DocsBook a demo

Platform

Platform overviewArchitectureWorkflow orchestrationGitOps configurationGovernance and AAAAI and MCPKnowledge and contextRuntime and executionEvidence and monitoring

Use cases

All use casesProduction incidentRelease preparationHotfix to productionSecurity scan triage
Why NopsAIIntegrationsSecurity

Resources

All resourcesAI agent governanceMCP governanceMCP securitySelf-hosted platforms
PricingGitHub

Company

How a run worksAboutContactBook a demo

Storage and persistence

What is durable, what is per-run, and what the default topology does not persist.

ConceptAdministratorOperator

Key points

  • PostgreSQL holds durable execution, audit, credential, monitoring, and setup state.
  • Docker runners create named Docker volumes; Kubernetes runners create an agent-owned PVC-backed workspace per run.
  • Step volumes reuse existing storage and create it when missing, in volume:/mount/path form.
  • The reserved runtime output path /nopsai/outputs cannot be used as a volume mount target.
  • Backups are gzip-compressed JSON Lines files under DATA_BACKUP_DIR, defaulting to /data/backups.

How it works

The default Compose topology does not mount the backup directory as durable storage, so production deployments must mount or export it. This is the single most common persistence surprise.

Docker step containers leave image-provided temporary directories such as /tmp and /var/tmp unchanged, so images that rely on them behave normally.

Examples

See what holds state on a Compose installbash
# the only durable volume the control plane needs
docker volume ls --filter name=db

# volumes created by pipeline steps, which are caches rather than state
docker volume ls --filter label=nopsai.io/runner-id
Result

The database volume is the one that must be backed up. Step volumes can be recreated by re-running the work.

Limits

Current behavior
  • There is no S3 or generic object-storage artifact or backup backend.
  • Backup creation, listing, checksum metadata, download, and cleanup exist; a verified product-managed restore workflow does not.

Implementation evidence

  • config/config.go

    DATA_BACKUP_DIR and runtime output limits.

  • doc/kubernetes-runner.md

    Agent-owned PVC workspace behavior.