The dashboard is a local operational surface over the SLM daemon. It helps an operator inspect and control a deployment; it does not replace CLI/MCP traces or make a health signal a guarantee of recall quality.
V3.8.0 reorganized and expanded the navigation. The former "Operations" section is now Governance, with dedicated sub-tabs for access control, data privacy, and audit. A new Integrations group was added for MCP and tool management.
| Workspace | Product surface | What to verify separately |
|---|---|---|
| Memories | memory browsing, filtering, update/delete and provenance-oriented inspection | write identity and scope policy for mutations |
| Knowledge Graph | graph neighborhoods, entities, scenes and relationships | slm trace for query-time graph participation |
| Entity Explorer | compiled entity summaries and timelines | source facts and entity recompilation behavior |
| Workspace | Product surface | What to verify separately |
|---|---|---|
| Brain | consolidation, behavioral patterns, feedback/outcomes, reward and soft-prompt state | mode/configuration and the underlying data tables |
| Skill Evolution | opt-in lineage, budgets, trigger and verification outcomes | the operator's evolution policy and model/provider boundary |
The former "Operations" pane is now "Governance". Sub-tabs cover all compliance and access controls in one place.
| Sub-tab | What it covers | What to verify separately |
|---|---|---|
| Access & Users | user list, role assignments (admin/member/viewer), workspace membership, login gate toggle | effective RBAC role on each API call; test with slm doctor |
| Data Privacy | PII redaction toggle, retention rule list (gdpr-30d / hipaa-7y / custom), GDPR export button, erasure with confirmation | erasure completeness — run slm diagnostics export after erasure |
| Audit | hash-chained audit trail viewer (store/recall/mutate/erase events), tamper detection status | slm evidence verify for bundle integrity |
| Lifecycle & Trust | ingestion-operation state and traces, maintenance work, consolidation triggers | a representative slm remember --sync receipt |
| Workspace | Product surface | What to verify separately |
|---|---|---|
| MCP & Tools | active MCP profile name and tool count, profile switcher (runtime, no restart), per-IDE MCP config snippets for 15 IDEs | daemon restart is not required for profile changes; verify with slm status |
| Cloud Backup | backup destinations, schedule status, last backup timestamp, manual trigger | actual backup file presence and restore path |
| Workspace | Product surface | What to verify separately |
|---|---|---|
| Dashboard | runtime summary, daemon identity, storage/runtime health, diagnostics and recent activity | slm status, slm doctor |
| Health | dependency, math/lifecycle and process health | slm health, deployment resource limits |
| Mesh Peers | configured peers, inbox/outbox, pending messages and locks | authenticated transport on the real peer topology |
| Workspace | Product surface | What to verify separately |
|---|---|---|
| Settings | mode, provider and product configuration | effective configuration and secret management |
| Optimize | cache/compression controls, cache state and savings telemetry | actual proxy/MCP routing and invalidation behavior |
Panels can be empty when a feature is disabled, no data has been produced, or the selected mode lacks a configured dependency. Recalled or displayed memory is evidence, not an instruction authority. SLM keeps dynamic memory out of high-trust IDE instruction files and renders it through the bounded context path at runtime.
The Access & Users tab requires require_login = true to be meaningful in
a multi-user deployment. In personal mode (single user, loopback owner), the
RBAC controls are visible but authentication is not enforced.
For a release or deployment witness, validate each enabled surface end to end:
slm doctor
slm health
slm remember "dashboard witness" --sync --json
slm trace "dashboard witness"For optional systems, also execute a real cache invalidation, peer authentication exchange, provider call, adapter import, or Scale Engine prepare/verify/promote/rollback cycle as applicable. Do not infer that these subsystems are active solely because their dashboard workspace is visible.