For security architects & teams
Evaluate the action the agent assembled — not just the model.
You already know the seam: authentication works, retrieval works, the policy engine works — and the action the agent composed from all of them still violates intent, because none of those controls was built to reason about the composed action. Apeiris defines the layer that evaluates that action, and the evidence it should produce — under whose authority, over what data, within what policy — not a verdict on the model and not a gate that decides for you.
Where to start
Follow the thesis, then a real action end to end.
The compositional idea
Security verifies components; enterprises deploy composed actions. The seams, the ownership test (visibility / authority / responsibility), and why mapped ≠ satisfied.
One action, fully traced
Walk a single autonomous action from actor to authority, data, knowledge, behavior, effect, and a signed ledger — the evidence chain a composed action creates.
Incident-to-assurance
The method on a real incident — JADEPUFFER agentic ransomware — mapping observed behavior to the controls that would have produced evidence.
Agentic · Identity · Authority
The control families closest to autonomous action: agent identity and delegation, tool-call authorization, human-in-the-loop gates, approval limits.
Agentic exposure check
A quick self-assessment of where an agentic deployment can reach, and which controls answer for each new execution path.
Explore the knowledge graph
From a control to its threats, ATLAS techniques, frameworks, and normative sources — provenance on every edge.
How Apeiris differs from the tools you already run
Apeiris is not a policy engine, CNAPP, SIEM, identity provider, or AI gateway, and it does not replace them. It defines the open control-and-evidence model those tools produce and consume evidence against — the model for reasoning about the action a workload assembles across identity, authority, data, knowledge, and behavior, and composing a reviewable verdict from evidence. It is an evidence layer, not another sensor and not a choke point. See where Apeiris fits vs GRC, gateways, model eval, policy engines & observability →
Available today vs. in development
The open corpus is usable now. The runtime platform that watches your environment is being built — we hold that line explicitly.
Available today
- The compositional thesis, one-action walkthrough, and incident method
- The Agentic, Identity, and Authority control families with evidence and blocking posture
- A local platform prototype (deterministic trust engine, CLI, MCP, Ed25519 attestation)
- The knowledge graph and ATLAS model, queryable over MCP
In development
- Runtime detection of capability drift in your environment
- Telemetry ingestion and per-tenant evidence storage
- Output connectors (ServiceNow, Jira, Splunk, Sentinel, …) and a connector SDK
Trace a real action
See the six checks a composed action must pass, and the evidence each one produces.
Not your role?