Basilisk

Basilisk is an AI-native intelligence platform that autonomously investigates entities across public and private data sources, builds a temporal knowledge graph with full provenance, and generates analyst-grade intelligence. It reasons like a human investigator - but at machine speed, across every available source, without forgetting.

The tools available to investigators have not fundamentally changed in a decade. Palantir evolved, but priced out the market. Everyone else is still doing enhanced Google searches.

The Intelligence Gap

The intelligence industry is bifurcated. At the top: enterprise platforms with eight-figure contracts and six-month deployments. Powerful, but inaccessible to anyone outside the Fortune 500. At the bottom: browser tabs. An analyst copy-pasting between corporate registries, sanctions lists, blockchain explorers, and search engines. Manual correlation. Manual verification. Manual everything.

Nothing exists in between. No AI-native platform that autonomously investigates, builds persistent knowledge, handles temporal complexity, overlays private data onto public intelligence, and is accessible to a five-person compliance team - not just a five-thousand-person agency.

The consequences of this gap are real. Danske Bank processed $230B in suspicious transactions. Wirecard hid $2.1B in fabricated revenue for years. FTX collapsed while the data that would have revealed its insolvency existed across dozens of public sources - no system could assemble the picture fast enough.

Three Differentiators

Temporal Knowledge Graph

Every fact in Basilisk's graph is versioned. Every relationship carries its history and provenance. The graph captures not just what is true now, but what was true at any point in time - and when the system learned it. This temporal dimension turns the graph into a historical record that can be queried at any point, revealing patterns that current-state snapshots miss entirely.

• Point-in-time queries - What did this corporate network look like on a specific date?
• Delta queries- What changed in this entity's ownership structure between two dates?
• Provenance tracking - Where did this fact come from? How confident should we be? What was the extraction method?

Autonomous Investigation Engine

Basilisk doesn't wait for queries. Given a target, it investigates autonomously - reasoning through an eight-phase cognitive loop that mirrors how the best human analysts work:

This is a loop, not a pipeline. Each pass deepens understanding. The investigation continues until confidence is reached, the budget is exhausted, or diminishing returns set in. Competing hypotheses are maintained simultaneously - the engine doesn't tunnel-vision on the first theory.

Open Intelligence + Private Data Overlay

Public sources - corporate registries, sanctions lists, court records, news, blockchain explorers - form the foundation layer. This open intelligence graph is available to all users. Private and proprietary data layers on top: encrypted, tenant-isolated, entity-resolved against the open graph. Provenance scopes never mix - the system always knows where a fact came from and who can see it.

Architecture

Basilisk is built as a four-layer stack. Each layer is independently valuable. Together, they create compound intelligence.

• Data Ingestion - Connectors to public registries, sanctions lists, court records, news sources, blockchain explorers, and more. Private data plugs in through file upload and database connections.
• Knowledge Graph- Temporal, entity-resolved, provenance-tracked. Cross-source entity resolution ensures the same real-world entity isn't fragmented across the graph.
• Reasoning Engine - Hypothesis generation, evidence evaluation, pattern detection, anomaly analysis, and narrative generation. Produces structured dossiers with full evidence chains.
• Interfaces - REST API for programmatic access. Natural language queries for ad-hoc investigation. A visual graph explorer for navigating entity networks and timelines.

Entity Resolution

Entity resolution is the core quality gate. The same person appears as “John A. Smith, Director of Acme Holdings Ltd (BVI)” in one registry and “J. Smith” in a Reuters article. The same company is registered under variant names across jurisdictions. Adversarial actors deliberately obfuscate - similar names, layered ownership, family proxies.

Basilisk uses a multi-stage resolution pipeline: probabilistic blocking and scoring to narrow candidates, then AI-powered disambiguation for ambiguous cases. The pipeline maintains a full audit trail - every merge and split decision is recorded with justification. False merges create phantom connections that poison investigations. False non-merges leave the graph fragmented. Quality here is everything.

Why Now

Three tailwinds make this possible today:

• LLMs enable autonomous investigation- Two years ago, an AI that could read a corporate filing, extract directorships, reason about entity identity, generate hypotheses, and explain its reasoning with citations was science fiction. Now it's engineering.
• Regulatory pressure creates urgent demand - AMLA (EU), MiCA, FinCEN CTA, FATF Travel Rule expansion. Every financial institution, every crypto company, every law firm doing cross-border work faces increasing compliance obligations with increasing penalties.
• OSINT has gone mainstream- Bellingcat, open-source flight tracking, blockchain forensics. The public data surface has expanded dramatically, but the tools to navigate it autonomously haven't kept pace.

Positioning

Roadmap