Tasukiba — Project platform that turns search into creation

No credit card registration is required at sign-up. The default payment method during and after the trial is invoiced bank transfer. Unless you upgrade to Expert / Pro, no charges occur.

After the trial ends, the tenant moves into read-only mode, and you can still view data and export. 90 days later (180 days total from tenant creation), data is automatically deleted, so please upgrade, export, or cancel before that.

If you decide to come back after deletion, you can only restart on the Expert / Pro plan (the Beginner plan cannot be restarted). See Q6 for details.

There are three main differences.

Seat limits (unlimited) and monthly call limits (unlimited) are identical. Pro suits those who want a deeper AI and reason explanations; Expert suits those who want to use it heavily.

For each suggestion Tasukiba returns (“this looks related”), the “Why?” button asks the AI to explain in natural language why it was judged related.

• It summarizes “why this knowledge is similar to the current project,” using tags, purpose, scope, and historical context
• Useful for executive reports or external proposals where you need reasoning, not just relevance
• The explanation is generated by Pro’s Claude Sonnet

The Expert plan does not include the “Why?” button (the button itself is not shown on the UI). Suggestions themselves are available on both Expert and Pro, but the “explain the reason in words” experience is Pro-only.

Immediately. From the next API call after the upgrade, the new price and AI model apply.

Example: switching from Expert (Haiku, ¥5/call) to Pro (Sonnet, ¥15/call) → from the next operation onward, Sonnet runs and ¥15/call is counted.

If you switch mid-month, the month-end invoice shows a period-by-period breakdown (e.g., “Pro period ¥15 × 10 + Expert period ¥5 × 20 = ¥250”).

Also immediately. From the next operation after switching from Pro to Expert, Haiku (light & fast AI) and the ¥5/call rate apply.

Notes:

• The “Why?” feature is not available on Expert. The moment you downgrade from Pro to Expert, the “Why?” button disappears from the UI.
• Downgrading to Beginner is not possible. See Q6.

Expert ⇄ Pro switches are possible at any time. However, once you have selected Expert / Pro, you cannot revert to Beginner (the free plan).

This is a deliberate rule we set for the sake of operational sustainability. Whether you cancelled yourself, the trial expired and the tenant was deleted, or the system administrator deleted the tenant — in any case, restart is only possible on Expert / Pro. We do not re-provision a Beginner tenant.

Please think of Beginner as “for people trying Tasukiba for the first time only.” Export is always available, so we recommend backing up locally before cancelling (see Q16).

We plan to offer invoiced (bank transfer) and credit card as two options.

Organizational customers typically use invoiced payment. Individuals are currently on the same invoiced flow (we’ll reach out once credit card becomes available).

🧾 About qualified invoices (Japan)

The operator is not registered as a qualified invoice issuer at this time, and does not plan to register. For Japanese corporate customers using input tax credits, please rely on the transitional measures (80% credit through September 30, 2026 / 50% credit through September 30, 2029) or follow your company’s accounting policy.

🤔 Common follow-up questions

For invoiced payment, we operate on the following schedule.

Example: May usage

• Cutoff: May 31
• Invoice issued: by June 15
• Payment due: June 25

The invoice PDF is sent to the Billing Contact Email registered in tenant settings.

If the 25th falls on a weekend or holiday, the deadline is extended to the next business day.

Applies to Expert / Pro plans. Beginner is free, so no invoice is issued.

Don’t worry. The service does not stop abruptly. We respond in stages based on elapsed days.

Capabilities retained during read-only mode:

• ✅ Login and view (projects, knowledge, issues, retrospectives)
• ✅ Data export (your data ownership is our highest priority)
• ❌ New writes / edits / deletes / API-generated suggestions

Once payment is confirmed, we lift the restriction immediately and writing resumes from that moment (no loss of past content).

🧾 About late-payment interest

We do not currently set late-payment interest, prioritizing trust during the β phase. We may set it later at the statutory commercial rate; we will notify in advance if so.

🔐 Data protection

• Even if it comes to tenant deletion, we always preserve a 14-day export window after the deletion notice (formally preserved even when communication is lost)
• After physical deletion, invoice PDFs / monthly API-call aggregates / payment-clearing records are retained for 5 years (based on retention obligations under Japan’s Commercial Code and tax law)

Don’t worry. Tasukiba itself continues to work normally. Instead of “stop everything when the cap is hit,” we switch into a gentle “degraded mode” where only the AI behind the scenes pauses temporarily.

When does degraded mode start?

How is it lifted?

• Automatic: On the 1st of the next month (JST 09:00), counters reset to 0 and normal mode resumes automatically
• Manual (instant): Beginner can upgrade to Expert / Pro; Expert / Pro can raise the budget cap. Either lifts it instantly.

See Q11 for behavior during degraded mode.

In one line: It’s like an “energy-saving mode” where the AI takes a break, but normal work continues.

Think of it as similar to your phone switching to low-power mode when the battery is low. Because AI calls have a cost, when the cap is reached we pause only AI features while letting normal work continue, so the bill never blows past expectations.

