ARGORIX LANG
An open-source language built in Rust for secure, traceable and verifiable AI-agent communication.
Argorix Lang explores structured execution for AI-agent systems: tools, providers, guards, traces, bytecode and runtime control — designed for systems that need to be inspected, tested and governed.
Three papers, one runtime
Each view animates one result from the ArgorixLang papers: the compilation path, the fail-closed provider boundary and the offline-verifiable EvidenceBundle.
Solid paths denote implemented compilation and runtime behaviour. Metadata layers do not imply external verification, live federation, legal compliance or real-world identity proof.
Why Argorix Lang?
AI-agent systems need more than prompts, wrappers and scattered permissions.
Most AI-agent systems today are built on fragile layers of prompts, tools, wrappers and provider-specific logic. That can work for prototypes, but it becomes difficult to audit, verify and govern when systems become more autonomous. Argorix Lang explores a different path: structured, auditable and programmable execution for AI-agent systems.
Secure by Design
Execution rules, guards and controlled flows are treated as first-class design concerns.
Traceable Execution
Every meaningful agent action should be inspectable, logged and reviewable.
Verifiable Behavior
Move beyond prompt-only control toward structured semantics and testable execution.
Provider Boundaries
Separate tool and model execution through explicit provider interfaces and permissioned calls.
Five separations that define the claim
These are not caveats appended at the end — they are the method. Watch each pair come apart.
From Language to Controlled Execution
A compilation pipeline designed for auditable AI-agent systems.
Source syntax enters the language pipeline.
Structured representation of programs.
Intermediate form for semantic checks.
Compact execution format.
Controlled runtime execution.
Inspectable execution evidence.
parser.ok → ast.valid → ir.checked → bytecode.emitted → vm.running → traces.recorded
Argorix Lang is designed around a compilation pipeline that can evolve toward sandboxed providers, deterministic execution, semantic checks and auditable agent communication.
One request, compiled and checked
Agent, passport, provider contract, policy and evidence requirements travel through one validated pipeline — and the request stops at the provider boundary.
VM-level denial is a control-flow property of the implemented path. It does not prove OS sandboxing, container isolation, host network containment or absence of side effects.
A verdict has a type, not a colour
Unknown rules and malformed policy objects are their own outcomes instead of collapsing into ordinary violations. Distribution over 57 deterministic controlled cases.
Every complete v0.1 session records policy_passed=false and review_required=true, with 1,188 detailed findings encoded as “unknown policy rule”. A bundle can verify while policy fails: integrity and approval are orthogonal axes.
Sovereign metadata, compiled
31 recognised country codes produce same-country PASS outcomes; EU is recorded only as a regional residency zone, never as a country. The evaluated sessions declare country and jurisdiction CL with residency CL and EU.
What the artifact actually shows
The snapshot is structurally repeated: these counts are evidence of deterministic artifact production, not workload diversity, prevalence estimates or security certification.
See it running
A recorded walkthrough of the current MVP — compilation, runtime control and the evidence it produces. Full detail on the Live MVP page.
Open Source from the Beginning
Secure AI infrastructure should be inspectable.
Argorix Lang is public because secure AI infrastructure should be discussed, tested and improved in the open. The project is early, but the direction is clear: build language-level infrastructure for AI-agent systems where safety, traceability and verification are part of the foundation.
GitHub Repository
Explore the source code, examples and implementation.
Contribute
Open issues, propose ideas and review the language design.
Roadmap
Follow the evolution from compiler pipeline to runtime safety infrastructure.
Apache-2.0 License
A permissive license for open collaboration and real-world adoption.
Part of the Argorix security ecosystem
From governance platform to language-level infrastructure.
ARGORIXLANG
Argorix Lang extends the Argorix philosophy into language-level infrastructure: governance, control, assurance, runtime evidence and secure AI-agent execution.
Argorix is focused on AI security, governance, runtime guardrails, red-team evaluation and audit-ready evidence. Argorix Lang is the open-source exploration of how secure agent behavior can be expressed closer to the language and runtime layer.
Build safer AI-agent systems from the language layer.
Argorix Lang is early, open and designed for people thinking deeply about AI safety, agents, compilers, runtime governance and verifiable execution.