Nutshell

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 small-team CRM that replaces Nutshell.

The rule that decides whether a CRM is used: **if a salesperson has to type what they already sent, they will not.** Build the capture first and the reporting second.

STACK
- Node 20+ with Fastify, server-rendered HTML with a little vanilla JS
- SQLite through better-sqlite3, WAL mode, FTS5 over notes and messages
- IMAP or a mail API for sync
- Caddy in front

THE DATA MODEL
- companies: id, name, domain, size, notes
- people: id, company_id, name, email, phone, role
- deals: id, company_id, title, value_cents, currency, stage, probability, expected_close, owner_id, created_at, closed_at, outcome, lost_reason
- stage_changes: id, deal_id, from_stage, to_stage, at, actor — **append-only, and every pipeline report is computed from it**
- activities: id, subject_kind, subject_id, kind, body, at, actor — calls, meetings, notes
- messages: id, deal_id, person_id, direction, subject, body_text, message_id, at
- tasks: id, subject_kind, subject_id, title, due_at, done_at, owner_id

EMAIL CAPTURE, WHICH IS THE WHOLE ADOPTION QUESTION
- Sync the salesperson's mailbox and match messages to people by address, and to deals by the person
- **A blind-copy address that files a message against the right deal**, for anything sent from a phone
- Match ambiguity resolved by asking, never by guessing. A message filed against the wrong deal is worse than an unfiled one
- **Only sync messages with people who are in the CRM.** Syncing an entire mailbox pulls in a salesperson's private correspondence, which is a privacy problem and an employment one

THE PIPELINE
- Stages with a probability, and a deal value in integer minor units with its currency
- Weighted forecast from value times probability, and a second figure from the historical conversion rate at each stage. **Show both** — the second is usually the true one and the gap between them is the sales manager's most useful number
- Time in stage, and an ageing list of deals that have not moved. That list is the weekly meeting
- A lost reason, required on close. **Without it you learn nothing from losing**, and it is the cheapest research a company can do

WHAT TO REPORT
- Conversion between each pair of stages, from the stage change history
- Median days to close, and the distribution rather than the mean
- Activity per deal against outcome, so 'more calls' can be checked rather than asserted

MAKING IT USED
- One screen per deal with everything on it: the people, the messages, the notes, the next task
- A next action on every open deal, and a list of deals without one. **That list is the only nagging a CRM should do**
- Fast keyboard entry for a note, because that is the action performed twenty times a day

WHAT MATTERS MOST
Automatic email capture, a required lost reason, and a next action on every deal. The first makes it used; the other two make it worth using.