The Pitch How it works Architecture Live preview Glass box What we claim Verify Talk to us
Architecture overview

Identity. Trust. Intent.

A commercial overview of how AIF wraps cryptographic governance around your AI without changing your stack. Written for the C-suite. Detailed enough for your architect.

01
Your agents call the spine
Any framework · any LLM
02
Identity is read from your IdP
Azure AD · Okta · Keycloak
03
Intent is declared and signed
Before any action happens
04
The action is governed and forwarded
To the LLM provider you chose
05
Every step is signed into the chain
Tamper-evident · independently verifiable
06
A regulator can query it in plain English
Answer arrives as a signed credential
The three pillars

A spine, not another AI.

Brains are commodity — you can switch from Claude to GPT to Mistral. The Identity-Trust-Intent spine is what makes any of them accountable to your regulator.

PILLAR 01

Identity

Who acted?

Every AI agent has a verifiable identity tied to a public key — not a pseudonym. AIF reads the identity your existing corporate login already issues. Auditors can verify any agent without access to your servers.

Verifiable identity per agent
PILLAR 02

Trust

Can we verify it?

Every action is cryptographically signed and chained. Edit one byte and the system refuses to start. Any auditor can verify the entire history using only your public key. No trust in Predict AI required.

Mathematical tamper-evidence
PILLAR 03

Intent

Why did it happen?

Before any action, the agent declares its purpose. The declaration is signed. The action is linked back to the declaration. A regulator can ask plain-English questions and receive answers as W3C Verifiable Credentials.

Purpose declared, signed, queryable
What we deliver

A regulator-ready governance layer for any AI stack.

Six outcomes. Each is testable by your architect from a laptop in five minutes. Each is non-negotiable in production.

Outcome 01

Every AI action carries a signed, tamper-evident record.

The audit chain is built using public-standard cryptography — the same primitives used by governments and financial institutions. Edit one byte anywhere in the history and the chain breaks visibly. Any auditor can verify the full record using only your published public key.

Outcome 02

Zero code changes.

Your agents change one URL in their configuration. Everything else stays as it is — including the framework you've already built on.

Outcome 03
100%

of audit records are signed at the moment they are created. There is no unsigned path through the system.

Outcome 04

Self-hosted in Europe.

One package on your infrastructure. The audit trail and signing keys never leave your control.

Outcome 05

Independently verifiable.

Any auditor, regulator, or court can verify the chain without access to your servers and without trusting us.

Outcome 06

A regulator can ask, in their own language, and get a signed answer.

Plain-English questions about the audit history return structured answers wrapped as W3C Verifiable Credentials — independently verifiable by any standards-compliant verifier. Your compliance team's job becomes review, not investigation.

Four specialist agents inside the spine

One always-on. Three optional. Each governs a different slice.

Compliance is the always-on core. Safety, cost, and observability come bundled but can be enabled, disabled, or purchased independently as your needs grow.

CORE · ALWAYS ON

Compliance

Signed audit trail. Tamper-evident chain. Regulation-mapped exports. Per-agent kill-switch. The non-optional foundation.

OPTIONAL · STANDALONE

Safety

Blocks malicious inputs before they reach the model. Redacts sensitive data before responses leave. Jailbreak and injection guards.

OPTIONAL · STANDALONE

Cost

Tracks token spend per agent and per business unit. Enforces daily budgets with hard stops. No invoice surprises.

OPTIONAL · STANDALONE

Observability

Continuous health checks on every agent. Cross-agent learning so successful patterns spread. Governed memory across the fleet.

Integration paths

Three ways your agents reach the spine.

Most customers start with the proxy — it works with whatever you've already built. The SDK and starter kit are for specific extension needs.

PATH 02 — SDK

Programmatic control.

A typed SDK for registering agents, writing governed memory, exporting compliance reports, and operating the kill-switch. Used alongside the proxy when you want fine-grained programmatic access.

Effort:  npm install Best for:  custom operations Languages:  TypeScript · Python
PATH 03 — STARTER KIT

Agents included.

For customers without existing agents, the spine ships an optional starter kit with five pre-configured business agents — research, communications, content, operations, finance — under the same governance, ready to deploy.

Effort:  config flag Best for:  greenfield AI rollouts Replaceable:  swap in your own
Regulatory mapping

Built around the regulations your auditor will reference.

AIF ships the record-keeping mechanics that the major AI regulations expect. Your auditor still does the certification — we make their review straightforward.

EU REGULATION

EU AI Act · Article 12

Detailed, auditable record-keeping for high-risk AI systems. AIF maps audit records to each requirement with gap analysis for what's missing.

INTERNATIONAL STANDARD

ISO/IEC 42001

The AI management system standard. AIF provides the audit-trail mechanics that an AIMS implementation needs as evidence.

RISK FRAMEWORK

NIST AI RMF

The US risk-management framework. AIF supplies the auditable trail behind every governance decision, mapped to RMF functions.

An honest scoping note. AIF is the technical foundation that record-keeping requirements expect — it is not, and does not claim to be, a compliance certification. Certification is your auditor's job; we make their work straightforward. See the claims register for what we do and do not claim.
Who this is for

One spine. Three audiences. One source of truth.

Premium governance reads differently for different audiences. The same AIF surface answers all three questions.

For the C-suite

How do we stop the fine?

EU AI Act Article 12 record-keeping is built in. Tamper-evident by mathematics. Self-hosted in the EU. Priced at roughly 2.5% of your fine exposure for the enterprise tier.

For the architect

How does it fit our stack?

One package on your infrastructure, behind your IdP, in front of your LLM provider. Your existing agents change one URL. Zero migration. Existing identity, existing audit pipelines, existing observability.

For the regulator

How do we verify it?

Ask the system a question in your own language. Receive an answer wrapped in a W3C Verifiable Credential, signed by the deployment root, independently verifiable using only the deployment's published public key.

The next step

A 20-minute conversation is enough to know if AIF fits.

We respond within one business day. For your architect, the technical pack — including the full deployment runbook and source-repository access — is signed under NDA in your browser.