Resolvd

An end-to-end inbox operator. Most "AI support" tools just draft a reply and make a human re-read and click every action. Resolvd triages, drafts, and acts within policy, issues the refund, sends the order status, closes the ticket, and escalates only the cases that genuinely need a person, with the proposed action already attached.

Demo

▶ Watch the demo · Live: https://resolvd.agentpostmortem.com

Flow

1. A message hits POST /api/inbound (helpdesk webhook or email forwarder).
2. Triage classifies category, urgency, sentiment, and drafts a reply (Claude when ANTHROPIC_API_KEY is set; a keyword heuristic otherwise).
3. Policy (the guardrail) decides:
   • order_status with an order id -> auto-resolve (live Shopify lookup + reply)
   • refund at or under REFUND_AUTO_LIMIT, pinned to one order -> auto-issue + reply
   • refund over the limit, ambiguous orders, complaints, or negative+high-urgency -> escalate with the proposed action attached
   • anything needing an unconnected integration -> escalate with instructions instead of acting
4. A human approves/rejects escalations via POST /api/approve. Approving an executable action runs it for real (refund/order lookup) and records the result.

Integrations

Integration	Env	Without it
Shopify store	SHOPIFY_SHOP_DOMAIN, SHOPIFY_ACCESS_TOKEN (custom app: read_orders + write_orders)	order/refund work escalates with instructions
Reply email	RESEND_API_KEY, RESEND_FROM_EMAIL (verified sender)	replies saved as drafts for a human to send

Refunds resolve the order by explicit reference first, then by sender email only on an unambiguous single match. Multiple matches escalate rather than guess.

The dashboard shows every ticket, the auto-resolution rate, what action was taken (or proposed), and the reason.

Stack

Next.js 14 + Supabase (rv_* tables in the shared project) + Cloudflare Workers (OpenNext).

Run

npm install
cp .env.example .env.local   # SUPABASE_*, RESOLVD_INBOUND_TOKEN, REFUND_AUTO_LIMIT
npm run dev
npm run deploy

Apply supabase/schema.sql in the Supabase SQL editor once. On an existing database, also apply supabase/migrations/002_add_source.sql so demo rows get an explicit source = 'demo' tag (the dashboard also recognizes the built-in demo senders, so this is optional but recommended).

Demo tickets are badged DEMO in the dashboard and can be hidden with the "Hide demo tickets" toggle, so evaluation traffic never mixes with live work.

npm test   # policy + triage unit tests

Examples

# auto-resolved: order status with an order id
curl -X POST "$URL/api/inbound" -H "x-resolvd-token: $TOKEN" \
  -H "content-type: application/json" \
  -d '{"sender":"sam@x.com","subject":"where is my order","body":"status?","orderId":"1042"}'

# auto-resolved: small refund under the limit
curl -X POST "$URL/api/inbound" -H "x-resolvd-token: $TOKEN" \
  -H "content-type: application/json" \
  -d '{"sender":"jo@x.com","subject":"refund","body":"please refund $20 for the damaged item"}'

# escalated: refund over the limit -> waits for human approval
curl -X POST "$URL/api/inbound" -H "x-resolvd-token: $TOKEN" \
  -H "content-type: application/json" \
  -d '{"sender":"al@x.com","subject":"refund","body":"I want a $900 refund now"}'