You ask your AI. We make the work hold up¶

You ask your AI to do compliance work in plain language. The Verifier checks every proposal server-side, and you approve with one click, outside the conversation. The change lands in your canonical record and the chain captures the full reasoning. No opt-out, no bypass. The trust layer enforces the workflow, and the workflow is what makes the work defensible.

Three things on your side, nothing else to install¶

Bring your AI¶

Anthropic Claude, OpenAI, Google Gemini, AWS Bedrock, or Azure OpenAI. You keep the contract with the AI vendor, and Kanonik never sees your conversations.

Load our skills¶

Your AI pulls compliance skills on first sign-in, and they turn a generic frontier model into a defensible worker that knows your framework and what an auditor expects.

Keep the receipts¶

Every change to your record is checked on our servers, gated by a signed human approval, and sealed into a tamper-evident, append-only record.

How a request becomes a defensible trail¶

You ask¶

In the AI tool you already use, in plain language.

It does the work¶

Using Kanonik's skills and the data you point it at.

We check it¶

The Verifier runs server-side before anything is real.

You approve¶

One click from a signed-in person's account, on the approval link or in the dashboard. Nothing your AI drafts reaches your compliance record without an approval from a signed-in person's account. Kanonik cannot tell whether the person, or software acting in their signed-in browser, made the click. The few things saved without an approval are listed here.

We seal it¶

Into a record that cannot be quietly changed.

Six steps. None of them skippable¶

You ask your AI¶

Inside your AI client you describe the work: draft a policy for ISO 27001:2022 A.5.1, assess a change-management control for Q1. Anything you would ask a senior compliance analyst.

Your AI calls Kanonik¶

Loaded with the matching Kanonik skill, your AI uses a curated set of MCP tools. Read tools return instantly; write tools always generate proposals, never direct writes.

The Verifier checks¶

Server-side, inside every commit tool, before any write can land: a rule layer (schema, scope, framework reference) plus an LLM cross-check that catches hallucination and out-of-policy reasoning, and neither your AI nor you can bypass it.

You approve, out of band¶

Verified proposals generate a signed approval token delivered by email to the person who decides. If you connect Slack, it posts a notice that a request is waiting, with no approval link. The approver sees the proposal, the reasoning, the Verifier verdict, and the exact diff. One click. In-conversation approval is never sufficient.

The change lands¶

Once approved, the write commits to your canonical record. Versioned, bitemporal, reversible. The skill version that produced it is pinned to the event.

Chained to the audit log¶

Proposal, verification, approval, commit: each captured as a signed event chained to the one before. The full chain exports on demand as a sealed package your auditor verifies offline.

Your first output the same afternoon¶

Sign up at kanonik.ai and paste a single setup command into your AI client. OIDC sign-in with PKCE handles the rest. About two minutes.

Pull the skill library¶

Your tenant fetches the signed skill bundles on first sign-in. Core skills today: policy architect, control assessment, audit narrative, and audit readiness. More ship continuously.

Configure approvals¶

Pick who approves, and approval links go by email to that person. If you connect a Slack channel, it gets a notice that a request is waiting, never the link.

No new tool to learn and no console to fill in. You ask in plain language; Kanonik is the part that checks the answer, holds it for your approval, and seals it into a record an auditor can trust.

The proof is the product.