ablute_ Investor CRM
Investor-outreach CRM for a seed round that enforces outreach discipline — it doesn't just store data.
Built for ablute_'s Seed Round 2026 (€1.3M); multi-tenant by design so other founders can be onboarded later.
What's inside
| Area | Where |
|---|
| Pipeline table (home) | / |
| Today / Next Best Action queue | /today |
| Entity & person profiles (pre-flight "Can I contact?") | /entities/[id], /people/[id] |
| Meeting prep one-pager | /people/[id]/prep |
| Log-an-interaction flow + live message linter | /log |
| Agenda (month view, ICS export) | /agenda |
| Dashboard (funnel, round progress, pass reasons, overrides audit) | /dashboard |
| Documents & Data Room (folders, grants, engagement) | /documents |
| Investor portal (external, magic-link) | /portal |
| Automations (draft_review / full_auto) + Outbox approval queue | /automations, /outbox |
| AI Review (paid, Anthropic API) | Settings + /api/ai-review |
Business rules enforced in code (src/lib/rules.ts)
- Pre-flight before any outbound: researched hook · no open hard filter · 14-day contact lock
(one approach per entity) · seniority order · verified-email-only for email sends (no override) ·
volume caps (5/day, 20/week) · never a third unanswered message (no override) · official
submission channel first · do-not-contact hard stop (no override).
- Message linter: kill words, 900-char LinkedIn limit,
/edit links blocked, generic-line-1 warning, one-small-ask reminder.
- Pass reasons required; same category at 3+ entities → "the pitch may be the problem" alert.
- Overrides always demand a justification and are written to an audit log.
- Automations run in
draft_review (Outbox approval) or full_auto — full-auto executes only
when pre-flight is green; otherwise the run falls into the Outbox with the reason. LinkedIn has no
official send API, so LinkedIn automations always produce ready-to-paste drafts.
Running it
npm install
npm run dev # http://localhost:3000
Demo mode (default): with no env vars the app runs entirely in the browser with the real
ablute_ pipeline seeded (14+ entities / 20 people, hard filters, kill words, waves). State persists
in localStorage. Settings → "Reset demo" restores the seed.
Connecting Supabase:
- Create a Supabase project, then run
supabase/migrations/0001_init.sql (SQL editor or CLI) and supabase/seed.sql
(regenerate it any time with node scripts/gen-seed-sql.mjs).
- Copy
.env.example → .env.local and fill the Supabase keys.
- The data layer swap point is
src/lib/store.tsx (documented inline) — action semantics
(locks, follow-up tasks, overrides, runs) are pure functions in src/lib/rules.ts, shared by both modes.
Deploy (Vercel): import the repo, set env vars, deploy. vercel.json schedules the hourly
automation tick (/api/automations).
Phase keys (all optional until the phase is used): Resend (email sends; verify the ablute.pt domain
first), Stripe (billing), ANTHROPIC_API_KEY (AI Review).
Deliberately out of scope
- Automated LinkedIn sending (no official API; automating risks the account).
- Sending to guessed/unverified emails — never, no override (30–40% bounce destroys sender reputation).
- LinkedIn scraping.
GDPR
B2B legitimate-interest basis; per-person data_source field (Art. 14); do-not-contact toggle
purges research fields immediately and blocks outbounds permanently; investor-portal views are
logged and disclosed.
