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

Security Agent

**Security Agent** is DeplAI’s end-to-end application security workflow: scan a connected project, review grouped findings, configure an AI remediation agent, approve proposed fixes, and—when you use GitHub—open a pull request and verify the result with a follow-up scan.

Security Agent is DeplAI’s end-to-end application security workflow: scan a connected project, review grouped findings, configure an AI remediation agent, approve proposed fixes, and—when you use GitHub—open a pull request and verify the result with a follow-up scan.

Open it from Services → Security Agent after selecting a project in the workspace nav. The URL is /dashboard/security-analysis/{projectId}.

DeplAI is not a black-box auto-patcher. Every change to source code waits for your Review approval. Pull requests use the GitHub App, not your OAuth login token.

---

At a glance

QuestionAnswer
What does it scan?Source (SAST), dependencies (SCA/SBOM), secrets, IaC, containers, Kubernetes, CI/CD, APIs, optional DAST, and cloud posture when available.
What does it fix?Proposes source patches for selected findings; you approve before anything is written.
What does it need?A GitHub project or ZIP upload, completed scan findings, and an available DeplAI platform remediation model. No BYOK key is used for remediation.
What is the output?Grouped findings, proposed diffs, optional GitHub PR, PDF report, and a Session record.

---

Pipeline overview

The stage rail is labeled Pipeline. The header shows progress such as 01 / 06 · Scan. Later stages stay locked until the previous stage completes.

StageUI labelWhat goes inWhat you get
01ScanProject + selected modulesRaw findings from all enabled scanners
02ResultsFinished scanKPIs, grouped findings, export, remediation entry
03Agent setupPlan + model accessConfigured AI agent (page title: Configure AI Agent)
04RemediationSelected findings + modelProposed diffs only—nothing written to GitHub yet
05ReviewProposed diffsApprove, reject, or copy patches manually
06GitHub & verifyApproved diffsPR opened/updated + verification rescan

---

Stage 1 — Scan

Quick-start presets

When you launch a scan, choose a preset or customize modules:

PresetWhat runs
SASTStatic application security testing—injection, XSS, and similar code patterns.
SCADependency inventory and CVE matching (supply chain).
Full Scan (recommended)SAST + SCA + secrets + infrastructure checks when matching files exist. DAST is configured separately.

Pipeline modules

On the Scan stage you can enable individual modules:

ModuleWhat it checksWhen it is skipped
SASTStatic code analysisNo scannable source for the configured languages
SCAKnown vulnerabilities in dependenciesNo lockfiles or manifests detected
SBOMSoftware bill of materials inventorySame as SCA prerequisites
SecretsHard-coded credentials and tokensNo matching paths
IaCInfrastructure-as-code misconfigurationsNo Terraform/CloudFormation/etc. files
ContainersContainer image and Dockerfile issuesNo container artifacts
KubernetesWorkload and manifest securityNo Kubernetes manifests
CI/CDPipeline and workflow securityNo CI config files
API SecurityOpenAPI/Swagger and API surface risksNo API specs detected
DASTDynamic tests against a verified HTTP targetNo verified target linked—see DAST
CloudLive AWS posture (Results tab)Shown after deploy when cloud context exists

Skipped modules display a reason in the scan output—they are not treated as failures.

Severity model

Findings use four severities: critical, high, medium, and low.

Remediation’s default major scope includes critical and high only. You can widen scope when starting remediation.

Running the scan

  1. 1Select your project in the workspace nav.
  2. 2Open Security Agent.
  3. 3Choose presets or enable modules on Scan.
  4. 4If using DAST, select a verified target (configured under Services → DAST).
  5. 5Click Run first scan (or equivalent run control).
  6. 6Keep the browser tab open while the scan runs. When it completes, the pipeline advances to Results.

---

Stage 2 — Results

Results surfaces

The Results stage organizes evidence for triage:

SurfaceContents
OverviewSummary KPIs and entry to remediation
FindingsPrimary vulnerability list with filters
SecretsExposed credentials and sensitive values
Supply ChainPackages, CVEs, fix versions
InfrastructureIaC and cloud configuration issues
CloudPost-deploy AWS posture when available
APIsAPI specification findings
Dynamic TestingDAST results from verified targets
RiskAggregated risk view
AssetsDiscovered asset inventory
Attack PathsChained risk visualization

Saved views

Filter quickly with built-in views such as All open, Critical, Secrets, Exploitable, and Infrastructure.

KPIs

The header shows counts for Critical, High, Medium, Low, and Auto-fixable findings.

