open-dataroom

Built by Grona — and shared as MIT for any founder, fund, or operator who needs an investor data room they can host themselves.

A self-hostable investor data room. Invite-only document portal, PDF / video / spreadsheet viewers, RAG-grounded AI chatbot, and a full admin control plane. Built on Next.js 16, Mongo, and your choice of Gemini / OpenAI / Anthropic. Fork it, brand it, deploy it.

---

The repo ships as the Acme Capital demo — make it yours with AI

main ships with a fully-applied Acme Capital brand kit (burnt orange + mustard, mid-century industrial catalogue, cartoon team photos, "Top Secret Anvil Project" persona). This is on purpose: it gives you a real working portal to clone-and-customise from, with a complete BRAND.md already documented.

The fastest way to make it yours is to ask an AI agent. Open Claude Code, Cursor, or your editor's AI assistant inside this repo and paste the prompt in REBRAND.md along with whatever you know about your company. The agent will:

• Rewrite dataroom.config.ts (brand name, tagline, persona, tier names, team bios)
• Swap the colour ramp in src/styles/theme-brand.css
• Generate new SVG logos + favicon
• Re-render the OG card via the existing Playwright template
• Replace the team page content
• Write you a new BRAND.md to match

A full rebrand from "Acme" to "your company" takes one prompt and ~10 minutes of AI work. See REBRAND.md for the prompt template.

Prefer manual? Each layer is a single file:

Layer	File
Brand name, tagline, persona, tiers, team	dataroom.config.ts
Colour ramp	src/styles/theme-brand.css
Logo, favicon, OG image	public/assets/branding/
Team photos	public/assets/team/
Local-knowledge markdown	content/*.md
Brand guidelines (for AI agents to read)	examples/acme-capital/BRAND.md

Just kicking the tires? Two other complete brand kits ship as examples — try the one closest to what you want and modify from there:

npx tsx bin/use-example.ts greenwood-holdings --yes    # Penguin Classics, dusty navy + brass
npx tsx bin/use-example.ts lighthouse-labs --yes       # Indie SaaS, indigo + amber
npx tsx bin/use-example.ts acme-capital --yes          # back to the default

Each has its own BRAND.md documenting voice, palette, type, and don'ts — useful inputs to an AI rebrand prompt.

---

Why this exists

Most investor data rooms are commercial SaaS — fine, but you pay per seat and you put your most sensitive documents on someone else's infrastructure. The ones that aren't SaaS are usually a glorified Dropbox folder with no audit trail, no per-investor access control, and no easy way to answer the questions investors will ask.

This template is the middle ground: a tiny, opinionated codebase you can run on your own Vercel + Mongo Atlas account for a few dollars a month, with everything an investor needs (gated documents, team page, videos, RAG chatbot) and everything an operator needs (invites, revocation, sync, audit log).

It's deliberately small. The whole src/lib/ is ~3000 lines. One person can hold it in their head.

What's in the box

For investors

• Invite-only signup (one-time tokenised link, password choice on accept)
• Document portal with PDF / video / spreadsheet viewers
• Team page, videos page, optional both
• AI chatbot grounded in your pitch documents (cites sources, falls back gracefully when it doesn't know)
• Per-tier access — different investors see different documents

For operators

• Admin shell separate from the portal
• Invite, revoke, restore investors (with audit trail)
• Google Drive sync on a cron — write documents in Drive, they appear in the portal
• "Re-index knowledge base" button for the RAG layer
• Activity log of every login, view, and admin action

For developers

• Single-file config (dataroom.config.ts) — brand, tiers, providers, persona, navigation
• Pluggable LLM provider (Gemini / OpenAI / Anthropic for chat; Gemini / OpenAI for embeddings)
• Interactive setup wizard (npm run setup) with connectivity tests — handles .env.local only; the brand is yours to own
• Pre-flight validator (npm run check) — catches misconfigured env (and reminds you if you're still on a shipped demo brand) before you boot
• Three fully-branded example datasets (Acme Capital · Greenwood Holdings · Lighthouse Labs) — each ships a BRAND.md with voice, palette, type, logo system, applications, and don'ts; pasteable into an AI agent for a custom rebrand

Quickstart (10 minutes)

Prerequisites: Node 20+, a free Mongo Atlas cluster, an API key for one of Gemini / OpenAI / Anthropic.

git clone https://github.com/my-username/open-dataroom
cd open-dataroom
npm install
npm run setup       # operational wizard — writes .env.local (DB, API keys, admin seed). Leaves the Acme demo brand intact; rebrand via REBRAND.md.
npm run check       # pre-flight validation (✓/✗/⚠/· per category)
npm run dev         # http://localhost:3000

Log in with the admin email/password the wizard printed. From there:

1. Go to /admin/invites and invite yourself a second email — verify the invite link arrives (or check the server log if you skipped SMTP).
2. Drop a PDF into your configured Drive folder. Click "Sync now" in the dashboard.
3. Click "Re-index knowledge base". Open the chat panel, ask a question grounded in the PDF.

