What Jents discovers — and how to resolve it
Discovery surfaces AI usage that Jents can tell is yours but can't yet tie to a registered record. Every item is one of a small set of known cases — here's what each one means, and the single fix that clears it.
The one idea
Every AI call either ties to something you've registered — an agent you run, or a person on a tool you track — or it doesn't. The ones that don't land in Discovery, grouped by source (so 50 developers on one tool show as a single item, not fifty rows).
Discovery has two sides, matching the two ways AI reaches Jents:
- Agentic — traffic routed through the Jents gateway by an agent you run.
- CLI/IDE — usage captured from a tool your team uses, like Claude Code.
Recognized usage never shows here. A known tool used by known people lives in Observability → Subscriptions, not Discovery.
Agentic (gateway)
🔴 Untracked key
A gateway key is making calls, but it isn't tied to a registered agent — usually a key minted outside Jents, or a deleted agent's leftover key still running.
The fix: Register as agent to bring it into the Hub so its calls attribute — or Revoke key to block it at the gateway if it shouldn't be running at all.
CLI/IDE (captured)
🟣 New tool
A tool is sending telemetry that Jents doesn't recognize — a shadow or homegrown app you haven't added yet.
The fix: Register tool to add it to your inventory so its usage and cost are tracked. (Or dismiss it if it's expected.)
🔵 Unmapped person
A known tool, used by someone who isn't in your People directory yet — for example a new contractor on Claude Code.
The fix: Add to People so their usage attributes to them.
🟠 Identity mismatch
A real teammate — but their telemetry reports a non-work identity (an OS username like dana-mbp, or a personal email), so it reads as a stranger. This is the single most common source of noise.
The fix: Map identity — point that identity at the right person. Their usage folds into them, and it stops showing here for good.
⚪ No identity
Usage on a known tool with no user attached at all — often an automated process or a CI server running with no personal login.
The fix: Register as agent if it's an automated workflow, or dismiss it.
Quick reference
| Side | What you see | What it means | The fix |
|---|---|---|---|
| Agentic | Untracked key | A gateway key with no registered agent behind it | Register as agent · Revoke key |
| CLI/IDE | New tool | An unrecognized tool reporting telemetry | Register tool |
| CLI/IDE | Unmapped person | A known tool, person not in People yet | Add to People |
| CLI/IDE | Identity mismatch | A known person on a non-work identity | Map identity |
| CLI/IDE | No identity | Calls with no user attached | Register as agent |
Keep things out of Discovery
The cleanest way to never land here is to attribute calls at the source:
- Agents: route through a per-agent provisioned key, or pass a
jents_agent_idon every call — then the call ties to its agent automatically. - People: make sure everyone reports under their work email — the same one that's in your People directory — so their tool usage maps to them with no extra step.
Anything you genuinely don't want to track — Dismiss it. It won't come back unless you choose to surface it again.