Operations guide
Operating a Private AI Workspace
Run service health, queues, storage, model capacity, audit history, governance, diagnostics, recovery, and promotion as one operating practice.
Key considerations
Observe
Health, dependencies, queues, storage, capacity, and failures.
Govern
Users, roles, capabilities, keys, configuration, and audit history.
Recover
Predefined safe actions, bounded retries, backups, and rollback.
Architecture at a glance
Operate the whole dependency chain
Application readiness depends on more than the web process. Database, Valkey, object storage, workers, model endpoints, media services, proxy behavior, and capacity all contribute to user-visible health.
Treat queues as product state
Scheduled tasks and media jobs require queue depth, age, failure counts, retry behavior, and output history. Stale work should be recovered through bounded operator actions rather than hidden deletion.
Keep diagnostics useful and safe
Logs and incident evidence should preserve request and service context while redacting secrets and limiting the amount of data exposed to an administrator or model-assisted diagnostic workflow.
Make governance routine
Review users, roles, keys, enabled product capabilities, storage policy, runtime configuration, and audit history on a defined schedule.
Promote releases with evidence
A release should carry test results, visual signoff, dependency checks, known risks, rollback notes, and owner approval. Production promotion is an operating decision, not only a build result.
Private deployment consultation
Review Pulsar against your environment.
Bring the infrastructure, security boundaries, model runners, and use cases. The Pulsar team will map the appropriate deployment path.