Descalate reads the ticket on your screen, redacts personal information in your browser before sending anything to our servers, and returns an analysis plus a suggested reply. We never store the ticket text. We store only the signals we extracted, so we can show you your usage and improve the product. Our analysis provider, Fireworks AI, does not train on our inputs or outputs. The terms are the same on Free and Pro.
1. What data we receive
The extension reads the text of the support ticket you have open. Before anything is sent to our backend, it removes personally identifiable information using pattern-based redaction. The Transparency page lists every category in detail. They currently include:
- Names (in greetings, closings, self-introductions, parenthetical transliterations and CJK scripts)
- Email addresses
- Phone numbers (US, international and masked variants like “+81 90 XXXX 1842”)
- Credit card numbers and “ending in XXXX” phrasing
- SSNs, passport numbers, driver’s license numbers
- Dates of birth
- Customer IDs, order, invoice and case numbers, reference IDs
- Usernames and logins
- Addresses, including Japanese block-style addresses and apartment numbers
- Postal codes (US, Canada, UK, Japan)
- Labeled fields like “Mobile — X”, “DOB — X”, “Address — X”
Redaction is pattern-based, so we call it best-effort: language is messy and we will miss novel formats. If you spot one we don’t catch, email us and we’ll add it. We treat redaction misses as bugs.
What reaches our backend is the redacted ticket text, the cultural signals detected in your browser (an inference like “likely Japan”, never the name it came from), and your account identifier.
2. Who processes your data
We publish the current list of sub-processors, and the date we last reviewed each one, on Transparency. We give at least 30 days’ notice, by email and a dated changelog entry, before adding or changing one.
Fireworks AI
Hosts the model that produces your analysis (MiniMax M3.1 Flash). Based in the United States. Per their published terms, Fireworks does not train on API inputs or outputs, and does not keep prompts or responses beyond operational logs for abuse prevention and reliability.
Supabase (database and auth)
Stores your account record (email, plan, signup date), your usage counter and signal metadata, never ticket text. Hosted in the United States. Sign-in codes are delivered through Resend.
Vercel (hosting) and Resend (transactional email) are also listed on Transparency. Neither keeps ticket text: Vercel processes it in memory for one API call and never writes it to disk, and Resend only handles account email.
3. What we do not do
4. What we do store
Each analysis writes one row of signals to our database. It contains only:
- Your user ID
- Timestamp of the analysis
- Platform (for example, Zendesk)
- Sentiment class (calm / concerned / frustrated / angry)
- Escalation signal IDs detected (for example,
cancellation_threat) - Overall escalation risk (low / medium / high)
- Recommended response-time band (15 min / 1 h / 4 h / 24 h / 48 h)
- The customer’s inferred cultural region, if any
- Later, opt-in only: whether you copied, modified or ignored the suggested reply
This record isn’t reversible. From the signals alone, no one can reconstruct what the customer wrote, who they were, or what the transaction was.
Retention. Signals are kept for 24 months or until you delete your account, whichever comes first. Operational logs are kept for 30 days.
5. How the analysis flow works
- You click Analyze in the Descalate sidebar on a ticket.
- The extension extracts the ticket text locally and runs the redaction patterns.
- The redacted text and cultural signals are posted over HTTPS to our API with your access token.
- Our API forwards them to Fireworks AI, which runs MiniMax M3.1 Flash and returns a structured analysis.
- Our API writes the signal row (section 4), returns the analysis to your browser and discards the ticket text.
- The sidebar shows the analysis. You decide whether and how to reply.
Round trip is typically 3–5 seconds. Ticket text lives in our backend’s memory for step 4 and no longer. Walk through it step by step →
6. Your rights and controls
For data requests, email privacy@descalate.app. We aim to reply within 7 days.
7. GDPR, UK GDPR, CCPA
We rely on legitimate interest to process ticket content for an analysis you’ve explicitly requested, and on contract and consent for account creation and billing.
EU and UK users: your data is processed in the United States. Our sub-processors maintain Standard Contractual Clauses for transfers. You can exercise your rights of access, rectification, erasure, restriction, portability and objection at privacy@descalate.app.
California users: you have the right to know what we hold, ask for its deletion, and not be discriminated against for doing so. Descalate does not sell personal information.
8. Security
All data in transit is TLS-encrypted. Our backend runs on Vercel’s serverless platform. Our database is managed Supabase Postgres with row-level security, so users can only read their own signals. API access is gated by signed JWTs issued at sign-in.
We’re a small team and honest about it. We don’t currently hold SOC 2 or ISO 27001. If your employer requires them for vendors, Descalate isn’t the right tool for you today. We expect to pursue SOC 2 Type I within 12 months of launch.
9. Children
Descalate is a professional tool for support agents, not intended for anyone under 18. We do not knowingly collect data from children.
10. Changes to this policy
If we materially change how we handle your data (a new sub-processor, different retention, a change to the signal taxonomy) we’ll email you at least 30 days before it takes effect and post a dated entry on the Transparency changelog. Minor edits are dated at the top of this page.