AIDR is the runtime security foundation for All Agents
on Every Endpoint
Start with runtime visibility and detection, investigate through session-level evidence, then expand into identity, data, execution, and remediation controls as each response capability is validated.
System / AIDR_01
Evidence / session-native
Decision / policy-bound
Delivery / phased
Traditional security tools can't understand Agents
Enterprises are running more Agents on endpoints, but security teams can't see their actual behavior chains.
Evidence hierarchy / verified trace
session:sess-7f3a
└─ turn:018 / identity:developer
└─ step:041 / ReadFile / policy:recorded
Traditional EDR tracks what processes do
AIDR reconstructs what Agents did and what they affected
Process → File → Network 5-tuple
Session → Identity → Toolchain → Model → Semantic Trace
Signature-based detection
Runtime detection + Investigation evidence + Response controls
From Visibility to a Response Loop
Runtime Visibility
DeepTracing real-time tracking of tool calls, model loads, file access, network activity, MCP communications, and permission decisions — every alert traces back to session-level context.
Detection & Investigation
AI-BOM, behavioral baselines, and drift rules locate anomalies; session evidence then shows which identities, data, and execution paths were affected.
Phased Response
Audit is available today. Warn, Approve, and execution blocking are delivered in stages and shown with explicit availability states.
Evidence path
Available
Response path
Phased delivery
Six Differentiating Capabilities
Full Endpoint Agent Observability
Unified collection of six event types, supporting IDE Agent, CLI Agent, MCP-driven, and custom Runtimes.
DeepTracing Evidence Chain
Every alert traces back to session-level Trace, not just process or network 5-tuple.
AI-BOM Asset Graph
Model, toolchain, MCP, and Agent dependency graph with diff support and compliance inventory.
Detection & Investigation Loop
Use baselines to locate drift, then return to session evidence to confirm identity, data, and execution impact.
Transparent Response Roadmap
Audit is available; Warn, Approve, SoftBlock, and HardBlock are delivered only after validation.
Three Deployment Models
Full SaaS, hybrid, and private models define different data boundaries; hybrid and private keep raw events customer-side.
Cover Every Integratable Agent Runtime
Beyond MCP: cover IDE, CLI, MCP, and custom Agents, while labeling each runtime by its actual observability depth.
IDE Agent
Cursor / Copilot
CLI Agent
Terminal Automation Agent
MCP Agent
Cross-tool Orchestration Agent
Custom Agent
Enterprise Custom Runtime
SDK/Sidecar integration, outputs six core events, supports policy feedback
Process/session identifiable, partial semantic events
Only system calls visible, added to onboarding backlog
Data Sovereignty, Boundaries Defined by Mode
Hybrid and private deployments keep raw events customer-side by default. Full SaaS stores and processes them inside the customer's selected cloud tenant boundary.
| Data Type | Storage Location | Outbound | Notes |
|---|---|---|---|
| Raw Event Details | Deployment-dependent | By mode | Cloud tenant for SaaS; customer-side for hybrid/private |
| Masked Event Summary | SaaS + Customer | Yes | Hashed paths/domains |
| Alert Metadata | SaaS + Customer | Yes | No plaintext sensitive fields |
| AI-BOM Asset List | SaaS + Customer | Configurable | Supports stats-only sync |
| Policy Config | SaaS → Customer | Yes | Requires signature verification |
Measured by integration, not vanity metrics
11
Agent runtime integrations in the current matrix
OTel
OpenTelemetry-aligned telemetry
3
SaaS / hybrid / private deployment models
AIDR / next evidence record
Start Securing Your Agents
Start with endpoint runtime visibility and detection, then build an Agent security foundation that is investigable, auditable, and ready for phased response.