The tools we build
Software and infrastructure that make evidence about AI capturable, verifiable, and publicly checkable — for enterprises, for certification bodies, and for the commons.
Evidence Fabric
Weaves the evidence you're already producing into one auditable record — without replacing anything you use.
AI solution/model cards live in one repo, evaluations in a spreadsheet, sign-offs in email, monitoring in a dashboard nobody exports. When a regulator, customer, or board asks "show me," teams scramble to reassemble a story from fragments.
Evidence Fabric isn't another platform to install. It connects to the tools your teams already use — and to the governance features your cloud, model, and orchestration vendors already ship natively — and weaves what they produce into one continuous, versioned evidence record.
One question it answers instantly: "What evidence do we have that this system is safe, compliant, and performing — as of today?"
No installation
Connects to what you already run instead of replacing it. Nothing to rip out, no migration, no new system of record.
Vendor-neutral
Works across whichever model, cloud, orchestration, and governance tools you've chosen — or are still evaluating.
Platform-native friendly
Complements the governance capabilities your platforms already ship. It reads them; it doesn't duplicate or compete with them.
Decision-proof
Because it doesn't depend on your vendor choices, it survives them — including switches made mid-selection.
Capture
Pull evidence from your existing tools — repos, eval runs, monitoring, approvals — instead of duplicating work.
Structure
Map every artifact to a control or competency in your chosen framework, so gaps are visible at a glance.
Verify
Flag stale, missing, or contradicted evidence automatically — assurance is a living state, not a one-time audit.
Version
Every record is timestamped, so you can show what was known and true at any point in time.
Export
Generate an evidence pack for a regulator, customer, or board in minutes — not a two-week fire drill.
Connect
Feeds from our evaluation and assurance work, and from your systems' observability.
Status: early access, shaped with design partners.
Evidentia assessment infrastructure
The technical backbone certification bodies run their credentials on.
Evidentia is the assessment infrastructure credential programs run on. A serious AI-oversight credential can't be tested with multiple-choice questions. It has to assess whether someone can actually evaluate an agent, configure a guardrail, and read an observability dashboard.
Evidentia is the infrastructure that makes that possible — built on open primitives so it stays portable, and paired with a public register so any credential issued on it is verifiable by anyone. The aim is the oversight workforce AI deployments quietly assume: the person ready to take over, and the local validators every adopting site needs before it can trust a system in its own population.
How we build programs on itEvaluation harnesses
Reproducible environments for testing agent behavior, tool use, and edge-case handling.
Guardrail labs
Hands-on environments where candidates configure and defend real safety controls.
Observability dashboards
Live monitoring surfaces candidates must read, interpret, and act on.
Badging & public register
Issuance plus a public, checkable registry — verification as trust infrastructure.
Public-goods tools
Part of our public-goods layer — free, openly licensed, and built with nonprofit, academic, and foundation partners.
Open-source readiness tool
A self-assessment mapping a person or team against oversight competencies, surfacing gaps and ways to close them.
AI Oversight Workforce Index
A free, annual, methodology-transparent picture of which oversight and assurance skills are growing, shrinking, and missing.
Open assessment framework
A public, versioned, citable framework for defining and testing AI oversight competence — free for any body to adopt.
See the tools in action
Book a walkthrough and tell us what "show me the evidence" looks like for you.