Approved source control
Customers approve what the agent can use before publish, and noisy content like blogs can be excluded by default.
Atithya AIPortal previewSecurity
Source approval, domain controls, and instant pause from day one.
Atithya should build trust without overclaiming compliance. V1 focuses on practical controls: approved sources, tenant separation, server-side keys, rate limits, usage caps, retention, deletion, incidents, and pause.
P0pause-first incident posture
Agent live
Domain allowlist
Instant pause
Incident trail
Data deletion
What matters
Trust language must be accurate.
Atithya can say it is designed with controls. It must not claim HIPAA, SOC 2, ISO, or GDPR compliance until those programs are implemented and reviewed.
Domain allowlist
Free supports one customer domain, Pro two, Plus five, and Custom as agreed. Demo and localhost are controlled separately.
API key protection
AI keys stay server-side. Browser code should never expose secrets or private routing logic.
Incident response
Pause the agent, preserve safe logs, diagnose the issue, fix configuration or knowledge, rerun QA, and republish.
Workflow
V1 controls
Security is framed as practical operating control, not big-company theater.
- Source approval
- Domain allowlist
- Server-side keys
- Tenant metering
- Rate limiting
- Usage caps
- Retention
- Deletion
Practical trust controls
Security is a customer-visible operating system, not a promise hidden in a footer.
Customers approve the pages and documents the agent can use before publish.
Live agents answer only on approved customer, preview, and development surfaces.
Keys stay behind the gateway. Browser code never receives secrets or private routing logic.
Agent knowledge, leads, transcripts, and settings are scoped to the owning customer.
Request controls reduce abuse without making genuine visitors feel blocked.
Answer allowance and alerts keep spend predictable while lead capture remains available.
Out-of-scope prompts, unknown domains, and repeated suspicious activity are handled defensively.
The customer can pause an agent immediately while keeping a visitor-friendly handoff path.
Domain allowlist
Plans define where the agent is allowed to answer.
Free1 customer domain
Pro2 customer domains
Plus5 customer domains
Customcustom domains
- demo environment allowed
- localhost only in development
- unknown domains blocked
- per-domain usage metering
Incident response
Severity is defined before something goes wrong.
P0unsafe answer, data leak, API abuse, wrong customer data
P1important hallucination, lead capture broken
P2UI issue or minor incorrect answer
P3copy/design issue
Data handling
Conversation retention and deletion need a visible workflow.
Conversation retention
Keep transcripts for analytics and quality review only as long as the customer policy allows.
Data deletion workflow
Customer requests should remove leads, transcripts, source snapshots, and agent settings from the active workspace.
Response flow
Every incident should create a diagnosis trail.
- pause agent if needed
- notify Sherry immediately
- preserve safe logs
- diagnose cause
- fix knowledge, guardrail, or config
- re-run QA
- republish
- send customer note if needed
Product principle
What Atithya does not claim yet
No formal HIPAA, SOC 2, ISO, or GDPR claims until the product, process, legal terms, and audits support those statements.