Dashboard

Advanced views#

Ten tabs behind the Advanced disclosure. Each answers a narrower question than the top-level pages, and several only apply to some runtimes.

Schedules#

Cron jobs with status, schedule expression, last and next run, and duration. Full CRUD through the gateway for runtimes that support it — OpenClaw, NemoClaw and Hermes — plus Cline's own schedule table with per-run tokens and cost.

curl -s localhost:8900/api/crons -X POST -d '{}' -H 'content-type: application/json' | jq
curl -s localhost:8900/api/cron-health | jq
curl -s localhost:8900/api/cron-run-log -X POST -d '{}' -H 'content-type: application/json' | jq

Cron management is one of the four surfaces that can affect an agent — see How it works. Creating, editing or triggering a job writes through the gateway RPC, so it is a real change to the agent's schedule.

Memory#

The persistent files an agent reads on boot — the memory documents, agent instructions and daily notes. Browse them, see what changed and when, and see which files the agent actually read during a session.

Memory is per runtime, not per workspace: each runtime has its own memory convention and its own directory, and merging them would produce a file list that belongs to nobody.

curl -s localhost:8900/api/runtimes/memory-catalog | jq
curl -s 'localhost:8900/api/runtimes/<runtime_id>/files' | jq
curl -s localhost:8900/api/memory-access | jq

Notifications#

What ClawMetry has told you and through which channel, so a missed alert is diagnosable rather than mysterious.

Logs#

The live log stream, filtered and searchable. Primarily OpenClaw's structured logs, plus any runtime that writes a log ClawMetry can follow.

curl -s 'localhost:8900/api/logs?limit=200' | jq

Security#

The node's security posture: credential exposure scanning, policy events, integrity verification status, threat signatures and an audit trail.

curl -s localhost:8900/api/security/posture | jq
curl -s localhost:8900/api/security/credential-scan | jq
curl -s localhost:8900/api/security/policy-events | jq
curl -s localhost:8900/api/security/integrity | jq

Posture grading is honest about what it can and cannot verify — a closed-source component is graded on what is observable from outside, not assumed compliant. → Hardening

Tool permissions#

Which tools each agent may run, where they run, and what was approved or blocked. This is the pre-tool axis, and it is deliberately separate from Guard policies: different table, different timing, no shared state.

curl -s localhost:8900/api/tool-policy -X POST -d '{}' -H 'content-type: application/json' | jq

→ Tool permissions

Skills#

Discovered skills across every runtime that has a skills convention, with usage counts and fidelity — how often a skill was invoked versus how often its guidance was actually followed.

curl -s localhost:8900/api/skills | jq
curl -s localhost:8900/api/skills/fidelity | jq

Self-Evolve#

Findings the system has produced about its own operation, and — where enabled — proposed fixes. Off by default; CLAWMETRY_SELFEVOLVE_AUTO controls whether anything is attempted automatically.

curl -s localhost:8900/api/selfevolve/status -X POST -d '{}' -H 'content-type: application/json' | jq

Version impact#

What changed when a runtime version changed. Agent runtimes ship fast, and a regression in cost, latency or success rate that lines up with an upgrade is worth seeing as a correlation rather than discovering as a mystery.

curl -s localhost:8900/api/version-health | jq

NemoClaw#

Appears only on a node with NemoClaw present. NeMo Guardrails policy state, the rules in effect, and the approval queue for actions the guardrails held.

curl -s localhost:8900/api/nemoclaw/status -X POST -d '{}' -H 'content-type: application/json' | jq
curl -s localhost:8900/api/nemoclaw/pending-approvals | jq

nemo_governance is a free feature and works on any runtime; NemoClaw is what makes it interesting.