Audit & Telemetry
Fleet provides two distinct categories of observability data: audit logs (what happened in your org) and telemetry (optional usage signals from your Harness). This page explains both, who can see what, and how consent works.
Audit log
The audit log records actions taken within your Fleet organization — member invitations, role changes, policy updates, SSO configuration, and similar administrative events. It is not a record of your AI conversations. Conversation content reaches Fleet only through the features that carry it: context and memory sync, team session handoff, and the audit archive below.
Who can access it
| Role | Access |
|---|---|
| org_owner | Read and export |
| org_admin | Read and export |
| org_member | No access |
What's in it
The audit log captures: who performed an action, what the action was, when it happened, and the IP address or originating device. You can browse the log from the Fleet dashboard under Audit and export it as a download.
Audit archive (Team and Enterprise)
On Team and Enterprise plans, each signed-in Harness streams its local audit events to Fleet. Each batch is signed with the device's key, and each event carries the hash of the one before it (the same chain as the Harness audit log). Fleet records whether each batch's signature verified and whether the chain continued without a gap; admins see those results in the dashboard. Event payloads are redacted on the device before they are written, so the archive holds the redacted record. Members can see whether the archive is on in the Harness under Settings → Compliance.
Telemetry
Telemetry is entirely opt-in. Nothing is collected or sent from your Harness without your explicit consent.
Telemetry classes
Fleet tracks consent per category of Harness and Sigil data. Every class is off until you, or your org's policy, turn it on:
| Class | What it covers | Default |
|---|---|---|
harness.usage_counts | Session and turn counts | Off |
harness.tool_calls | Which tools were invoked (not their content) | Off |
harness.errors | Harness error reports | Off |
harness.diagnostics | Diagnostic information for support | Off |
sigil.heuristics | Heuristic signals from the Sigil component | Off |
sigil.predictions | Prediction metadata from Sigil | Off |
sigil.suggestions | Suggestion usage signals from Sigil | Off |
harness.errors carries stack traces and error codes, never your conversation content. Like every other class it is off until you turn it on.
Managing your own opt-ins
Every member can manage their own telemetry preferences from Account → Telemetry in the Fleet dashboard or Harness settings. Your choices are saved per-account and respected across every Harness you sign in to.
Org-level policy (Team and Enterprise)
On Team and Enterprise plans, org owners and admins can require telemetry for all members. When a class is required at the org level, individual members cannot opt out of it, and the source shown in their preferences changes to org_policy to make this visible.
Members can always see their current opt-in state and its source (user_self, org_policy, or install_default) in their account settings.
Analytics for admins (Team and Enterprise)
Admins on Team and Enterprise plans can access aggregated analytics in the Fleet dashboard:
- Usage — session counts, turn volume, model usage over time.
- Adoption — active users across the org, per team.
- Cost — estimated spend by provider and model.
Analytics are aggregated — the dashboard doesn't expose individual conversation logs.
Privacy posture
- No conversation content in telemetry. Telemetry carries counts, timings and error codes, not prompt or response text. Model calls go from the Harness to your AI provider, not through Fleet's servers. Content reaches Fleet only through sync, session handoff and the audit archive, described above.
- No silent collection. Every telemetry class is opt-in.
- Transparent policy. If an org admin requires a telemetry class, members can see that it's org-required rather than a personal choice.
- Data stays in your org. Audit log and analytics data belong to your organization. Staff at Kameas can access it only as explicitly documented in your subscription agreement.
Next: Emergency Lockdown — the break-glass tool for lost or compromised devices (Team and Enterprise).