Help building guardrails your team can trust.

Why this matters now

The field has moved from "can agents call tools?" towho governs what those tool calls are allowed to do. Microsoft shipped the Agent Governance Toolkit v4.0.0 (Jun 2026) with 992 conformance tests and 110 contributors — the enterprise standard for Azure teams. OpenTelemetry is standardizing GenAI and MCP spans. OWASP's GenAI work gives teams a risk vocabulary, and EU AI Act transparency obligations start taking effect in August 2026.

Most teams deploying agents have model governance (evals, bias testing, red-teaming). Almost none haveagent governance — the layer between "the model is fine" and "the agent has production access." That gap is where side effects, approvals, audit trails, and recovery procedures either exist or fail.

WhiteMagic's position: Microsoft AGT is the right choice for large Azure estates. WhiteMagic's Dharma Engine is the lightweight, framework-agnostic alternative for teams who need governance without the infrastructure overhead — or who need to extend AGT with custom policy logic, audit formats, or non-Azure runtimes.

What I deploy

A framework-agnostic middleware layer that every agent action passes through:

• Policy engine — declarative rules evaluated at agent runtime with deterministic, sub-millisecond latency
• Audit ledger — every action, with inputs, outputs, policy verdicts, and agent identity, hash-chained so tampering is detectable
• Approval workflows — high-autonomy actions route to human reviewers with configurable quorum rules
• Identity & trust — every agent has a signed identity; trust decays on violations and recovers on clean operation
• Kill switch — single-operator action to halt all agents, isolate specific ones, or roll back to a known-good state

Why me, specifically

I built this layer — Dharma rules engine, Karma audit ledger, eight-stage middleware pipeline — inside WhiteMagic starting in late 2025. Recent SDK and standards work validates the shape: guardrails, traces, tool-call policy, and evidence generation are becoming table stakes. You get a consultant who has already walked those design trade-offs in a working codebase.

For teams already committed to Microsoft's toolkit, Databricks Unity AI Gateway, or similar: I can integrate with and extend those rather than replace them. The goal is governance that works, not governance with my name on it.