| sidebar_position | 4 | ||
|---|---|---|---|
| sidebar_custom_props |
|
AI Inventory helps teams maintain visibility into AI assets and AI activity sources across Browser Extension, Endpoint Agent, LLM Gateway, MCP Gateway, and compliance API coverage. It shows where AI usage is happening, what assets or applications are involved, what guardrails apply, and where additional review may be needed.
Use AI Inventory when you need to:
- Review AI assets and applications discovered across the organization.
- Compare browser, endpoint, gateway, MCP, and compliance API sources.
- Inspect asset metadata, guardrails, tags, status, source, and integration points.
- Investigate endpoint coding inventory such as agents, skills, MCP servers, models, hooks, permissions, plugins, and repositories when endpoint telemetry provides it.
- Review all applications discovered on endpoints through asset inventory, inspect OS type, privilege, approval status, and execution policy, and update execution policy inline.
- Review LLM Gateway API keys and MCP Gateway servers from an inventory perspective.
- Track adoption, request volume, blocked activity, sensitive activity, and source-specific trends.
- Switch between Browser Extension, Endpoint Agent, LLM Gateway, MCP Gateway, and Compliance APIs source views.
- Review source-specific counts and table metrics.
- Search and filter inventory rows within the active source.
- Open a source-aware details drawer for overview, interaction, and configuration context.
- Review Browser Extension applications with the same app-management controls used by the Applications page, scoped to browser-discovered apps.
- Review Endpoint Agent application groups with request, sensitive, detection, blocked, and user metrics.
- Review endpoint coding inventory where available, including agents, skills, MCP servers, models, hooks, permissions, plugins, and repositories.
- Review endpoint-discovered applications with the Discovery sub-view: filter by name, user email, category, source, OS type, approval status, and criticality, and update execution policy inline.
- Review LLM Gateway API keys with request, blocked, anonymized, model, last-used, and posture context.
- Review MCP Gateway servers with tools, scopes, DLP action, status, and activity metrics.
- Review the Compliance APIs source through Overview, Assets, Usage, Governance, and Security tabs to monitor organization-wide metrics, browse ChatGPT conversations, Codex sessions, workspace agents, and apps, analyze user adoption and model-usage rankings, track token consumption and sharing risk, and investigate security and audit events.
The Browser Extension source focuses on applications observed through browser activity. It uses the same app intelligence and app-management experience as the Applications page while keeping the inventory view scoped to browser-discovered apps, including the same review-before-apply screen and background Pending bulk actions status indicator for bulk Approve, Unapprove, Block, and Unblock operations described in Applications.
The Endpoint Agent source focuses on endpoint-observed applications. It summarizes requests, sensitive activity, detections, blocked activity, and users. Where coding inventory is available, the detail view can also show related agents, skills, MCP servers, models, hooks, permissions, plugins, and repositories for the selected application group.
A Discovery sub-view within the Endpoint Agent source lists all applications observed on endpoints through asset inventory, including applications that have not generated AI-specific activity. Discovery rows show application name, associated users, device count, OS type, privilege, category, source, and execution policy. Admins can filter by name, user email, category, source, OS type (macOS, Windows, Linux), approval status (Needs Review, Approved, Blocked), and criticality (Critical, Not Critical), and toggle off OS system processes. Execution policy (Allowed or Blocked) can be updated inline for each row.
The LLM Gateway source lists protected LLM application keys. It shows gateway-native metrics such as requests, blocked requests, anonymized requests, configured models, last use, DLP action, Guardian Agent status, and key status. The drawer provides logs-backed overview and interaction context plus embedded gateway configuration sections.
The MCP Gateway source lists gateway-managed MCP servers. It shows server identity, transport, auth mode, tools, scopes, DLP action, status, and activity metrics. The drawer provides MCP logs, analytics, interaction details, and embedded General, Guardrails, and Tools configuration sections.
The Compliance APIs source provides provider-specific inventory for compliance integrations. The OpenAI Compliance view is organized into five tabs:
- Overview: Shows organization-level metrics for the selected time range — observed users, events, ChatGPT conversations and messages, files, Codex sessions, agent runs, connector calls, and total tokens — plus snapshot cards for ChatGPT inventory totals, Codex activity, and user adoption highlights, and a data-freshness and source-status panel.
- Assets: Lists inventory rows for each supported product with search, filters, sort,
pagination, refresh, and source-aware detail drawers. Products and their asset tabs:
- ChatGPT: Conversations, Projects, GPT Usage, GPT Inventory, Knowledge Files, Conversation Files, and Users. GPT Inventory lists stateful GPTs with visibility, sharing, capabilities, knowledge files, and external actions; the detail drawer shows full configuration sections and top-user lists. Knowledge Files lists files linked to GPTs with link count, GPT count, and owners. Conversation Files lists files shared in conversations. Users shows per-user activity with a 360 detail view.
- Codex: Sessions, Tasks, and Environments.
- Spreadsheets: Sessions — Microsoft Excel add-in sessions with event counts, content events, model usage, and token totals.
- Agents: Agents, Connectors, and Skills. Skills shows the workspace skill catalog with creator, file count, and agent cross-references; skill name chips in conversation and agent run views link directly to the Skills tab.
- Apps: Apps — OpenAI app activity, auth events, returned resources, and per-app user reach.
- Usage: Provides Users, Sources, and Rankings sub-tabs with ChatGPT top-user lists sorted by conversations, Codex top-user and model rankings, source adoption breakdowns, source overlaps, and cross-source transition flows.
- Governance: Organizes compliance risk data across three sub-tabs. Token Governance shows total token consumption, input/output/cached composition, cache-efficiency ratio, token field coverage, a daily consumption trend, and consumption rankings by user, model, source, and client. Sharing Blast Radius surfaces ChatGPT sharing activity — conversation shares, GPT access grants, and scope-change signals — with risk highlights and a paginated event log. Egress shows network egress data for the selected period.
- Security: Organizes OpenAI admin-layer security events across seven sub-tabs: Audit Log, Logins, Denied (blocked and failed events), Sessions, Admin Access, Anomalies, and Egress. Each event sub-tab includes a posture header with key stats, an action or outcome composition bar, a filterable event table with per-row user and geographic context, and a country-based signal callout.
The default time range for the Compliance APIs view is 7 days. Claude Compliance is represented as a provider option when available but may be disabled until configured for the tenant.
- Open AI Inventory and choose the relevant source.
- Review source counts, table metrics, and posture signals.
- Search or filter the active source table.
- Open an inventory row to review overview, interactions, and source-specific details.
- Use embedded configuration sections or related platform areas when an item requires action.
AI Inventory requires AI asset access. Source visibility depends on tenant configuration and permissions for the underlying source, such as Endpoint Agent, AI Gateway, MCP Gateway, or Compliance. Endpoint Agent inventory appears only when endpoint coverage is enabled for the tenant.