AI agents

Built for AI agents.

Security is filling up with agents that act on what they're told. Whisper gives them ground to stand on: a live graph they can query directly, answers that admit what they don't know, and evidence behind every verdict.

The connection

One endpoint, the agent does the rest

Whisper speaks the Model Context Protocol. An agent connects once and has the whole graph in front of it — no per-question API to wire up, no bespoke tool for each pivot. Because the graph is queried in Cypher, and language models already write Cypher, the agent expresses a hunt in its own words and refines it as the evidence comes back.

That turns a static lookup into an investigation. The agent asks who runs a host; if the answer is thin, it walks to the neighbours; if a neighbour is listed, it pivots into the campaign around it. Each step is a query it composed, against data it can cite.

An agent can even provision its own access — sign up and get a key programmatically, then connect over MCP.

"Using WhisperGraph, identify what acme-login.co
 actually is, assess it with coverage, and if the
 graph has no data, walk to its nearest known
 neighbours and tell me their threat posture."
A natural-language prompt the agent turns into live Cypher.
Grounding

An answer that admits what it doesn't know

The most dangerous thing an agent can do is treat silence as safety. Most feeds answer yes-or-no, so a host nobody has reported reads as benign — and a brand-new piece of attacker infrastructure sails through. Whisper’s verdicts carry coverageas a first-class field, so the agent can tell the difference between “checked, clean” and “never seen.”

assess(host)one callKnown-cleanchecked — nothing adverse?No dataunseen — not safe, just new!Listedwith the feeds behind it
Coverage is a first-class field. An agent can tell “checked, clean” from “never seen” and refuse to vouch for what the graph has no data on.
Known-clean
We looked and found nothing adverse. A safe conclusion the agent can act on.
No data
We have no information on this host. Not safe — just new. An agent that treats absence of evidence as evidence of safety is exactly the failure this prevents.
Listed
The host appears in threat intelligence, returned with the feeds and factors behind the verdict so the agent can weigh it.
Auditability

Evidence on every verdict

A score with no reasons can’t be defended, and an agent that produces one can’t be trusted. Every verdict Whisper returns comes with its factors and the exact source feeds, dated — the chain of evidence behind the conclusion, not just the conclusion.

For the agent, that’s a check against its own confidence. For the analyst reviewing its work, it’s the difference between “the model said so” and a sourced finding that goes straight into a ticket. See agent-grounding investigations →