Nuix launches the Neo AI Agent: agentic investigations with an audit trail built in

• Who: Nuix (ASX: NXL), a Sydney-based investigative analytics and intelligence software company; CEO John Ruthven.
• What: the Nuix Neo AI Agent — agentic AI on the Nuix Neo platform, built on the Nuix Neo Gen AI Framework, which answers plain-language questions over case data.
• When: launched Wednesday, October 21, 2026 (announced via PR Newswire from Sydney).
• Capabilities: reads documents, images and other items; follows connections between people, accounts, communications and transactions; draws on 20+ tools; auto-generates knowledge graph queries; every answer linked to its source.
• Controls: administrators can cap spend, track usage, and set guardrails on prompts and responses; role-based access controls enforced; the full conversation with steps and findings is exportable.
• Model layer: connects the customer's chosen supported LLM to Nuix's curated, forensic-grade data intelligence layer — outputs grounded in source material, meant to accelerate, not replace, professional judgement.

Investigative analytics company Nuix on Wednesday launched the Nuix Neo AI Agent, an agentic AI built for the high-stakes legal, regulatory and criminal matters its software serves. The pitch is aimed squarely at the trust problem dogging AI agents: the agent answers plain-language questions over case data, links every answer back to its source document, and records each step it takes — so investigators can show courts and boards exactly how a result was reached. (Nuix; Morningstar)

An agent that reads the case file

Built on the Nuix Neo Gen AI Framework, the agent responds to natural-language queries, reads and sifts through documents, images and other items, and finds and follows connections between people, accounts, communications and transactions — work that today consumes hours of manual review. It draws on more than 20 tools inside the Nuix Neo platform, operates across entire case inventories with role-based access controls enforced, and integrates with the Nuix knowledge graph, where it can automatically generate queries — cutting the need for investigators to hand-write graph queries. Once an answer is produced, the full conversation, including the steps taken and the findings, can be exported.

The audit trail is the product

In regulated work, an agent's output is only as useful as its provenance. Nuix's launch message makes the audit trail the headline feature rather than an afterthought: trace every answer to its source, review every step. “For more than two decades, organisations have trusted Nuix with their high-stakes matters. Agentic AI should raise that bar, so we built the Nuix Neo AI Agent to show its work. Trace every answer to its source. Review every step. Rely on forensic-grade data processed by the Nuix Neo platform,” said John Ruthven, CEO of Nuix. The company's positioning is that the agent's outputs are grounded in source material and designed to be verified against it — supporting and accelerating, not replacing, the judgement of qualified professionals.

Governed, and model-agnostic by design

Under the hood, the Neo Gen AI Framework is an enterprise-grade infrastructure and governed orchestration layer that connects a customer's chosen supported LLM to the curated data intelligence layer delivered by Nuix's patented data processing engine — a bring-your-own-model design that lets regulated customers keep their existing model approvals. Administrators can cap spend, track usage, and set guardrails on prompts and responses, which matters when agent usage bills against tokens and every query touches sensitive evidence.

Why it matters

Nuix (ASX: NXL), headquartered in Sydney, sells to organisations whose decisions face the highest scrutiny — government agencies, law firms, financial institutions, corporations and regulators. In those rooms, “the AI said so” is not an acceptable provenance, which is why the launch reads as a template for how agents will have to ship into regulated industries: source-grounded answers, step-level audit trails, spend and prompt guardrails, and human judgement explicitly kept in the loop. The question for buyers will be how the agent performs under real adversarial casework — messy evidence, conflicting documents, incomplete records — where an audit trail proves what the agent did, but not whether it missed what mattered.