Skip to content

Vei (Virtual Enterprise Infrastructure)

Summary

Vei is a virtual enterprise simulation framework developed by Rohit Krishnan at Strange Lab. It provides a persistent, stateful simulated company environment — including service tickets, billing cases, dispatch state, email, Slack, a wall clock, and exception registers — that can be used to test how AI agents behave in realistic multi-agent organizational settings. Vei was the experimental substrate for the "Helios Field Services" multi-agent misalignment experiment.

Details

What Vei Provides

According to Krishnan's essay, Vei provides:

  • Persistent company state — service tickets, billing cases, dispatch state, documents, Slack, mail, etc. that change as agents act
  • Role-bounded agents — each agent sees and touches different surfaces of the enterprise system
  • Official state fields — agents can change or preserve specific operating state fields (e.g., SLA clock posture, work-order state), making drift observable
  • Replayable seeds — the same scenario can be run with the same random seeds to compare multi-agent vs. single-agent behavior
  • Artifact capture — all agent outputs are logged for post-hoc audit

Why Vei Matters for Multi-Agent Research

Multi-agent evaluation is notoriously difficult. Krishnan documents several experimental designs that failed because they accidentally coached agents toward truth (too-visible evidence, integrity language in role prompts, shared documents that eliminated distributed-organization dynamics). Vei provides a realistic enough environment that authentic multi-agent failure modes can be elicited:

  • Agents can have different information surfaces (they see different things)
  • State is persistent and authoritative (not reconstructed each round)
  • The "right answer" can be verified against ground truth independently of agent outputs

Without Vei, Krishnan writes, the narrative drift finding would have required rebuilding the infrastructure from scratch. The point of Strange Lab was to have better ways to test agents in simulated enterprises.

Relationship to Strange Lab

Vei is the primary infrastructure tool of Strange Lab, Krishnan's research project. The Helios Field Services experiment ran on Vei.

Known Limitations

  • Designing scenarios that elicit authentic failure without coaching is extremely difficult — Krishnan documents multiple failed experimental designs
  • Whether Vei generalizes beyond service-operations companies to other enterprise topologies is unknown

Key Claims & Data Points

  • Provides persistent company state including service tickets, billing cases, dispatch state, email/Slack — [source: rohitt-krsnan-aligned-agents-misaligned-organisations-2026.md]
  • Enables replayable seeds for comparing multi-agent vs. single-agent runs on the same scenario — [source: rohitt-krsnan-aligned-agents-misaligned-organisations-2026.md]
  • Was the experimental substrate for the Helios Field Services multi-agent misalignment experiment — [source: rohitt-krsnan-aligned-agents-misaligned-organisations-2026.md]

Open Questions

  • Is Vei open-source or publicly accessible? What is the repository URL?
  • What language and stack is Vei built on?
  • What other experiments have been run on Vei beyond Helios Field Services?
  • Can Vei be configured for non-service-operations organizational topologies?

Sources