Freshdesk

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me the help desk that replaces Freshdesk.

The centre of this is email threading, and it is harder than it looks. **Get the threading wrong and the whole product is wrong**, so it comes first.

STACK
- Node 20+ with Fastify, server-rendered HTML with a little vanilla JS
- SQLite through better-sqlite3, WAL mode, FTS5 across tickets
- IMAP or a mail webhook for inbound, SMTP for outbound
- Caddy in front

THE DATA MODEL
- tickets: id, subject, requester_id, assignee_id, status, priority, opened_at, first_response_at, resolved_at, sla_due_at
- messages: id, ticket_id, direction, author, body_text, body_html, message_id, in_reply_to, references_json, is_internal, at
- requesters: id, email, name, org_id
- events: id, ticket_id, kind, actor, detail_json, at — **append-only, and this is how 'what happened to this ticket' is answered**
- macros, tags, and sla_policies as small tables

EMAIL THREADING, WHICH IS THE HARD PART
- Store Message-ID, In-Reply-To and References on every message, and thread on those first. **That is the only reliable signal**
- A token in the reply-to address as a fallback, because plenty of clients break the headers
- Subject matching last, and only with a strict pattern. Loose subject matching merges two customers' unrelated 'Question' emails into one ticket, which is a data leak
- Strip quoted history when displaying, keep it in the stored body. Every client quotes differently and the heuristics never fully work — so **never delete what you stripped**
- Detect auto-replies and out-of-office by their headers and do not reopen a ticket for them. An auto-responder loop between two help desks is a classic and embarrassing incident

SENDING
- From your own domain with SPF, DKIM and DMARC configured properly, or your replies land in spam and the customer thinks you ignored them
- Set References correctly on outbound so the customer's client threads it
- Every outbound message queued with retries and a dead-letter list

SLA AND AUTOMATION
- A policy per priority: first response and resolution targets, business hours aware with the right timezone
- **Compute the due time from business hours, not wall clock.** A four-hour target starting at five on Friday is not due at nine on Friday night
- Automations as ordered rules with a run cap per ticket, so two rules cannot trigger each other forever

WHAT MANAGERS ACTUALLY NEED
- First response and resolution times as medians and 90th percentiles, by hour and weekday
- Backlog age, and the count of tickets with no reply at all — the number that justifies a hire
- Reopen rate per agent, which finds premature closing

THE PORTAL
- Ticket history by token link, no account required, and a searchable knowledge base with the zero-result queries logged

WHAT MATTERS MOST
Header-based threading and business-hours SLA. The first keeps conversations correct and private; the second keeps the targets meaningful.