End-to-end AI agent observability

Every AI agent. Every plane.
One source of truth.

AI agents act across three planes no single tool can see. Spectis is the only platform that observes all three — external API, internal network, and endpoint — correlates them by identity, and holds it all in one platform you control. Observe by default — enforce by choice.

SpectisLive
4,255events correlated today — across all three planes
Held in one platform · no third-party data sharing
P1External APIOpenAI · Anthropic · Bedrock
P2Internal networkMCP · tool calls · data
P3Endpointconfigs · skills · processes
Correlated by identityusername × host × agent_id
3
Planes observed
21
AI clients covered
Zero
Latency added
GDPR + CCPA aligned
SOC 2 · in progress
01 The blind spot

No single tool sees all three
planes of AI agent activity

AI agents reach across external APIs, internal networks, and local endpoints. Secure Access Service Edge (SASE) platforms see the first. MCP gateways see part of the second. Code-level scanners catch a sliver of the third. Nobody sees them together — and that gap is where shadow agents live.

PLANE 01

External API

Agent calls to model providers — OpenAI, Anthropic, Bedrock. Your SASE platform sees opaque TLS, never the prompts, tool calls, or decisions inside.

SASE sees the connection, not the conversation
PLANE 02

Internal network

MCP servers reaching databases, directories, Splunk. Deliberately bypassed from inline inspection to avoid latency — so it goes completely unmonitored.

Most sensitive AI traffic, zero visibility
PLANE 03

Endpoint

Process-to-process stdio, config files, skills, hooks, credentials — local to the workstation. Code-level scanners glimpse it; none correlate it back.

Where shadow agents are born and hide
02 Why Spectis

The only end-to-end view.
Everyone else covers part of one plane.

Each adjacent category was built for a pre-agent world. Even the newest AI-SPM tools, which now reach code-level endpoints, stop at coding agents and source repos — they never touch enterprise SaaS agents, and none correlate across planes. Spectis was designed to join all of them by identity.

Tooling categoryAPINetworkEndpointCross-plane
correlation
SASE / SSEpartial
MCP gateway / proxypartial
EDR / XDRpartial
AI-SPMcode & coding agentspartialcode-level*
Spectis

* AI-SPM now reaches coding agents (VS Code, Claude, Codex) and source repos, but not enterprise SaaS agents (M365 Copilot, Copilot Studio) — and none correlate the endpoint back to the API call and the human. Spectis joins all three planes on username × hostname × agent_id.

03 The platform

See every agent. Know every
action. Trust every outcome.

A telemetry and correlation layer — not a gateway, not a proxy. Nine capabilities, one shared identity graph, held and analyzed in one platform you control.

Endpoint agent

Scans MCP configs across 21 AI clients — VS Code, Cursor, Claude, Codex, Windsurf, Zed, JetBrains and more — finds running servers, probes tool inventories, surfaces shadow agents. macOS · Windows · Linux.

AI inventory + exposure graph

Every artifact, runtime, and the inferred flows between them. See what an agent can actually reach — endpoint → cloud model, repo → production — scored for blast radius.

Cloud AI runtimes

Read-only inventory of Bedrock & SageMaker, Azure OpenAI & AI Foundry, and GCP Vertex — guardrails, public endpoints, and who can invoke what.

M365 Copilot inventory new

Tenant feature flags, license adoption, Copilot Studio agents, Graph connectors, sensitivity/DLP posture and usage — the enterprise SaaS surface code-level tools miss.

Supply-chain scanner

Crawls GitHub orgs for AI artifacts in source — agent definitions, skills, prompts, leaked tool grants — before they ship to production.

Agent identity + registry

Every agent gets a registered, scoped identity; every action is attributed to both the agent and the human who triggered it — the full delegation chain, audited.

Safety LLM

Async behavioral analysis that catches what rules miss — reconnaissance, exfiltration through read-only access, prompt injection, and privilege creep across sessions.

Your AI system of record

Everything is correlated and retained in Spectis — one source of truth you own. We stream only Spectis's own audit log to your SIEM; your correlated intelligence never leaves the platform.

Risk classification engine

A deterministic rule pack plus an optional LLM tier scores every artifact — benign, enterprise, suspicious, harmful — with a rationale you can audit.

Developers & AI clients
CursorClaudeVS CodeCodexWindsurfZed+15 more
Spectis Correlation Engine
joins by · username × hostname × timestamp × agent_id
Endpoint
21 clientsMCP scanTool prober
Orchestrator
Agent identityRegistryTool calls
Cloud + SaaS
Bedrock · VertexM365 CopilotGitHub
Network
SASE / FWDNS logsFlow data
Spectis dashboardAnalyticsRetentionAudit log → your SIEM
04 How it works

From scattered agents to
one source of truth.

Spectis observes across all three planes, joins everything by identity, and holds it in one platform you control — nothing in the request path, zero latency, no data handed to third parties.

