One governed platform for public-sector AI operations.
Every module and channel runs on the same identity, multi-tenant isolation, audit, and model-governance layer. Add a real model provider per deployment; everything else — accountability, access control, data handling — stays consistent.
How a request flows
1 · Authenticate
Bearer-token session; the tenant and roles come from the token, never the request body.
2 · Retrieve
Only approved, active, tenant-scoped sources are retrieved and passed as grounding — never as instructions.
3 · Answer or escalate
The governed gateway answers with citations, or fails closed to a human on low confidence or a disallowed model.
4 · Audit
Every step is written to an append-only, tenant-scoped log for later review and export.
The shared platform layer
Identity & access
Password sign-in → signed bearer token → request context with roles. Idle + absolute session timeouts, admin re-authentication, login lockout. MFA-ready schema.
Tenant & department isolation
Every tenant-scoped table carries a tenant id; the data layer always filters by it and re-checks each returned row.
Append-only audit
User, admin, and agent actions are recorded and only ever inserted — tamper-evident at the application layer.
Model governance
A central gateway enforces an allow-list, circuit-breaker, timeouts, no-training default, and per-agency spend caps — fail-closed to human escalation.
Provider-neutral data
An async database interface runs on SQLite for local work and PostgreSQL in hosted environments — the same code, no lock-in.
Accessibility & i18n
Semantic, keyboard-operable interfaces with visible focus; English and French copy (French flagged for human validation).
One assistant, every channel
Consistent governance across web, mobile, and phone.
The same grounded, audited assistant answers on your website, in a mobile app, and on an inbound AI phone line — each opening with a clear AI disclosure and escalating to a person on request.
The five modules
| Module | What it does |
|---|---|
| Staff Assistant | Grounded answers with citations, drafting from templates, summarization, document comparison, human escalation. |
| Citizen Assistant | Approved-source public answers, department routing, human handoff, emergency & prohibited-topic screening. |
| Knowledge | Document ingestion, grounded retrieval, supersede/expire lifecycle so retired content never grounds an answer. |
| Document & Case Intake | Extraction, completeness validation, PII/SSN redaction, case creation, queue routing, human-approval gate. |
| Workflow & Governance | Approvals, audit review, retention, tenant/department/user admin, assistant versioning, export & deletion. |
Procurement-ready evaluation
See the platform with your scenarios.
We can seed a fictional evaluation tenant that mirrors your permitting, records, or 311 workflows.