MaiGuard Assist

Overview

MaiGuard Assist is MaiGuard's embedded assistant for fraud and compliance operations. It is purpose-built for financial workflows — not a general-purpose chatbot. Ask questions in plain language and receive structured answers grounded in your tenant's transaction data, case records, rule configuration, and platform signals.

MaiGuard Assist goes beyond answering questions. Analysts can use it to take direct enforcement actions — block a transaction, freeze an entity, escalate a case, or notify the compliance team — all within the same conversation flow. An intent classifier routes each request to the appropriate agent persona (fraud, AML, compliance, or executive), ensuring the right tools and context are applied automatically.

Transaction lookups

Query live and recent transaction data without writing SQL or API filters. Examples:

• "Show transactions with riskScore above 70 flagged in the last 48 hours"
• "List REVIEW decisions for customer cust_abc123 this week"
• "Find transfers over ₦500,000 from new devices today"

Responses include inline transaction cards with IDs in monospace, amounts with tabular numerals, and links to open the full record. MaiGuard Assist respects the same tenant scoping as the rest of the dashboard — your API key and role determine what data is visible.

Anomaly explanations

When a transaction is flagged, analysts often need the "why" behind the score. MaiGuard Assist breaks down contributing signals — velocity breaches, geo anomalies, behavioral biometrics, entity graph links, and ML feature contributions — in human-readable form.

• Explain which rules or models fired for a specific transaction ID
• Summarize feature importance bars from adaptive ML scoring
• Compare shadow-mode output against the live decision when shadow rules are active
• Investigate linked entities in the graph (shared device, IP, or beneficiary patterns)

Policy Q&A

Ask how your configured rules, thresholds, and workflows apply to a scenario. Useful for onboarding new analysts and aligning ops teams on enforcement policy.

• "What happens when riskScore is between 40 and 69?"
• "Which velocity rule applies to card-not-present transactions?"
• "When should I escalate to manual review vs block?"

On Scale and above plans, MaiGuard Assist uses vector-indexed policy documents to ground answers in your uploaded compliance materials — internal policies, regulatory guidelines, or onboarding playbooks. Answers reference your tenant's active rule set where possible. See Policy Documents for supported formats, document limits by plan, and how indexing works.

Actions

MaiGuard Assist can take enforcement actions directly from the conversation — no separate UI step required. Each action is confirmed with the analyst before execution and logged to the audit trail.

Analytics queries

Ask for live platform metrics in plain language. MaiGuard Assist queries ClickHouse — the same analytics backend powering the MaiGuard dashboards — and returns structured summaries:

• "What is our block rate for the last 7 days?"
• "Show me the top 5 rule IDs by triggered volume this month"
• "How many REVIEW decisions are pending right now?"
• "Compare fraud rates in LIVE vs TEST environments this week"

Analytics queries are available in both entity-scoped sessions and freeform mode. Metrics are returned as structured data with a plain-language narrative — suitable for quick executive briefings or ad-hoc operations reviews.

SAR drafting

Within a case session, MaiGuard Assist can generate a SAR narrative draft grounded in the case record, linked transactions, and entity data. The output surfaces as a structured SAR card with a FinCEN-aligned layout:

• Subject identity and account details from the case brief
• Activity period with total suspicious amount
• Narrative paragraph suitable for SAR Part V (Activity Description)
• Copy-to-clipboard for pasting into your filing system

Alert triage

The Assist command center includes an alert triage view — a live queue of REVIEW-status transactions waiting for analyst decision. From the triage view, analysts can open any queued transaction directly into an MaiGuard Assist session without leaving the page.

• Sorted by risk score descending — highest-priority items surface first
• Real-time count reflects current REVIEW + PENDING queue depth
• One-click to open a transaction assist session with full context pre-loaded
• Block or escalate from within the conversation — queue depth updates immediately

Access the triage view at /assist/alerts or from the MaiGuard Assist sidebar icon in the dashboard navigation.

Freeform mode

Freeform mode lets analysts ask questions without starting from a specific transaction or case. Use it for cross-cutting queries, operational reviews, and compliance questions that span the whole platform:

• "What were our top fraud patterns last week?"
• "Summarise today's block rate and flag any anomalies"
• "When is our next SAR filing deadline?"
• "Which compliance documents cover high-value cash transactions?"

Freeform sessions are persisted separately from transaction and case sessions. Session history is available for replay from the MaiGuard Assist sidebar. The intent classifier automatically routes freeform queries to the most appropriate agent persona — compliance, executive, or fraud — without requiring analyst selection.

Drawer vs full-page

Role-aware prompt chips appear in the empty state (for example, high-risk queue, explain last BLOCK, shadow rule comparison). Chips disappear after the first message is sent. On mobile (below 768px), the drawer navigates to the full-page view automatically.

Data & safety rules

• Transaction IDs render in monospace with primary styling — same as the rest of the UI
• Monetary values use tabular numerals and standard currency formatting
• Raw JSON, API error codes, and stack traces are never shown in chat output
• A mandatory disclaimer appears below the input: verify critical financial data before acting
• Send is disabled while a response is streaming; use stop to cancel long-running queries
• All sessions are tenant-isolated — no data crosses tenant boundaries
• Action tool calls (block, freeze, escalate, notify) are logged to the platform audit trail with analyst identity and timestamp

MCP & agents

For programmatic and agent integrations outside the dashboard, MaiGuard publishes machine-readable product context and runs an MCP server at mcp.maiguard.com/v1. External agents can call the score_transaction tool directly and fetch structured schemas — complementary to the in-app MaiGuard Assist experience.

• →Product context JSON
• →Agent scoring schema
• →Developers hub (MCP package)
• →llms.txt index