AI agents & clients

Every agent across your org
ClaudeCursorCopilotCodexM365

Spectis observes

Read-only, all three planes
APINetworkEndpoint

Correlate by identity

One agent, one human, one story
username × host × agent_id

See, retain & govern

Your source of truth — held in Spectis
DashboardAnalyticsAudit log → SIEM
05 Hosting

Fully managed.
Live in minutes.

Spectis is hosted, secured, and run by us — nothing for your team to deploy, patch, or babysit. Connect your AI sources, see and govern everything in one platform, and you're live the same day.

Secure by design, minimal by default

Your platform runs on dedicated, encrypted, single-tenant infrastructure that we operate and monitor. And because Spectis is data-minimized at the source, the sensitive material never reaches us in the first place — we collect metadata, never your secrets or content.

  • Nothing to run — fully managed and maintained by Spectis; onboarding in minutes, not months.
  • Secure & isolated — dedicated single-tenant, encrypted at rest and in transit (TLS 1.3); SOC 2 on the roadmap.
  • Minimal by design — env var names only, never values; no raw commands or file contents. Spectis is your system of record; only Spectis audit logs are streamed to your SIEM.
Spectis Cloud · status
HostingFully managed
IsolationDedicated · single-tenant
EncryptionAt rest + TLS 1.3
TelemetryMinimized · names only
System of recordSpectis
SIEM exportAudit logs only
OnboardingSame day
✓ Managed & monitored — nothing for you to run
06 Enterprise

Enterprise-grade,
from day one.

Security, governance, and isolation that satisfy the teams who sign off — without a six-month procurement cycle.

Dedicated isolation

Dedicated single-tenant infrastructure — no shared environments, no co-mingled data, ever.

SSO, SAML & SCIM

Connect your identity provider for single sign-on and automated, policy-driven user provisioning.

Granular RBAC & audit

Scoped roles and a complete, attributable audit trail for every agent and every human action.

Compliance-ready

Data minimization by design, full audit logging, and GDPR alignment — SOC 2 on the roadmap.

SIEM audit-log streaming

Stream Spectis's audit trail to Splunk, Sentinel, CEF or webhooks for compliance — while your correlated intelligence stays in Spectis.

Dedicated support & SLA

White-glove onboarding, a named technical contact, and enterprise service-level agreements.

Ready to roll out Spectis across your org?Talk to our team about a dedicated enterprise deployment.
Request access
06 Security posture

Built to the standard
it holds others to

A security product has to be the most secure thing in the room. Spectis is engineered that way — minimal data, modern crypto, least privilege, and enforcement only where you choose.

Observe by default

Observability adds zero inline latency and never touches the request path — no conflict with your SASE or proxy stack. Enforcement intercepts only what you choose to block.

Data minimization

Env var names only, never values. No raw command lines. No file contents unless explicitly flagged.

Zero-trust agent comms

TLS 1.3, Ed25519-signed commands with replay protection, short-lived agent identities. Argon2id at rest.

Auditable by design

Every action attributed to both agent and human. Inspectable verdicts, full audit trail, evidence per alert.

08 Coming soon

One platform for observability,
compliance & enforcement.

Observability ships today. Compliance and policy enforcement are next — to our knowledge, no single company is building all three in one platform. Here's where Spectis is going.

MCP engine liveApp SDK soon

Policy & enforcement engine

Intercept and block non-compliant agents, skills, and tool calls in real time. Hooks for Claude, Codex, and VS Code today, plus the MCP policy engine. Coming: a drop-in SDK that brings the same enforcement — and prompt-injection detection — to your production apps.

Prompt injection · PII · policy violations · real-time block
Coming soon

AI usage metrics

Connect Claude, Codex, OpenAI, M365 Copilot, and GitHub Copilot for org-wide metrics — adoption, spend, model & tool mix, top users, and trends over time.

Claude · Codex · OpenAI · M365 Copilot · GitHub Copilot
Coming soon

AI compliance & analytics

Ingest server-side ground truth through the OpenAI and Anthropic compliance APIs, retain conversations, and turn them into analytics — who's using AI, how, and why — with LLM-powered summarization and risk scoring.

OpenAI & Anthropic compliance APIs · conversation analytics
Observability → Compliance → Enforcement. One platform.
07 Research

The Three-Plane Blind Spot

The first systematic mapping of MCP configuration paths across every major AI client, and a formal agent-identity model for enterprise security.

The Three-Plane Blind Spot: Why Enterprise Security Cannot See AI Agents

We show that SASE, MCP-gateway, EDR, and AI-SPM tools each cover at most one plane of AI-agent activity, formalize a cross-plane identity model, and evaluate Spectis against a reference enterprise environment.

EditionVersion 2 · 2026
Categorycs.CR · cs.AI
Read the whitepaper
Plane coverage by category
ToolAPINetEndpoint
SASE~
MCP gateway~
EDR / XDR~
AI-SPM~~
Spectis

Built for the agentic era

Bring end-to-end agent observability to your enterprise — held, correlated, and governed in one platform you control.