Shadow Warden AI Analytics Hub AI Analytics Hub
One API. Nine semantic models. Every dashboard, every AI agent, every external BI tool queries through a single deterministic SQL interface — so changing a metric definition updates every surface instantly.
How It Works
LLM (Claude Haiku) translates natural language → QueryObject → deterministic SQL — no hallucinated numbers, no raw DB access exposed.
9 Built-in Models
All models available immediately on Pro+. Expand any model with custom metrics via the Self-Service Catalog.
Filter Events
filter_events
Block rate, P99 latency, verdict distribution across the 15-layer pipeline.
Entity Risk Scores
ers_scores
Session risk scores, shadow-ban events, high-risk session count.
Security Incidents
incidents
STIX-linked incident journal — severity, status, avg resolution time.
Billing & Quota
billing_usage
Requests consumed, cost, quota utilisation, overage charges.
AI Cost Allocation
ai_spend
Per-department AI spend vs budget — the source of truth for Budget Guardian.
Agentic Commerce Orders
agentic_orders
Purchases by AI agents: spend by merchant, protocol (UCP/AP2/MCP), status.
Vendor Contracts & DPA
vendor_contracts
DPA expiry countdown, risk tiers, active vs pending vendor contracts.
Compliance & Training
compliance_attestations
Employee AI training completion %, overdue renewals by department.
Sovereign Tunnel Sessions
tunnel_sessions
MASQUE tunnel compliance rate per jurisdiction — EU, US, UK, CA and more.
Three Pillars
Commerce Budget Guardian
- ✓ Reads actual MTD spend from ai_spend model
- ✓ Per-transaction limit enforcement
- ✓ Monthly budget ceiling check
- ✓ Approval gate for large purchases
- ✓ Slack alert when budget exceeded
GET /business-community/commerce/budget Self-Service Catalog
- ✓ Register custom models via API or Streamlit
- ✓ Persisted to SQLite, hot-loaded instantly
- ✓ Access rules enforce tenant isolation
- ✓ OSI 1.0 export/import for external BI
- ✓ SOVA can discover and query any model
POST /semantic-layer/models/catalog SOVA AI Tools (74 total)
- ✓ semantic_query() — NL or structured QueryObject
- ✓ list_semantic_models() — discover available models
- ✓ check_commerce_budget() — pre-flight payment check
- ✓ get_spend_summary() — MTD spend report
- ✓ #70-74: write_handoff_memory, read_handoff_memory, semantic_listing_search, get_protocol_schema, send_order_proposal
Via SOVA agent tool dispatch Query Example
Structured API call
POST /semantic-layer/query
{
"model_id": "ai_spend",
"metrics": ["total_cost_usd"],
"dimensions": ["department"],
"filters": [{
"dimension": "tenant_id",
"operator": "=",
"value": "your-tenant"
}],
"limit": 100
} Generated SQL (deterministic)
SELECT
department AS department,
SUM(amount_usd) AS total_cost_usd
FROM cost_allocation_entries
WHERE (tenant_id = %s)
GROUP BY department
LIMIT 100 The same query works for SOVA agents, SOC Dashboard, Streamlit, external BI — and results are Redis-cached for 10 minutes. SQL never reaches the LLM; only metric names and descriptions are in the prompt.
-- Month-to-date seller net earnings
SELECT
seller_agent_id,
SUM(gmv_usd) AS gross_revenue,
SUM(platform_fee_usd) AS take_rate_paid,
SUM(seller_net_usd) AS net_profit
FROM agentic_orders
WHERE cleared_at >= date_trunc('month', now())
GROUP BY seller_agent_id
ORDER BY net_profit DESC;Ready to centralise your analytics?
AI Analytics Hub is available on the Pro plan ($69/mo). Register custom models, query with natural language, and share OSI models with your community — all from a single API.