ValarIQ — technical overview

Architecture

Agent runtime (REST · SDK · MCP)
    │  tool call + context (amount, counterparty, …)
    ▼
ValarIQ  ── verify identity · evaluate authority · apply policy
    │
    ├─ ALLOW  ──► payment / core systems
    ├─ DENY   ──► blocked + audit
    └─ ESCALATE ──► human approval ──► ALLOW | DENY
                          │
                          └─► decision evidence object + audit trail

Models and ledgers stay yours. Integration via HTTPS APIs and @valariq/gateway (wrapTool).

Policy engine (shipped)

Refund policy (REFUND-POLICY-001): evaluate context.amount_gbp against workspace thresholds → ALLOW / ESCALATE / DENY. High-risk scopes without policy rules default to ESCALATE.

• Enforce mode — decisions block or hold production
• Observe mode — log would-be ALLOW/DENY/ESCALATE, allow through (for pilot modelling)
• Kill switch — workspace-wide agent freeze (fail closed)

Core objects

• WorkspaceTenant + policy settings + kill switch
• Agent certificateIdentity + fingerprint + scopes; revocable
• Policy decisionALLOW | DENY | ESCALATE (+ policy_id, reason)
• ApprovalPending ESCALATE outcome; 30-min TTL
• Evidence packSigned decision evidence object

Control API (summary)

• POST /api/v1/verify — fingerprint → active agent?
• POST /api/v1/decisions — Financial Action → ALLOW | ESCALATE | DENY (200/202)
• PATCH /api/workspace/policy — thresholds, observe/enforce, kill switch
• UI / API approvals — resolve ESCALATE queue

Full reference: https://valariq.com/docs

Integration pattern

1. Create workspace; configure refund policy; start in observe mode.
2. Issue agent with tools:refund; create API key.
3. Wrap payment tools with gateway — every call hits /v1/decisions first.
4. Review observe logs; switch to enforce when Risk approves.
5. Export decision evidence packs for pilot success criteria.

Roadmap (design partner)

• Richer Agent Authority Profiles (counterparty, transaction age)
• Multi-level approvals / segregation of duties
• SIEM webhooks · SSO · payment rail connectors

Security posture: /security

Pilot technical success

Refund path wired end-to-end; observe → enforce transition demonstrated; ALLOW/ESCALATE/DENY decisions logged; at least one approved ESCALATE with printable evidence pack; kill switch demonstrated.