How It Works

透明性のある、仕組み。

How it actually works.

What we touch, what we don't. Which AI runs the show. Where your data lives. We document SekiHQ at spec-sheet depth — this is not a black-box AI product.

From signup to running, in 4 steps

1

Sign up2 min

Email + password, or one-click with Google. A personal organization is auto-created with you as the owner. Invite teammates later, anytime.

  • 7-day free trial (card required at signup)
  • Cancel before day 7 — never charged
  • Multi-tenant by design: your org's data is fully isolated from others (row-level security)
2

Connect Gmail1 min

One-click OAuth with Google. SekiHQ pulls only subject lines, senders, and timestamps — message bodies are never accessed. Auto-syncs every 5 minutes, automatically building contacts + interaction history from new mail.

  • Read-only scope only (we never request gmail.send)
  • Disconnect anytime — we delete stored tokens immediately
  • Initial sync = last 7 days; ongoing sync = new mail only
3

AI analyzes your relationships~5 min

From the synced metadata, the AI computes warmth scores (0-100), keigo register, and recency per contact. Click "Enrich from emails" on any contact and the AI proposes job title, company, and a relationship summary.

  • Never overwrites your existing values — only fills empty fields
  • Keigo + summary always need your accept / edit / dismiss
  • Powered by Anthropic Claude — your prompts are not used for training
4

Draft, send, track

Generate keigo-correct draft replies. Review, edit, and send from your normal mail client. SekiHQ logs the send automatically and refreshes the warmth score in real-time. The Ma Timing engine surfaces the right next moment daily.

  • SekiHQ doesn't send mail — you send from your own Gmail / Outlook
  • Sent mail is auto-logged to interaction history
  • Warmth score recomputes in real-time

What we touch / what we don't

In Japanese B2B, data handling is a trust question, not a contract question. Full disclosure.

We do collect
  • Email metadata: subject, From, To, CC, timestamps
  • Contact info: name, company, title, phone, notes — whatever you input or import
  • Interaction summaries: email snippets (first 280 chars), interaction type (sent / received / meeting)
  • Account info: email, hashed password, Google account ID
  • Usage stats: logins, feature use, errors (non-identifying)
We do NOT collect
  • Full email bodies: never accessed, never stored. Snippets are Gmail's own snippet field (first 280 chars)
  • Attachments: never touched
  • Credit card details: handled entirely by Stripe — never stored on our servers
  • Your data for AI training: not sent to Anthropic or Google for training, ever
  • Third-party ad trackers: zero. No ad cookies. No cross-site tracking.

Data flow architecture

From your inbox to the dashboard, every step of the pipeline.

① Source
Gmail / Outlook
  • OAuth 2.0
  • Read-only scope
  • Sync every 5 min
② Processing
SekiHQ Edge
  • Vercel Edge Functions (US + Tokyo)
  • Metadata extraction only
  • Bodies discarded immediately
③ AI analysis
Anthropic Claude
  • Sonnet 4.6 model
  • Enterprise agreement
  • No training on your data
④ Storage
Supabase Postgres
  • Encryption at rest
  • Row-level security (RLS)
  • Org-level isolation
⑤ Delivery
SekiHQ App
  • HTTPS / TLS 1.3
  • JWT auth
  • Session cookie only
⑥ You
Your browser
  • Warmth visualized
  • Keigo drafts drafted
  • You send as normal

Security

Encryption in transit

All traffic encrypted with TLS 1.3. HSTS enforces HTTPS at the browser level.

Encryption at rest

Database (Supabase) and OAuth tokens are stored under enterprise-grade encryption.

Tenant isolation (RLS)

PostgreSQL row-level security keeps orgs physically isolated — even on shared tables, your data is invisible to other tenants at the SQL layer.

Password storage

Passwords are bcrypt-hashed by Supabase Auth. Plaintext is never stored, never logged.

Access controls

Three-tier role system inside each org: owner / admin / member. Enterprise plans add SAML SSO and audit logs.

Business continuity

Daily backups, 99.9%+ uptime target, 30-day data-export window after cancellation. Enterprise plans get custom SLA terms.

AI transparency

We disclose the AI model, contract terms, and data-handling policy in full. No vague "latest AI" claims — verifiable specifics only.

AI provider
Anthropic, PBC (San Francisco, USA)
Model in use
Claude Sonnet 4.6 (2026)
Contract
Enterprise API (zero data retention)
Used for training
No — Anthropic does not train on prompts sent via SekiHQ
Prompt retention
30 days (abuse-detection only), then permanently deleted
What's sent
Target contact's metadata, last 15 email snippets, your prompt
What's NOT sent
Credit cards, passwords, other users' data, full email bodies

Now that you've seen how it works, try it.

7-day trial · Card required at signup · Cancel before day 7, no charge

Start Free Trial