Agent Composer – Build Custom AI Risk Analysis Agents
Build agents that know what they don’t know.
Agent Composer is the visual builder for Superconductor-powered workflows. Assemble judgment-capable agents from a library of pre-built skills, connect them to 41 MCP tools, and deploy with confidence gating and MentisDB audit trails included by default.
AGENT COMPOSER · UFLPA SCREENING WORKFLOW
Trigger: New supplier onboarding
webhook · TPRM system
Skill: Entity Resolution
resolve_entity · cross_jurisdiction_match
Skill: Supply Chain Trace
tier_n_traversal · uflpa_screen
Skill: Sanctions + BIS Check
screen_sanctions · check_bis_entity_list
Output: Risk report + MentisDB ID
confidence: 0.94 · sources: 7 records
THE BUILDER ARCHITECTURE
Four layers. Visual. Judgment-capable from the start.
Agent Composer organizes the agent-building experience into four distinct layers – from trigger to judgment to output to audit. Every layer is judgment-aware: your agents don’t just run, they know when they’re uncertain.
Connect to Your Systems
Start an agent workflow from a webhook, a scheduled scan, an API call, or a manual trigger from an analyst. Agent Composer connects to your TPRM, ERP, sanctions screening, or case management systems via standard REST webhooks.
- REST webhook triggers
- Scheduled scans (cron)
- Manual analyst triggers
- Event-driven from MentisDB alerts
Pre-Built Judgment Skills
Drag and drop from the Skill Registry – pre-built, tested workflow units that combine multiple MCP tools into a coherent judgment task. Each skill is a named, versioned unit with defined inputs, outputs, and confidence thresholds.
- 25+ pre-built skills in registry
- Versioned, testable skill units
- Custom skill creation for your workflows
- Confidence threshold configuration per skill
Confidence Gating Built In
Every skill outputs a confidence score. Judgment gates in the workflow canvas route to human review when confidence falls below your threshold – so your agents never quietly produce low-confidence outputs that reach a compliance decision.
- Configurable confidence thresholds per step
- Human review routing on low confidence
- Escalation path configuration
- SLA-aware routing (fast path / full fleet)
MentisDB – Automatic
Every workflow run is automatically recorded in MentisDB. You don’t configure audit logging – it’s built into the platform. Every step, every confidence score, every source record consulted, every human override: all SHA-256 signed and append-only.
- Automatic audit trail, no configuration
- SHA-256 signed per record
- Human override history included
- Exportable for regulatory submission
SKILL REGISTRY
Pre-built judgment tasks. Drag. Drop. Done.
25+ skills. Add yours.
The Skill Registry is the library of pre-built workflow units that ship with Agent Composer. Each skill is a named, versioned combination of MCP tools with defined inputs, outputs, confidence thresholds, and recommended human review triggers.
Skills are designed by Sayari’s tradecraft team against real analyst workflows – not generic API call sequences. They’re tested against EconSecHELM before release.
Custom skills. Build your own skills from MCP tool combinations and contribute them to your organization’s private skill registry. Version, test, and share them across your agent workflows – with full MentisDB provenance for every custom skill execution.
OWNERSHIP
UBO Deep Trace
Identifies ultimate beneficial owner through full multi-tier corporate structure. Handles VIEs, nominees, and trust structures. Returns ownership chain with confidence.
SCREENING
Network Sanctions Scan
Screens entity and connected parties for OFAC, UN, EU, and UK sanctions exposure. Surfaces second- and third-degree network connections.
SUPPLY CHAIN
UFLPA Tier-N Screen
Traces supplier network through customs data to identify Xinjiang-region connections at any tier. UFLPA compliance-grade output with source citations.
TRADE
Export Control Assessment
Classifies goods against CCL, screens end-users against BIS Entity List, and flags dual-use technology exposure with enforcement context.
KYC
CDD Package Builder
Assembles a complete Customer Due Diligence package: entity profile, ownership structure, adverse media, PEP screening, and risk score – with source citations for every claim.
MONITORING
Continuous Monitoring Alert
Scheduled re-evaluation of a counterparty against latest registry data, sanctions lists, and adverse media. Triggers alert on confidence change above threshold.
AGENT TEMPLATES
Start from a template. Deploy in hours.
Agent Composer ships with pre-built workflow templates for the most common economic security agent use cases. Customize for your specific workflow, connect to your systems, and deploy.
Supplier Onboarding Screen
Full entity resolution, beneficial ownership trace, sanctions check, and UFLPA screen – triggered automatically on new supplier records from your TPRM system.
- resolve_entity
- trace_ownership + find_ubo
- screen_sanctions + check_bis
- uflpa_screen
- risk_score → MentisDB
Customer Due Diligence
Complete CDD package assembly for financial institution KYC workflows. Entity profile, beneficial ownership, PEP screening, adverse media, and regulatory risk score.
- resolve_entity + get_entity_profile
- trace_ownership + find_ubo
- pep_screen + adverse_media_search
- risk_score + explain_score
- CDD package → MentisDB
Continuous Counterparty Watch
Scheduled re-evaluation of active counterparties. Detects ownership changes, new sanctions exposure, and adverse media on a configurable cadence.
- scheduled trigger (daily/weekly)
- re-resolve entity
- delta: ownership + sanctions + media
- confidence delta check
- alert if delta > threshold → MentisDB
Build agents that know when to ask.
Agent Composer early access is available for enterprise teams building on the Superconductor platform. Request access to start building judgment-capable workflows.