Trust every AI agent by what it actually does

EDAMAME is the trust layer for the AI agents and machines behind your code — proving trust through both security posture and runtime behavior, on every host: local, remote, and headless CI/CD.

The problem

Is the agent doing what it declared — or something else?

Cursor, Claude Code, Codex, OpenClaw and Hermes read repos, call tools, and touch credentials. Identity and static posture tell you the machine is hardened — they don't tell you whether the agent stayed inside what it declared.

EDAMAME watches each agent from outside, at the host boundary — no SDK or plugin inside the agent. It scores how far the agent drifts from its declared intent, flags attack patterns, and rolls the same evidence up fleet-wide in Hub.

How it works

From host evidence to fleet-wide trust

Step 1 · Know locally

Score behavior on each host

EDAMAME compares what each agent declared with what it actually did — process, file, network, and secret activity — scores the divergence, and flags attack patterns. Observed from outside the agent.

Step 2 · See fleet-wide

Roll it up in EDAMAME Hub

The same evidence becomes fleet checks — divergence scores, attack-pattern findings, blast radius, and conditional access when an agent strays.

Posture, behavior, and fleet-wide evidence

Verify each AI agent on the host — security posture plus runtime behavior — then see the same evidence across your fleet in Hub.

Intent-divergence score

How far the agent drifted from what it declared — scored on real process, file, and network activity, not a prompt log.

Attack-pattern findings

Credential harvest, token exfiltration, tool poisoning, and malicious package pulls — caught from host telemetry, with the evidence trail.

Security posture

Disk encryption, screen lock, OS patch level, EDR presence — the static hardening of every host an agent runs on, local or headless.

Fleet trust & conditional access

Hub rolls posture, divergence, and findings up fleet-wide — and gates access to repos and resources when an agent or host drifts out of trust.

Trust agents on evidence, not assumptions

Trust agents on evidence, not assumptions

Score divergence, catch attack patterns, gate access — on every AI agent, fleet-wide

Score divergence, catch attack patterns, gate access — on every AI agent, fleet-wide

Trust agents on evidence, not assumptions

Score divergence, catch attack patterns, gate access — on every AI agent, fleet-wide