Product / how it works

Policy, execution decisions and evidence. In one defined path.

Clariva is a software control and evidence layer for AI workflows. Integrate at a request boundary your engineering team controls, then evaluate the requirement that your existing stack leaves open.

The software in context

Connect the decision to the request.

  1. Your LLM-bound requestYour application connects a supported request path to Clariva.
  2. Configured policyClariva evaluates the connected request and records a policy decision.
  3. Continue, hold or rejectThe decision determines whether the request can proceed toward eligible model execution.

Before eligible execution

Clariva checks a connected request against the configured policy for its organization and workspace. The supported path can prepare input before deciding whether it may proceed.

A record for review

An engineer can inspect the policy decision and recorded outcome. In the implemented request path, rejected requests and requests awaiting review stop before the later provider-execution step.

Source rechecked October 21, 2026. This describes the implemented path; runtime behavior, provider nonexecution and record persistence still need validation in the selected environment.

Read the source walkthrough
A useful starting point

One AI feature. One consequential requirement.

Example: an embedded support assistant

Your application assembles tenant context for a summary. A reviewer asks which information and provider route are permitted before the request is sent.

Agree the policy, integration point and expected outcomes. Test allowed, transformed, denied and held cases while checking that useful summaries remain possible.

Integration and coverage

The boundary is part of the product decision.

You own the execution seam

An engineering owner controls insertion, credentials, retries and fallback. Map each relevant model call and any path that could bypass the control.

Agree the supported behavior

Confirm request contracts, policy versions, tenant binding, provider eligibility, failures, task utility and data handling before committing to a deployment.

Review the integration requirements
Common product questions

What to know before you integrate.

Is Clariva another model provider?

No. It is a software control layer in a specified application workflow. It does not replace your application or model provider.

Why use Clariva with an existing gateway?

Only if an important requirement remains unmet. Compare the configured boundary, reviewable evidence and engineering effort; if existing controls suffice, the extra supplier may not be justified.

Does a recorded rejection prove no model was called?

Source ordering and a recorded decision are limited evidence. Stronger nonexecution claims require independent provider or egress observations for the exact tested path, including retries and bypasses.

What about latency and detection quality?

Measure them with representative cases and buyer-owned quality criteria. No universal latency, detection accuracy or zero-leakage guarantee is asserted.

Start with a conversation

One workflow. One clear question.

Bring the AI workflow, the reviewer’s requirement and the controls you already use. Start with a free 30-minute fit review.

Any paid technical evaluation is separately scoped.