Audit via n8n, not the agent. When something didn't fire, n8n's execution log is the source of truth. The agent thinks it succeeded if it got HTTP 200 back; n8n shows you what actually happened on the SaaS side.
  • Limit by namespace. Don't expose every workflow to every agent session. Use the plugin's check_fn to gate access — e.g. only enable Stripe-related tools when a billing skill is loaded.