Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.vorel.ai/llms.txt

Use this file to discover all available pages before exploring further.

Vorel splits its automation surface into two layers:
  • Per-turn dispatch (router → sub-agent → tool) runs in-code as TypeScript inside the Vorel web app. Type-checked, version-controlled, code-reviewed. This is the synchronous AI brain that handles every customer message + voice turn.
  • Around-the-brain workflows (post-booking nudges, nurture cadences, weekly digests) run in n8n — operator-editable visual workflows that consume Vorel’s public API + outbound webhooks.
This page covers the around-the-brain layer.
n8n service status: not yet deployed on Railway. The around-the-brain templates ship in the repo (handoff/n8n-workflows/*.template.json) and the operator-side discoverability surface lists them, but the n8n service itself is not standing up on Railway today. Stand-up runbook: handoff/docs/runbooks/n8n-setup.md (~30 min). Once deployed per-tenant templates are imported during onboarding.

​
The 3 v1 templates

Post-booking confirmation

Trigger: Vorel webhook on booking.created.Flow: When the agent books an appointment, fan out — Slack notify the assigned user, push to the tenant’s CRM via Vorel’s create-record endpoint, and schedule a 24-hour-before reminder via Vorel’s outbound conversation send. Appointment data comes from the webhook payload — no extra GET needed.Vorel API surface: POST /v1/crm/create-record · POST /v1/conversations/:id/sendSource: handoff/n8n-workflows/post-booking-confirmation.template.json.

Lead nurture (3-day)

Trigger: n8n schedule trigger, daily 09:00 cron (0 9 * * *).Flow: Daily — find leads stale-qualified for 3+ days, send a re-engagement WhatsApp template via Vorel’s outbound conversation send; escalate to human handoff after 7 days. The lead-list filter ?stale_for_days=N is the vendored shortcut for the cohort.Vorel API surface: GET /v1/leads?stale_for_days=3 · POST /v1/conversations/:id/send · POST /v1/leads/:id/handoff · PATCH /v1/leads/:idSource: handoff/n8n-workflows/lead-nurture-3-day.template.json.

Weekly QA rollup

Trigger: n8n schedule trigger, Monday 09:00 cron (0 9 * * 1).Flow: Aggregate QA scores + hallucination flags + handoff reasons from the prior 7 days; format the rollup; email or Slack the digest to the operator. Single-endpoint workflow backed by weekly-rollup.Vorel API surface: GET /v1/analytics/weekly-rollupSource: handoff/n8n-workflows/weekly-qa-rollup.template.json.
All three templates are status: ready — every endpoint they call is shipped on the public API. A CI test (apps/web/src/app/admin/automation/templates-validity.test.ts) cross-references each httpRequest URL in every template against the live OpenAPI spec; a renamed or removed endpoint would fail the build before the template silently breaks.

​
How operator deployment works

When n8n is stood up, the operator runs through this on each new tenant:
1

Stand up n8n on Railway

Stand up n8n as a 5th Railway service per handoff/docs/runbooks/n8n-setup.md.
2

Cloudflare Access in front of n8n

Configure Cloudflare Access on the n8n custom domain so only the operator team can hit it.
3

Issue a per-tenant Vorel API key

Create a per-tenant Vorel public API key at /admin/tenants/[id] → API keys with the scopes each template needs (leads:write, conversations:write, crm:write, read).
4

Import the .template.json files

Import each template into n8n; configure credentials (the API key) + the cron timezone (per tenant’s local time).
5

Wire webhook subscriptions

For webhook-triggered workflows (post-booking-confirmation), wire the Vorel-side webhook subscription at /admin/tenants/[id]/integrations/webhooks so the right event types POST into the right n8n webhook URL.

​
Around-the-brain vs. in-brain — the split

Per-turn agent dispatch is in code, not n8n:
  • The router → sub-agent → tool flow runs in-process via runChatDispatch / runVoiceDispatch.
  • Each tool route at /api/tools/* is JWT-authed (5-min TTL) and tenant-scoped via withTenantContext.
  • This stays type-checked, code-reviewed, and version-controlled — the synchronous AI brain is too load-bearing to live in a visual editor.
n8n handles the asynchronous, cross-conversation, operator-editable layer:
  • Post-booking nudges, nurture cadences, weekly digests, custom escalation flows.
  • Operators clone + tune templates per-tenant without a code deploy.
  • Each tenant’s workflows are isolated by API key.

​
Custom workflows beyond the v1 templates

Tenants who want bespoke workflows go in two directions:
  • Subscribe to webhooks at /(dashboard)/settings/integrations/webhooks — receive lead.created / booking.created / 10 other event types in your own automation system. HMAC-SHA256 signed; 6-attempt retry ladder.
  • Pull from the public API — every resource (conversations, leads, appointments, offerings, analytics) has GET (list) + GET (id) + POST + PATCH parity. Build whatever cohort / cron / fan-out you need.
The n8n templates are the starting point, not the limit. They’re deliberately small (1–4 nodes each) so the operator can fork them into tenant-specific variants.

​
What’s NOT supported

  • n8n itself, today. Templates exist on disk; n8n is not deployed on Railway. Around-the- brain workflows aren’t running until the runbook is executed.
  • Vorel-managed scheduling. Crons live in n8n; Vorel doesn’t run a cron service today.
  • Visual workflow editor inside Vorel’s dashboard. n8n is the editor; we don’t ship a bespoke one.
  • End-customer-facing workflow editing. n8n access is operator-team-only behind Cloudflare Access. Tenants describe what they want during onboarding; the operator builds the workflow per-tenant.