Product
INTREXA AXIS
The managed authorization layer for autonomous AI agents. Identity, authority, policy decisions, revocation, and signed evidence — all included. Your agents call our API. We handle the rest.
20
Open specifications
< 500 ms
Revocation propagation
Capabilities
Agent Discovery
Find AI agents before they find vulnerabilities
Traditional identity systems discover humans and applications — they have no concept of an autonomous AI agent. Agent Discovery continuously surfaces ungoverned digital workers and brings them into the governance lifecycle: identity, trust scoring, policy enforcement, and audit.
Agent Identity
Know exactly who — or what — is acting
Every agent gets a cryptographically verifiable identity based on the DID standard. Agent IDs are issued against human principals and carry ownership, scope, and credential metadata. Identity is the foundation every other governance layer is built on.
Trust Scoring
Quantify how much to trust each request
Trust Score combines three components: lineage score (delegation chain depth and quality), credential score (verification and attestation history), and anomaly score (behavioral deviation from baseline). Updated per session — not static.
Branch-Cut Revocation
Cut off access instantly — everywhere
Revoke one agent and the cut propagates to every downstream agent in the delegation tree in under 500 ms. Branch-cut semantics mean you never have a partial revocation state. Works across single agents and deep multi-agent chains.
Audit Artifacts
Signed, tamper-proof records of every decision
Every authorization decision produces a signed artifact: who asked, what was decided, why, what context was present, and the full delegation chain. Artifacts are replayable and suitable for regulatory compliance evidence.
Compliance Reporting
Meet NIST, ISO, and EU AI Act requirements
Pre-built compliance reports for NIST AI RMF, ISO 42001, and the EU AI Act. Policy drift detection alerts you when your governance posture changes. Enterprise tier includes dedicated audit assistance and incident runbooks.
Where AXIS sits
AXIS isn't another layer of AI observability.
AXIS connects the authorization gap between the agent and the systems it can act upon. Each layer in your stack already has a lane — AXIS is the missing one, not a replacement for the rest.
IAM
identity
Policy engine
policy
API gateway
network / API control
SIEM
observation
AXIS
agent authorization
Technical architecture
Six primitives, one authorization model.
Every decision AXIS makes is built from the same six primitives, whether the caller is an SDK, a gateway, or a direct API request.
Actor
The agent making the request, identified against its human or org principal.
Authority
The scoped permission the actor holds — never more than what was delegated to it.
Action
The specific operation being requested, e.g. purchase.approve.
Decision
Allow or deny, evaluated deterministically against active policy.
Receipt
A signed, replayable record of the decision and the context behind it.
Invalidation
Revocation or expiry — authority that was valid can stop being valid mid-session.
Where AXIS sits in the request path
Enterprise Systems
│
┌────────────┼────────────┐
▼ ▼ ▼
APIs MCP Tools Data
▲ ▲ ▲
│ │ │
└──────── AXIS ───────────┘
▲
│
Authorization
│
AI Agent / RuntimeTrust
What's actually shipped — labeled, not implied.
We label status honestly rather than implying everything is uniformly complete. PARTIAL means the capability is real but coverage is limited — not a euphemism for missing.
Explicit authorization
Every action evaluated against actor, authority, action, resource, and policy.
Deterministic decisions
Same request, same context — same decision, every time.
Delegated authority
ES256-signed delegation chains; a child can never exceed its parent's scope.
Revocation
Branch-cut revocation propagates to the full downstream tree in under 500ms.
Expiration
Delegation tokens carry expiry; execution-time checks reject stale ones.
Execution-time validation
Gateway-routed calls (MCP/A2A/HTTP) are re-validated by default before dispatch. Direct SDK callers outside the gateway must opt in.
Evidence receipts
Every decision produces a signed, replayable artifact.
Policy-driven enforcement
The Policy Decision Point evaluates every request — no implicit allow.
Framework / runtime integration
SDK integrations exist for LangChain, CrewAI, MCP, and browser agents; depth of coverage varies by framework.
Compliance mapping
Pre-built mappings for NIST AI RMF, ISO 42001, and the EU AI Act support audit preparation — not a certification or legal compliance guarantee.
Open specifications
Published under Apache 2.0 as agf-standards — open for anyone to implement against.
For developers
Authorization belongs in
the execution path.
AXIS is designed to integrate into existing agent infrastructure, not to require rebuilding it. One call before the action executes.
decision = agf.authorize(
agent_id="procurement-agent",
action="purchase.approve",
resource="po:48291",
)
if decision.allowed:
execute()Use cases
Consequential actions, not chat transcripts.
Financial Agents
What can it do?
Approve payments, refunds, or transactions.
What authority does it need?
payments:approve, refunds:issue — scoped by amount
Where does AXIS enforce it?
Policy threshold at decision time; re-checked before dispatch
IT Agents
What can it do?
Modify infrastructure or production systems.
What authority does it need?
infra:modify — scoped to environment and resource
Where does AXIS enforce it?
Decision-time policy, plus revocation if the session is compromised mid-change
Security Agents
What can it do?
Disable accounts, rotate credentials, change controls.
What authority does it need?
security:remediate — scoped to the incident
Where does AXIS enforce it?
Delegation chain traces the action back to what authorized it
Customer Operations
What can it do?
Issue refunds, modify customer records, send external comms.
What authority does it need?
customer:write, comms:send — scoped per case
Where does AXIS enforce it?
Every action leaves a signed evidence receipt for dispute resolution
Data Agents
What can it do?
Access sensitive datasets or export information.
What authority does it need?
data:read, data:export — scoped by classification
Where does AXIS enforce it?
Policy evaluates data classification and destination before export
Enterprise Workflow Agents
What can it do?
Approve, create, modify, or delegate business actions.
What authority does it need?
workflow:approve, workflow:delegate — scoped by process
Where does AXIS enforce it?
Delegation model — a sub-agent can never exceed what it was handed
How it works
One API call.
Full governance.
Your agent sends a single authorization request. INTREXA AXIS verifies identity, computes trust score, evaluates policy, and signs an audit record — all before returning the decision.
You integrate once. We handle availability, policy updates, and compliance. No infrastructure to provision or maintain.
API request lifecycle
Watch it work
Registration to a signed audit report,
end to end.
A full walkthrough of INTREXA AXIS: register an organization, register an agent, generate and upload its keys, mint a delegation token, write and activate a policy, evaluate a live decision, and inspect the signed audit record it produces.
Every command shown is real — the same CLI and API calls a developer would run.
Ready to evaluate AXIS?
Put one agent through the authorization boundary and see what it produces.