If all three work, you have a functioning data room. Move on to docs/deploy.md for the Vercel push.

Try a worked example

npm run example:lighthouse-labs       # no tiers, Gemini-only, markdown KB
# OR
npm run example:greenwood-holdings    # one tier, Anthropic chat + OpenAI embeddings
# OR
npm run example:acme-capital          # multi-tier, OpenAI, custom RAG weights

Each command swaps dataroom.config.ts and content/ for the example's contents. Restore with git checkout -- dataroom.config.ts content/ public/ (plus git clean -fd content/ if the example added untracked files).

Architecture, in one paragraph

Next.js 16 App Router. Edge middleware does JWT signature checks, then API routes do the DB lookup + tokenVersion reconcile. Mongo (Mongoose) for everything: users, invites, documents, knowledge chunks, sync history, access log. Google Drive is the canonical document store; sync is a read-mirror. Keyword RAG over chunks of PDF/spreadsheet/markdown text; embeddings are computed and stored but not queried (vector swap is documented as a 30-min extension). The chat panel streams NDJSON from /api/chat. The LLM call is wrapped in an adapter interface so the three provider SDKs are interchangeable.

See docs/architecture.md for the longer version.

Configuration

Everything user-tunable lives in dataroom.config.ts. The shape is documented in docs/customize.md — every field, every footgun, with examples. The .env.local file holds secrets (Mongo URI, provider API keys, JWT secret); .env.local.example is the committed template.

Security

The template defends what a small data room reasonably can: authenticated sessions, tier-gated access, JWT revocation, audit logging, CSP headers. It does not prevent a logged-in investor from extracting a PDF via DevTools — no browser-rendered document does. Read docs/security.md before you go live; it's honest about the limits and lists the production-hardening steps (Upstash rate limit swap, AccessLog TTL, etc.).

Vulnerability disclosure: SECURITY.md.

Deploy

First-class target is Vercel. docs/deploy.md walks the push, the Atlas setup, Gmail SMTP gotchas, free-tier ceilings, EU data residency, and the post-launch operational checklist. Self-hosted is supported (it's stock Next 16); we don't ship a Dockerfile but a 20-line one works.

Docs

• docs/architecture.md — what's where, request paths, the LLM adapter contract
• docs/customize.md — every config field with examples
• docs/security.md — threat model, what's defended, what isn't
• docs/deploy.md — Vercel quickstart + ops checklist
• docs/rag.md — RAG pipeline + vector-search swap recipe
• docs/google-drive.md — folder convention, service account setup
• SECURITY.md — vulnerability disclosure
• CONTRIBUTING.md — PR scope philosophy + style
• CODE_OF_CONDUCT.md — short, in plain English

Non-goals

Things the template intentionally does NOT include — see CONTRIBUTING.md for the full list:

• Vector search (keyword retrieval ships; vector swap is documented)
• SSO / OAuth / magic-link auth
• Internationalisation (English only)
• Dark mode (deliberately removed; restoration recipe in docs/customize.md)
• Multi-tenant (one company per deployment)
• DRM / watermarking / true no-download enforcement

Stack

• Next.js 16 App Router with Turbopack
• Mongoose on Mongo Atlas
• Tailwind v4 with CSS custom properties
• @clack/prompts for the wizard
• @google/generative-ai, openai, @anthropic-ai/sdk — all three, behind a unified adapter
• unpdf + ExcelJS for text extraction
• nodemailer for transactional email
• googleapis for Drive sync

License

MIT. Use it commercially, fork it, modify it, sell it, don't tell us. We'd love attribution if you ship a public deployment — set brand.poweredByCreditUrl to your own fork's URL or leave the default badge.

Contributing

PRs welcome. Read CONTRIBUTING.md before opening one — the scope bar for the upstream repo is deliberately high (changes should benefit most forkers, not solve one deployment's problem). Most forks should diverge, not contribute back.