Introduction

Hardalion Nexus is the deterministic execution spine for autonomous AI agents in production.

Nexus is not an LLM. It is the gateway that sits between your agent runtime (LangChain, CrewAI, OpenAI Agents, custom MCP clients) and your enterprise APIs, databases, and payment rails. Every tool call is inspected, authorized, and logged before it reaches your backend.

Release: NSR 0.11.0 · Policy standard: NPL 1.0.0 (Apache 2.0)

What Nexus does

What Nexus is not

• Not a prompt wrapper or secondary LLM safety classifier
• Not a replacement for your model provider
• Not optional at the infrastructure layer when agents touch regulated data

Prompt instructions are not enforceable under audit. Nexus evaluates execution payloads deterministically: same input, same ALLOW / BLOCK / REQUIRE_HUMAN decision, stable rule IDs.

Request flow

flowchart LR
  A[LLM Agent] --> B[Nexus MCP Gateway]
  B --> C{NPL Evaluation}
  C -->|ALLOW| D[Enterprise API / DB]
  C -->|BLOCK| E[Audited rejection]
  C -->|REQUIRE_HUMAN| F[Approval queue]
  B --> G[Hash-chained audit log]

Product surfaces

Who this documentation serves

Developers and architects integrate agents in minutes, inspect latency, and pin policies from the open NPL registry.

CISOs and compliance officers map Nexus controls to EU AI Act and DORA, export conformity bundles, and verify tamper-evident audit chains.