DeplAI
— credits
DeplAI

Application

Dashboard

HomeYour ProfileOrganizationsUsageDocumentation

Services

UI/UX customizerSecurity AgentDASTCloudDeployInstance ManagementCode ReviewerSoonSessions

BYOK

KeysCatalogCompareUsage

Account

BillingInvoicesCreditsRefer & EarnNEWIntegrations
Settings
DE

DeplAI

Free

— creditsFree
deplaiDocumentation

Start here

Services

Account and models

Help and reference

Services

Artifacts, state, and recovery

DeplAI workflows produce artifacts with different owners, lifetimes, and recovery guarantees. Knowing which evidence is durable prevents a live status indicator from being mistaken for the system of record.

DeplAI workflows produce artifacts with different owners, lifetimes, and recovery guarantees. Knowing which evidence is durable prevents a live status indicator from being mistaken for the system of record.

Artifact map

ArtifactProducerTypical storage or handoffReview concern
Repository contextRepository analyzerProject/run resultStaleness and conflicting evidence
Scan reports and findingsSecurity engineReport volumes and API resultScanner coverage, skipped modules, false positives
SBOMSyft-based scan pathSecurity report artifactPackage completeness and source revision
Remediation diffRemediation workflowRun state, review UI, optional GitHub PRCorrectness, scope, tests, unsafe behavior changes
Customization manifestConversation workflowCustomization runWhether intent and frontend boundary are accurate
Snapshot and previewCustomization serviceRepository snapshot and preview processPreview is not production validation
Architecture decisionReview and advisorProject deployment stateUser assumptions, budget, reliability, data needs
Diagram and estimatePlanning workflowStage review payloadEstimate freshness and excluded usage costs
Terraform bundle and planTerraform engineRun workspace and optional remote artifact/stateProvider scope, destructive changes, drift
Deployment recordDeployment executorMySQL plus execution service stateArtifact identity, environment, health result
Session and logsConnectorMySQLSummary may outlive live worker context

Four kinds of state

Durable product state

Users, organizations, memberships, projects, GitHub installations, sessions, AI configuration and usage, billing records, DAST assets, and deployment execution records live in MySQL. These records are tenant-scoped and survive normal process restarts.

Workspace artifacts

Repository clones, uploaded projects, scan reports, customization snapshots, and Terraform files live in configured workspaces or volumes. Their durability depends on deployment configuration, volume retention, and cleanup policy.

Live execution state

WebSocket subscribers, some orchestration dictionaries, preview processes, and in-flight tasks can be process-local. If the worker restarts, the durable session may remain while live execution is gone.

External state

GitHub pull requests, AWS resources, Terraform remote state, Secrets Manager entries, and provider usage belong to external systems. DeplAI records references and results, but those systems remain authoritative.

Session lifecycle

The common session statuses are deliberately small. Service-specific stages provide more detail. Deployment execution also uses statuses such as CREATED, COMPLETED, FAILED, CANCELLED, and ROLLED_BACK in its own record.

Retry, resume, restart, and rollback

ActionUse whenImportant condition
RetryA transient stage failedInputs and source revision are still valid
ResumeThe subsystem exposes resumable state for an interrupted executionA compatible durable checkpoint or execution record exists
RestartContext, source, credentials, or decisions changedCreate a fresh run and preserve old evidence for comparison
RollbackA deployment artifact must return to a recorded prior stateThe executor has a valid rollback target; infrastructure rollback is not automatic application recovery

Never use rollback as a substitute for understanding database migrations or external side effects. A previous application image cannot necessarily reverse data changes.

Recovery checklist

  1. 1Record the project, session, run, and deployment identifiers.
  2. 2Read the last durable event or log, not only the browser toast.
  3. 3Determine whether the worker is still running.
  4. 4Verify that source revision, credentials, and environment are unchanged.
  5. 5Check whether an artifact was partially created externally.
  6. 6Retry only the supported failed action; otherwise start a new run.
  7. 7For deployment, inspect Terraform state, bootstrap status, and HTTP health independently.

Production retention

Back up MySQL and any persistent artifact volumes required by your recovery objective. Configure and protect Terraform remote state. Retain GitHub and AWS audit evidence according to your organization policy. Do not assume session logs are a complete compliance archive.

Related: Sessions | Deployment | Production operations | Troubleshooting

On this page

Artifact mapFour kinds of stateDurable product stateWorkspace artifactsLive execution stateExternal stateSession lifecycleRetry, resume, restart, and rollbackRecovery checklistProduction retention