Observability / Observability
Observability
Monitor AI usage across the organization: costs, tool analytics, agent sessions, execution logs, data export, and employee enrollment.
The Observability section of the dashboard provides visibility into how AI is used across the organization. It answers four questions: what AI spend is and where it goes, which MCP servers and tools are used most, what an agent did in a session, and who in the organization has adopted the platform. Open it from the Observability group in the project sidebar.
Access requirements
Section titled “Access requirements”Viewing the pages in this section requires the org:admin scope. Access is
included in the default Admin
role but not the
Member role: the nav entries stay visible to Members, but opening a page shows
an access-restricted notice.
All Observability pages share a common filter bar (date range, user, agent, account type) and an AI-assistant mode that answers natural-language questions over the same data.
Sections
Section titled “Sections”Usage analytics for MCP servers and tools: call volumes, failures, and ranked lists of the most used servers, tools, clients, and skills, the tools with the most errors, and the busiest people. Every row links into Tool Logs for that slice, so the board is the starting point for seeing which tools deliver value and where failures cluster.
The raw stream of tool executions across every MCP server, gateway, skill, and local tool. Narrow it with the facet rail (server, client, status, type, agent, role, account, user) or arbitrary log attributes, brush the timeline to a window, expand traces, and inspect the full input and output of any tool call.
A searchable list of agent chat sessions with risk indicators. Open any session to review the full transcript, including tool calls, token usage, duration, and flagged risk findings.
Track AI spend across the organization. Headline totals for tokens, cost, active users, and AI clients, with breakdowns by role, employee, model, agent, and more. Drill from org-wide totals down to individual sessions, compare periods, and export session-level data.
Export normalized OTLP logs, metrics, and traces from each project to Langfuse, Datadog, Grafana Cloud, an OpenTelemetry Collector, or another OTLP/HTTP destination. Exports are configured under Organization settings > Data > Exports.
Getting data into Observability
Section titled “Getting data into Observability”Observability pages populate once the observability plugin is installed in the AI agents used across the organization. The setup dialog offers Claude Code, Claude Cowork, Codex, Cursor, opencode, and Pi; the OpenClaw plugin is installed from the Plugins page instead. Until data arrives, pages show a setup prompt with per-provider installation steps.
Neither OpenCode nor OpenClaw has an OpenTelemetry exporter, so their sessions, tool calls, tokens, and cost arrive entirely through the observability plugin. See distributing plugins to OpenCode and distributing plugins to OpenClaw for what that changes.
Related organization settings
Section titled “Related organization settings”The switches and integrations that feed these pages live under Organization settings:
- Logging Settings controls what the platform records from instrumented agents, from tool-call traces to session transcripts and tool input and output. Everything on the Observability pages is powered by these switches, and a disabled switch shows up here as missing data rather than an error.
- AI Integrations connect provider compliance and admin APIs, which supply the Codex, ChatGPT, and Cursor spend on Costs and the imported conversations in Agent Sessions that device telemetry alone cannot see.
- Data export is the outbound side of the same pipeline: the normalized telemetry behind Observability, delivered to an external OTLP destination from Organization settings > Data > Exports. Exports do not change what the Observability pages show.