Governance your agents can follow.

Purpose-built for the policies your agents read. Designed for the audit trail your auditor trusts.

AI agents are doing work in your name. Today, no one outside your team can verify what they're allowed to do, what they actually did, or which policies they followed. That changes here.

Dictiva is the missing identity, policy, and evidence layer for agentic systems. It treats agents as first-class principals — not handles in a session — and gives every team building with AI a way to publish what they commit to and prove it.

What we build

Three things that have to be true.

Identity

Every agent has a stable, verifiable name. Not a session token. Not a service account someone forgot to rotate. Each registered agent gets a W3C DID, an accountable human, and a place in the audit trail an auditor can actually follow.

Policy

Your policies are statements an agent can read. Scope, enforcement mode, and refusal rules live inside the statement — not buried in a PDF an LLM has to summarize. Authored once, served via MCP, honored by every agent that touches your tenant.

Evidence

Every commitment an agent makes is a Policy Commitment Attestation — a signed credential that composes W3C VC, in-toto, and OSCAL. It expires, it can be revoked, and it's verifiable by anyone. Your auditor doesn't have to trust us. They can check.

Where to start