Monitoring your tenant
Conduit gives every organization three surfaces for keeping an eye on its own tenant: a usage dashboard, a vendor health view, and an audit log. Each answers a different question. This guide explains what each one is for and points to the deeper reference.
Everything here is org-scoped — an organization only ever sees its own data. Platform-wide observability across all tenants is a WYRE-internal surface and is not part of the customer product.
Usage dashboard
Section titled “Usage dashboard”Answers: “How much are my AI agents using Conduit, and on what?”
The usage dashboard (part of the /org overview page in the Conduit web app) aggregates tool-call activity for your organization over a date range:
- Totals — total calls and active users, as stat cards at the top of the dashboard.
- Vendor breakdown — for each connected vendor: call volume, unique users, and its top tool.
- Estimated savings — an estimate of the tokens and cost saved by routing calls through the CLI wrapper instead of raw MCP, shown alongside the totals above. These figures are a heuristic estimate, not a metered measurement.
Access: the dashboard requires an active Conduit subscription (billing in good standing) and an admin or owner role. The underlying endpoints are documented under API § Usage Dashboard.
Vendor health
Section titled “Vendor health”Answers: “Are the vendor tools my agents depend on actually up right now?”
The vendor health view reports the current status of each vendor MCP server your organization has connected — healthy, degraded, down, or unknown — along with last-checked time and latency.
It is current-state only: there is no uptime history or sparkline. For the four health states, how they are measured, and what a down vendor means for tool calls, see the Vendor Health reference.
Audit log
Section titled “Audit log”Answers: “Which tools were used, by whom, and when?”
The audit log (/org/audit) records metadata for every MCP tool call — user, vendor, tool, status, latency, timestamp. A separate admin audit log records administrative actions (member changes, credential updates, allowlist edits).
Names in those tables are labels, not ids. Conduit shows a person’s name when it has one, otherwise their email, otherwise a service account’s name, otherwise a short fallback (the last four characters of the id). Hover a name to see the full id. Background jobs such as log shipping show up as a readable name (“Log shipper”), not a system: token. CSV export uses the same label in the user or actor column and keeps the stable id in user_id or actor_id.
Two things to know before you rely on it:
- It is metadata only. Conduit never logs prompts, tool arguments, or tool results. See Prompt Capture & Privacy for the exact privacy boundary.
- Access is gated by billing status (active subscription), role (
admin/owner), and org scope. CSV export has no separate gate — anyone who can view the audit log can export it. Entries are auto-purged after 90 days — for longer retention, configure log shipping to a SIEM.
The audit log API is documented under API § Audit.
Which surface for which question
Section titled “Which surface for which question”| Question | Surface |
|---|---|
| How much usage, and on what? | Usage dashboard |
| Is a vendor up right now? | Vendor health |
| Who used which tool, when? | Audit log |
| Is Conduit itself up? | The unauthenticated GET /health liveness endpoint — see Vendor Health § Related. |