Summary. InboxOS is an AI-assisted email triage system for regulated client operations — primarily wealth-management and private-banking support desks today. It ingests Gmail, classifies intent via a nine-agent LangGraph pipeline, extracts structured fields, and routes tickets to autonomous or human-review queues. Staff approve every outbound email. Deployed single-tenant via Docker on customer infrastructure.
Who it is for
Internal bank staff (Relationship Managers, Support Professionals, CSAs, Escalation Specialists). Customers interact by email only.
Shipped today
Gmail IMAP ingest & SMTP send
9-agent LangGraph classification
Wealth Management industry pack
RBAC, JWT auth, append-only audit log
Docker single-tenant deployment
Human review before send
Roadmap (not shipped)
Outlook & additional connectors
SSO / enterprise identity
Multi-tenant SaaS
Additional industry packs
PostgreSQL option
Decision engine (default)
IF confidence ≥ 7 AND intent_risk == "low" → Autonomous queue · ELSE → Human review queue
Thresholds are configurable — not hardcoded. Autonomous means internal routing without desk handling; it does not mean unsupervised client replies.
Security & data (FAQ)
LLM processing: Email content is sent to your configured LLM provider API (OpenAI today) for classification and extraction. Results and agent logs are stored in your instance database.
Auto-reply: Not enabled in current release. Author agent drafts clarifications; humans send.
Audit: Agentic AI Timeline per email + append-only audit log for auth and sensitive actions.
Access control: JWT + role permissions on all data routes.
Deployment: Single organization per Docker instance — not multi-tenant SaaS today.
What AI does not do
Send client email without human approval
Replace compliance sign-off on high-risk intents
Operate as a generic inbox assistant without industry taxonomy