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

Help and reference

How DeplAI fits together

DeplAI separates browser-facing control from long-running execution. The Connector owns identity and authorization; execution services receive project-scoped work only through controlled server routes.

DeplAI separates browser-facing control from long-running execution. The Connector owns identity and authorization; execution services receive project-scoped work only through controlled server routes.

Platform map

Layers and contracts

LayerResponsibilityInputOutput and stateSecurity boundary
Connector control planeUI, GitHub login, session, project and organization authorization, API facadeBrowser requestsMySQL records, scoped upstream requests, UI responsesEncrypted session or API token; ownership and permission checks
Project systemResolve GitHub or ZIP source into a project workspaceAuthorized project idRepository metadata and server-side pathBrowser cannot choose arbitrary filesystem paths
Repository intelligenceCollect bounded source evidenceProject workspaceTyped repository context and summaryRead-oriented analysis with file-size and exclusion rules
Agentic LayerRun scans, remediation, architecture, Terraform, and deployment workService-authenticated project requestEvents, reports, plans, apply resultsInternal service key and project-bound WebSocket token
Customization serviceTurn a design conversation into scoped frontend changes and validationAuthorized repository context plus manifestDiffs, snapshots, quality results, previewReached through Connector proxy; frontend-only mode is default
Security engineExecute scanners and normalize findingsProject source plus selected modulesReports, findings, skip reasonsContainerized tools; DAST and cloud need separate authorization or credentials
Deployment plannerConvert evidence and user answers into architecture and cost decisionsRepository context, environment, requirements, budgetTyped profile, alternatives, diagram, estimateAdvisory output has no apply authority
Terraform engineGenerate, validate, plan, and apply supported AWS configurationsApproved context and AWS credentialsTerraform bundle, plan, outputs, state, eventsPlan confirmation, workspace locking, timeouts, redaction
Model gatewayResolve model, credential, routing, policy, and usageTask, model or alias, access modeProvider response plus usage and normalized errorsProvider keys stay server-side and are masked in UI
Observability recordsPreserve run summaries, logs, events, findings, and deployment recordsWorkflow updatesSessions and subsystem recordsTenant-scoped reads and sanitized payloads

Request path

A browser action does not call an execution container directly.

Live security and deployment progress may use a short-lived token bound to the user and project. This token is distinct from the server-to-server key and cannot be reused as general API authority.

Control plane versus execution plane

The control plane answers who may do what to which project. The execution plane answers how a permitted workflow performs its bounded work. Keeping those decisions separate prevents a scanner, model, or Terraform process from becoming the identity system.

The execution plane can be powerful. It runs containerized tools and, in the deployment path, can call AWS. Production operators must keep it on a private network and treat access to the container runtime as privileged.

Data and durability

Durable records include users, organizations, projects, GitHub installations, sessions and logs, billing data, AI credentials and usage, DAST assets and scans, and deployment execution records. Repository workspaces, reports, and Terraform artifacts can also persist on configured volumes or state stores.

Some live orchestration context remains process-bound. A service restart can disconnect WebSockets or end an in-flight run even when the session row and generated files survive. Consult the service status and artifacts before retrying.

Deployment topology

Local development can expose Connector and execution services on local ports. Production uses a reverse proxy as the public edge; Connector, Agentic Layer, customization, database, and supporting services stay private. Browser WebSockets use the same public origin and a restricted route.

Boundaries that remain external

GitHub remains the source-of-truth for repository history. AWS remains the source-of-truth for cloud resources and charges. Model providers remain the source-of-truth for provider quotas and BYOK billing. DeplAI coordinates these systems but cannot override their permissions, outages, or retention policies.

Related: Repository intelligence | Agents and workflows | Security and data | Production operations

On this page

Platform mapLayers and contractsRequest pathControl plane versus execution planeData and durabilityDeployment topologyBoundaries that remain external