Glossary

Single source of truth for ArmorIQ-specific terminology. Every term here is also linked from the page where it's first used.

A

Agent

An AI software entity that calls tools, typically an LLM-backed program (Claude, ChatGPT, a custom agent, a framework like Google ADK / CrewAI / LangChain), invoking external functionality. In ArmorIQ, every agent is registered with an agent_id and is bound by a policy.

AIQraph

ArmorIQ's interactive topology visualization of agents, MCP servers, tools, and policies: how they connect and what's allowed to call what. See AIQraph.

API Key

The authentication credential the SDK and CLI use to talk to the ArmorIQ platform. Prefixed ak_live_ (production), ak_test_ (sandbox), or ak_claw_ (legacy ArmorClaw standalone; no longer issued). See API Keys.

ArmorClaude

ArmorIQ's integration product for Claude: installation and configuration to wire Claude agents through the ArmorIQ proxy. See ArmorClaude.

ArmorClaw

ArmorIQ's security plugin for OpenClaw agents: intent verification, drift detection and policy enforcement. Uses standard ak_live_ keys issued by the dashboard. See ArmorClaw.

ArmorCodex

ArmorIQ's integration for code-generation agents and editors. See ArmorCodex.

ArmorCopilot

ArmorIQ's integration for GitHub Copilot-style agents. See ArmorCopilot.

ArmorHealth

ArmorIQ's health-insurance-compliance product: coverage rules, claims processing, provider networks, and PHI access controls for AI agents. See ArmorHealth.

ArmorPay

ArmorIQ's financial-compliance product for AI payment agents: amount thresholds, velocity limits, and transaction controls. See ArmorPay.

armoriq.yaml

The config-as-code manifest for an ArmorIQ-protected agent. Declares identity, MCP servers, policy, and intent settings. Has a published JSON Schema. Point your editor at it for IntelliSense. See the YAML reference.

C

CSRG (Canonical Structured Reasoning Graph)

Cryptographic proof that the tool an agent actually invoked matches the tool the LLM reasoned about. Enforced by default; can be disabled per-config via intent.require_csrg: false (not recommended in production). See Intent Intelligence: How It Works.

D

Delegation Request

A human-in-the-loop escalation raised when a tool call exceeds what an agent's policy allows on its own (typically a monetary or authority limit) instead of failing the call outright. The plan is held until a human with a higher-ranked role than the requester approves or rejects it (a requester can never approve their own request); approval records a Delegate trust delta. Separate from IAP's own agent-to-agent trust chain. See Plans Governance.

Drift

The condition where an agent's runtime tool invocation diverges from its signed Intent Assurance Plan. The proxy detects drift and refuses the call. See Intent Intelligence.

I

IAP (Intent Assurance Plan)

ArmorIQ's core security primitive. The proxy intercepts every tool call between an agent and an MCP server, derives a cryptographically signed plan of what the agent intends to do, and refuses to forward any call that doesn't match the plan. See Intent Intelligence.

Intent Token

A short-lived token (default ttl_seconds: 300) issued by the proxy after a plan is accepted. The SDK passes it on every invoke() call; the proxy refuses calls without a valid, matching token. Get one via client.get_intent_token(plan).

K

Kill Switch

A per-agent, enforcement-time block: while active, every tool call the agent makes is refused with enforcementAction: block before policy or OPA evaluation even runs, regardless of what any policy would otherwise allow. Set for a preset or custom duration, or indefinitely, from the agent's Controls tab, or via POST /iap/agent/:id/kill and lifted via /iap/agent/:id/unkill. Separate from revoke, which disables an agent in the registry instead. See Agent Controls.

M

MCP (Model Context Protocol)

The open JSON-RPC protocol AI agents use to call external tools (file systems, databases, APIs, etc.). ArmorIQ's proxy speaks MCP on both sides: agents talk to the proxy, the proxy talks to your MCP servers. See MCP Servers.

MCP Server

A server that exposes tools to AI agents over MCP. ArmorIQ registers each MCP server under a short id (used in policy refs like travel.book_flight).

Merkle Tree

A hash tree the CSRG uses to commit to an agent's reasoning steps; the proxy verifies a Merkle proof on every tool call.

O

OPA (Open Policy Agent)

A third-party policy engine (openpolicyagent.org) that ArmorIQ integrates for context-aware enforcement analytics. Runs alongside ArmorIQ's native policy engine. See OPA Engine.

Organization

The top-level tenancy unit in ArmorIQ. All agents, MCP servers, policies, API keys, and audit logs belong to exactly one org. See Organizations.

P

PAP

A pre-flight refinement step the SDK can call before a plan is finalized, returning a decision of accepted, rejected, or escalated plus any policy predicates that failed. ArmorIQ persists these decisions and exposes them read-only through the /pap decisions dashboard and its stats endpoints. The acronym isn't expanded anywhere in the codebase or its docs: treat "PAP" as this step's name rather than a confirmed expansion. See Plans Governance.

Personal Organization

The workspace organization automatically created for every account the first time it verifies its email, signs in via OAuth, or completes the CLI device-code flow (ensurePersonalOrg), named after the user, with them as org_admin. Lets a brand-new account register an MCP server, add an agent, or generate an API key immediately, with no manual setup step. See Organizations.

Policy

A set of allow / deny rules over <mcp_id>.<tool_name> pairs that determines which tools an agent can invoke. Deny always wins over allow. Authored either in armoriq.yaml or via Policy Studio.

Policy Studio

The form-based dashboard editor for authoring policies with a live YAML preview. See Policy Studio.

Proxy

The ArmorIQ service that sits between agents and MCP servers, enforces policy, validates intent tokens, and forwards approved calls. Hosted at https://proxy.armoriq.ai by default; self-hosting is possible by overriding proxy.url in armoriq.yaml.

Q

Quick Scan

Ad-hoc URL vulnerability scanning: paste a URL, get a security report. Useful for evaluating an MCP server before registering it. See Quick Scan.

R

Reasoning Graph

See CSRG.

Revoke

The action, on an agent's header, that sets the agent's registry status to inactive via POST /agent/agents/:agentId/status. Unlike a kill switch, revoke has no timer or expiry (it stays in effect until the agent is re-enabled) and shows a confirmation dialog before applying. Use revoke for indefinite lifecycle management; use kill to immediately stop tool calls at the enforcement layer. See Agent Controls.

S

Skill

A reusable, named capability an agent can be configured with. ArmorClaw scans skills for vulnerabilities. See ArmorClaw.

T

Trust Delta

A signed, tamper-evident event-log entry in a plan's cryptographic chain of custody, not a reputation or trust score; there is no numeric trust rating for an agent or organization anywhere in the platform. Delta types include ReAnchor (the plan was modified after its initial commit), Delegate (a delegation request was approved), and Revoke (a token or delegation was revoked). See Plan Detail Tabs: Trust Deltas.