Operations and support document (not an SLA) — SiCusto AI Customer Support

1Guarantees and targets at a glance

Item	Category	Details
Prepaid spending cap	Guarantee	We do not bill beyond the amount topped up. If the monthly allowance has been exhausted and the prepaid balance is also insufficient to start a new support session, we accept an inquiry for a human contact instead of starting a new AI support session.
Billing for failed turns	Guarantee	Turns where AI response generation fails are not billed.
Data location	Guarantee	Storage is in the Seoul region, Republic of Korea. Data for response generation is sent only to the AI providers in the register (D-2).
Retention period	Guarantee	Automatic deletion after the periods in Annex 2 to the Data Processing Agreement.
Incident notification	Guarantee	We notify your company without undue delay upon becoming aware of a personal data incident (D-8).
Support hours	Guarantee	Requests accepted weekdays 10:00~18:00 (KST) · support@alienpeoples.com
Initial response time	Target	Within 4 hours for requests received during business hours · Requests at night or on weekends: next business day. For emergencies (entire support page unavailable), use [Urgent] in the subject.
Initial incident notification	Target	Within 24 hours of awareness (including an initial notice of matters still being investigated).
Incorrect response correction	Target	Knowledge and rules updated within 1 business day from receipt of an incorrect-response report.
Recovery point objective (RPO)	Target	Restore to a point within the preceding 1 hour.
Recovery time objective (RTO)	Target · Not yet measured	A value will be entered after a recovery drill. No numerical commitment is made before then.
Uptime (%)	No commitment	Separate agreement after actual measurement, if needed.

2Support

Contact — support@alienpeoples.com (weekdays 10~18 KST). For emergencies, use [Urgent] in the subject.
Operations lead — the representative handles this directly. AI responses, handoff emails and alerts run automatically even when the representative is absent.
Incorrect-response reports — use the “Report incorrect response” button on the admin page or email us. We collect and process them once each morning.

3Behavior during outages

Situation	What the support page does
Primary AI provider outage or delay	Automatically switches to another provider in the register (currently 3 providers · 2 fallback stages). Repeated consecutive failures temporarily block the affected provider while recovery is tested.
All three providers fail	Shows “Please try again shortly” without billing that turn · The human handoff form remains available at all times.
Monthly allowance exhausted · Insufficient prepaid balance to start a new support session	Accepts an inquiry for a human contact instead of a new AI support session (forwarded to the contact’s email).
Daily cap reached (new sessions · Cost)	Blocks new AI conversations for that day and accepts inquiries only · Alerts the operator.
Email delivery failure	Email queue retries automatically (up to 5 times) · Alerts the operator if a backlog develops.

4Alert system

A monitoring job running every 5 minutes checks the following. The same incident is notified only once during the cooldown period.

Alert	Condition	Recipient	Cooldown
Prepaid balance	50% · 80% · 95% of the latest top-up used (once per threshold)	Your company’s contact + Operator	Once per threshold
Daily cap reached	Tenant’s daily cap for new sessions or cost	Operator	12 hours
Response failure rate	Failure rate of at least 10% over the previous 1 hour (with at least 20 completed requests)	Operator	1 hour
Email queue backlog	At least 50 pending emails or oldest email waiting at least 10 minutes	Operator	1 hour
Incorrect-response reports	If at least 1 report was received the previous day (KST), combined into one morning email (09~11)	Operator	Once per day
Provider circuit open · Total cost	B2B circuit open · Total daily B2B cost at least 80% of the cap	Operator	Once per day

5Backup and recovery

Backups — Cloud Firestore point-in-time recovery (PITR) + daily backups retained for 7 days. Backups automatically expire when their retention period ends.
Recovery procedure — the production DB is not overwritten.
1. Restore the backup (or PITR point) to an isolated, separate DB
2. Select only B2B data (the relevant client’s paths) from the isolated DB
3. Reapply the deletion history since the backup point (so conversations requested for deletion are not revived by recovery)
4. After moving to the production DB, check that the automatic expiry (TTL) policy also applies to those documents
5. Notify your company of the recovery scope and point in time
Once recovery drill results (elapsed time) are available, enter them in the RTO field in §1.

6Deletion procedures and deadlines

Request	Who · How	Deadline
Delete one conversation	Your company directly uses the session deletion tool on the admin page (the session can no longer continue from that moment)	Immediately
Delete records of a specific end customer	End customer → Your company → Request to us (we are the processor) · Operator locates and deletes the records	Without undue delay after verifying the request · Report the result
Contract termination	Deliver export files (JSON · CSV) → Delete production originals → Notify of the result	Article 12 of the Data Processing Agreement
Backups · Provider-retained copies · Delivered emails	Backups expire after 7 days · Provider policies apply · Your company deletes emails in its own mailbox	Annex 2 to the Data Processing Agreement

7Audit records — what we record

Admin page: viewing full conversations · Downloading support CSV and ledger CSV files · Changing settings · Changing handoff status · Deleting sessions · Inviting and removing members · Applying and reverting knowledge updates.
Recorded values: who (member) · When · What (action type) · Target id · Count. Conversation bodies are not duplicated. Records are written before returning the result.
Our operators’ actions (viewing full conversations · Exporting · Deleting clients) are recorded in the operational audit log.
Retained for 400 days · Your company can view and download these under “Activity log” on the admin page.

D-8Personal data incident response and notification 【Legal review】

Stage	Action	When
① Intake	Channels: support@alienpeoples.com, subject [Security] · Alerts and monitoring · Provider notices · Third-party reports. The receipt time is recorded.	At all times
② Initial action	Block the source path (suspend the affected tenant · Rotate keys · Revoke access) · Preserve evidence (retain audit records, logs and the list of affected documents separately; do not overwrite)	Immediately upon awareness
③ Initial notification (processor → Your company)	Circumstances · Affected data and counts (including estimates) · Actions taken · What your company needs to do · Contact point. Unknown matters are marked “Under investigation”.	Target Within 24 hours of awareness
④ Determine the impact scope	Use audit records to establish which clients, periods and data are affected	As information is confirmed
⑤ Follow-up report	Confirmed scope · Prevention of recurrence · Draft notice to data subjects	Without undue delay after confirmation
⑥ Closure	Report completion of actions · Retain records	—

Distinguish statutory duties from contractual commitments — notifying data subjects and reporting to Korea’s Personal Information Protection Commission and specialized agencies (Article 34 of the Personal Information Protection Act; statutory requirements and deadlines apply) are your company’s statutory duties as the personal data controller. Our notification above is a contractual commitment, and we provide the materials needed for your company’s reports and notifications.
Backup procedure when the representative is absent — 【Representative to enter: alternate contact name · Contact details · Authority (tenant suspension · Key rotation) · Time threshold if unreachable】
If we receive an incident notice from an AI provider or cloud provider, we forward it to your company using the same procedure.