Overview#
An analyst is mid-investigation when she notices an unusual pattern in entity relationships, three shell companies with overlapping directorship and no apparent legitimate business purpose. She needs a quick risk assessment, a summary of what the platform knows about these entities, and a suggested next analytical step. Rather than navigating five separate modules, she types the question into the AI Assistant. It draws on case context, enriches the entities in real time, and returns a structured analysis with recommendations, all through a single interface.
The AI Assistant domain provides that unified interface for AI-powered operations across the platform. It routes requests to the appropriate AI provider, maintains investigation context, and delivers consistent responses regardless of which underlying model is processing the request.
Key Features#
- Conversational AI Interface: Natural language interaction for investigation queries, analysis requests, and data exploration.
- Data Enrichment: AI-powered enrichment of entities and data records with contextual intelligence and insights.
- Multi-Provider Support: Access to multiple AI providers through a single unified interface, with automatic failover if a provider is unavailable.
- Consistent API: Standardised request and response formats regardless of the underlying AI provider.
- Token and Cost Tracking: Centralised monitoring of AI usage with token counting and cost estimation across all operations.
- Investigation Context: AI interactions can be scoped to specific investigations for context-aware analysis and recommendations.
- Page Context Awareness: AI responses can incorporate context from the current page or workflow for more relevant assistance.
- Provider Fallback: Automatic failover to alternative AI providers ensures continuous availability of AI capabilities.
- Centralised Configuration: Model selection and parameters are managed centrally for consistent behaviour across the platform.
Use Cases#
Investigators use conversational AI to ask natural language questions about case data, receiving analysis and insights without needing to write complex queries or understand the underlying data structure.
Analysts enrich entity profiles with AI-generated summaries, risk assessments, and contextual intelligence from available data sources, adding analytical depth to POLE model records (Person, Organisation, Location, Object, Event) without manual research.
Investigation teams access AI assistance throughout workflows: hypothesis generation when a case opens, pattern identification during active analysis, and report drafting as cases approach closure.
Integration#
The AI Assistant domain connects to the platform's multi-provider AI orchestration system for model routing and fallback, the investigation domain for case-scoped analysis, and the usage tracking system for token and cost monitoring.
Open Standards#
- GraphQL (June 2018 specification): the entire AI Assistant query and mutation surface is exposed as a typed GraphQL API, allowing clients to request exactly the fields they need from a single endpoint.
- OpenAI Chat Completions message schema: all provider integrations (Anthropic, OpenAI, Cloudflare Workers AI, Gemini, Grok) exchange conversation turns as a
{role, content}message array, the cross-industry de facto format that enables seamless provider failover. - OAuth 2.0 Bearer Token (RFC 6750): every outbound call to an AI provider API authenticates with an
Authorization: Bearer <token>header, following the standard token usage pattern for HTTP APIs. - JSON (RFC 8259 / ECMA-404): all AI request payloads, response objects, structured extraction outputs, and usage export data are serialised as JSON, ensuring interoperability with any client.
- ISO 8601: all event timestamps, billing period boundaries, and export date ranges are encoded as UTC ISO 8601 strings (via Python
.isoformat()), giving consistent, timezone-unambiguous time references. - HTTPS / TLS (RFC 2818, RFC 8446): all network calls to external AI provider endpoints are made exclusively over HTTPS, enforcing transport-layer encryption for API keys and inference data in transit.
Last Reviewed: 2026-02-05 Last Updated: 2026-04-14