Hyperportal

An open-source dataroom you build and manage with an AI coding agent (Claude Code, Codex, Cursor — anything that can run a CLI). You own the code, the deployment, and the analytics.

Website: hyperportal.dev — overview, use cases, and the waitlist for the hosted version.

Share documents with investors, acquirers, landlords, or lawyers, and know exactly what's happening:

• One private link per recipient organization — no accounts, no sign-up wall for your readers. Their team shares one link internally.
• See who read what — page views, document downloads, and engagement per recipient (PostHog, in your own project).
• Revoke a link and it's actually dead — the CLI verifies against the live site before reporting success, instead of trusting a dashboard edit.
• Staged disclosure — start everyone with the curated room, then unlock financials or sensitive sections per recipient as trust grows. Sections can show as "on request" or stay completely invisible until unlocked.
• Content is just MDX in a git repo — which is exactly why an agent can manage the whole thing: write sections, add documents, grant and revoke access, and read the engagement stats for you.

One codebase covers fundraising datarooms, M&A datarooms, rental applications, and visa/green-card evidence binders — same machine, different content. Extracted from a data room used in a real transaction; the operational judgment here (what leaks, what silently fails) comes from production use.

Quickstart

npx -y @hypersocial/hyperportal new acme-fundraise
cd acme-fundraise && pnpm install

# local dev with a throwaway token
PORTAL_TEAMS='[{"token":"dev","org":"Local","active":true}]' pnpm dev
# open http://localhost:3000/r/dev

Then edit lib/sections.ts and content/*.mdx for your domain, wire Vercel, verify with npx -y @hypersocial/hyperportal doctor --url https://…, and grant real links with pnpm portal add "Org Name".

Using with an AI agent (recommended)

Hyperportal ships as an agent skill. With it installed, Claude scaffolds the portal, walks the disclosure design with you (who are the recipients, what unlocks at which stage, what must never leak), and then operates it day-to-day — granting links, revoking them, adding documents, pulling engagement stats:

npx skills add hypersocialinc/hyperportal   # or:
npx -y @hypersocial/hyperportal skill    # installs into ~/.claude/skills

Then just ask: "build me a fundraising dataroom".

What's in the box

Part	What it is
template/	A deployable Next.js portal: MDX sections, /r/<token> gate, tiers, pnpm portal CLI, PostHog analytics. Works out of the box with sample content.
bin/hyperportal	Scaffolding CLI: new stamps a portal, doctor verifies one, skill installs the agent skill.
skills/hyperportal/	The Claude skill: workflow plus judgment — access model, leak traps, operations, per-domain recipes.

The model

• One link per counterparty org — /r/<token> sets an HttpOnly cookie and redirects, so tokens never live in history or referrers. Tokens live in an env var, never in git, managed only by a CLI whose every mutation redeploys and probes the live URL before reporting success.
• Tiered disclosure — each section has minTier and belowTier: "teaser" | "hidden": below-tier recipients either see an "on request" row, or never learn the section exists. Enforced server-side.
• Leak discipline — images route through a media map so blob paths can't cross-name counterparties; the OG image is static and wordmark-only because unfurlers cache it forever, outside the gate.
• Analytics as courtesy, not gate — engagement is grouped per counterparty; nobody has to identify themselves to read.

See skills/hyperportal/references/ for the full write-ups — they double as documentation for humans.

License

MIT