✅ Still works (as usual)

• Writing: create / edit / delete projects, knowledge, risks, issues, retrospectives, memos
• Reading & searching: view and search past data
• CSV / Excel import: bring data in from files
• Recommendations: related-project suggestions continue (with slightly reduced precision; see below)
• Plan changes / cancellation / cap changes: always allowed (these are how you end the pause)

⏸️ Paused (the AI parts only)

• AI semantic understanding for new content: AI analysis of newly written memos / knowledge pauses (data created during degraded mode is saved without AI understanding)
• Auto-tagging: AI tag suggestion pauses (manual tagging still works)
• “Why?” button explanation generation (Pro only): Already-generated explanations remain viewable from cache; only new generation pauses

🎨 How does it look on screen?

• A yellow banner appears at the top of the screen so the mode is obvious at a glance
• Admins see an extra “Open settings” button for instant recovery via plan change or cap raise
• Regular users see “Please consult your administrator”

📊 What does “slightly reduced recommendation precision” mean?

The recommendation feature usually computes similarity on three axes:

Normal:    tag match 30% + text content 20% + AI semantic understanding 50%
Degraded:  tag match 50% + text content 50% + AI semantic understanding 0%

Memos created in degraded mode are stored without AI analysis, so recommendations rely on tag + text content alone. Connections like “the tags and text differ, but the meaning is close” become harder to find — but on the monthly reset (1st of each month, JST 09:00), missing AI analyses are backfilled in bulk, restoring full precision. (Up to 128 items per run; overflow carries into the following month.)

🔄 When does it lift?

• Automatic: monthly reset on the 1st (JST 09:00) returns you to normal mode
• If you want it back immediately (admin):
  • Beginner → upgrade to Expert / Pro
  • Expert / Pro → raise the monthly budget cap

Note: “Expert / Pro → Beginner” is not possible (see Q6).

One-liner: To prevent overspending, the AI takes a break — but normal work doesn’t stop. It automatically lifts next month, or instantly if the admin changes the plan or cap.

No. Cancellation completes in a few clicks from the tenant settings screen. No phone calls, no meetings, no back-and-forth with the operator. “Welcome those who come, and don’t block those who leave” is our policy.

It’s stored in a state of complete isolation per tenant. The design does not allow any view from another tenant. In addition, all data changes, authentication events, and permission changes are automatically recorded as an audit log.

See the “🛡️ Enterprise-grade security features” section above for a full list (MFA, account lockout, auto-deactivation of unused accounts, and more).

Short answer: It is not used for training. But during proactive suggestions and semantic search, data is sent to overseas AI services (US).

Here are the details.

📍 Where data is stored

The main database is in ap-northeast-1 (Tokyo region). The knowledge, risks, issues, and retrospectives you enter are all stored in Japan.

🌐 When cross-border transfer happens

When proactive suggestions or semantic search run, data is sent to the following overseas AI services.

🚫 Why it isn’t used for training (three parties)

① Tasukiba itself

• We never reuse your input for cross-tenant training or for building derived AI models
• The database is fully partitioned per tenant — only Company A’s people see Company A’s data

② Anthropic (Claude)

• Anthropic’s Commercial Terms of Service prohibit using customer data to train AI models
• “Not used for training” is guaranteed at the contract level

③ Voyage AI (text → vector for semantic search)

• Voyage AI’s default terms allow using data for training
• However, the Tasukiba operator has set the “do not use for training” opt-out, so data sent to Voyage AI is not used for training either

📊 Summary

🚨 Honest note: nobody can say “absolutely”

Theoretically, every service in the world carries some residual risk — AI-vendor hacks, future terms-of-service changes by AI vendors, account takeover via impersonation.

Tasukiba builds in measures to make those risks as small as possible:

• API keys are kept securely; never written into code or records
• Triple over-use blocking prevents abuse via mass-call flooding
• Tenant isolation is the top priority (treated as a hard fail if breached)
• Checks for other people’s personal information leaking into AI responses
• Automated checks that block merges of code with security scores below 90

📋 Each AI vendor’s official policies

If you want to verify yourself, or if your enterprise partners ask for the AI vendor’s terms (as of May 2026):

• Anthropic (Claude): anthropic.com — Commercial Terms of Service (explicitly excludes training) / Privacy Policy / FAQ “Is my data used to train your models?”
• Voyage AI: voyageai.com — Terms of Service (describes the opt-out) / Privacy Policy

Honestly: we’re currently operated by a single developer + Claude Code (AI), a small team. If we don’t reach monetization within two years of launch, we’ll stop onboarding new users, but the service itself is expected to continue because the developer uses it daily. If we ever have to stop providing the service, we will always set aside time for you to export your data first.

📊 About uptime and incident response

Given the small-team nature, we do not publish an explicit SLA (uptime guarantee) at this time. We’d rather be honest than commit to numbers we cannot back up.

That said, our operating principles are:

• Data protection is the top priority (automatic cloud DB backups + user-initiated export anytime)
• For serious incidents, we share the cause and recovery estimate on the Discord community
• Once we are operating at scale, we will formally set up SLAs and incident-response contacts

🤝 The developer is also a user

Tasukiba is a platform the developer uses daily for project operations. The fact that “the maker would suffer first if it went down” means the structural risk of abandonment or sudden shutdown is low.

Export is always possible. Regardless of plan limits, post-cancellation, or post-trial, you can download a full export from your tenant settings. We don’t yet have direct integrations with Backlog / Notion etc., but CSV import is partially supported for knowledge, risks, and issues — Excel and spreadsheet imports are fine.