How findings are grouped

CategoryGrouped byExample
Code security (SAST)CWE identifierCWE-79 (cross-site scripting) across files
Supply chain (SCA)CVE + packageCVE-2021-23337 on lodash@4.17.20
DASTRuntime check + endpointMisconfiguration on a verified URL

Counts reflect occurrences, not necessarily distinct root causes. Remediation groups the same root cause into one work item.

Export

Use Download PDF on Results to export a security report for stakeholders or compliance records.

Continue to remediation

When findings exist, an AI Auto-Remediation Available banner appears. Continue to Agent setup when you are ready to generate patches—not before you have reviewed what matters.

---

Stage 3 — Agent setup

This stage configures which model proposes fixes. It does not run another scan.

The page heading is Configure AI Agent; the pipeline rail still shows Agent setup.

Remediation model policy

Security remediation always uses DeplAI's platform OpenRouter route and an eligible free coding model. The model list is filtered for zero-priced, active coding variants that meet the workflow's context and output requirements.

Allowed for remediationNot used for remediation
An eligible free platform OpenRouter coding modelYour BYOK key or provider account
A replacement eligible free model when the saved choice is unavailablePaid OpenRouter models or paid-model opt-in
DeplAI-managed request limits and temporary cooldown handlingDirect provider SDKs, local models, or worker-held provider keys

This is intentionally narrower than the general BYOK and Compare features. Saving a provider key, changing profile routing, upgrading a plan, or selecting a paid model elsewhere does not change the remediation route.

Choose a model

  1. 1Select an available free remediation model from the list in Configure AI Agent.
  2. 2Start remediation. DeplAI checks availability and request capacity before generation.
  3. 3If the selected model became unavailable, DeplAI may use another eligible free remediation model. If none is available, wait and retry from the same stage.

The picker does not expose a BYOK, Auto, or paid-model remediation option. You do not need to provide an OpenRouter key.

GitHub PAT (optional)

GitHub PAT (Optional) on this screen is used only for pushing the fix branch on this run. It is not stored in the BYOK vault and is not your persistent GitHub login.

Use it when the GitHub App alone cannot push to the branch you need. Pull request creation still flows through the App after Review.

Organization context

If your project belongs to an organization, org Security policy may require certain scan types before downstream deploy steps. Security Agent itself does not bypass policy—you still approve every fix in Review.

---

Stage 4 — Remediation

Remediation turns selected findings into proposed diffs. Nothing is written to GitHub on this stage.

What happens

  1. 1Filters findings by the scope you chose (major = critical + high by default).
  2. 2Groups the same root cause across files into one work item.
  3. 3Sends grouped items to the configured model.
  4. 4Validates that diffs stay inside the project and address the requested severities.
  5. 5Presents patches for your decision: run another round or proceed to Review.

States you may see

StateMeaning
Awaiting your decisionPatches ready—choose another round or continue
Awaiting final approvalTransitioning toward Review
Remediation FailedModel error, access issue, or validation failure—see troubleshooting

Availability and usage

Remediation uses the platform's free-model route. It does not debit a BYOK provider account and does not send a remediation request with a key saved in BYOK → Keys. General platform/BYOK usage views may still show other product activity; they are not a way to select or fund a different remediation provider.

---

Stage 5 — Review

Review is the human gate. DeplAI does not persist fixes or open pull requests until you approve.

Your options

ActionResult
ApproveUnlocks GitHub & verify (GitHub projects) or saves diffs (ZIP projects)
RejectStops the pipeline; you can copy diffs manually if useful
Copy diffApply patches outside DeplAI

Inspect every hunk. Security Agent proposes fixes; you remain accountable for merged code.

---

Stage 6 — GitHub & verify

For GitHub-connected projects:

  1. 1DeplAI opens or updates a pull request with approved files.
  2. 2The PR uses the GitHub App installation you granted—not your OAuth token.
  3. 3DeplAI re-runs static and dependency scanners on the result.
  4. 4When clean, you may Continue to deployment from the security flow.

For ZIP uploads:

  • Approved diffs are saved on the upload record.
  • No pull request is created—connect GitHub if you need PR-based workflow.

Verification rescan

The follow-up scan confirms that critical/high items from the remediation scope are addressed. If issues remain, the UI tells you before you treat the run as complete.

---

Integrations

GitHub

StepIntegration
Sign inIdentity only (email, profile, org membership)
Repository accessGitHub App install on Account → Integrations or Your Profile → Integrations
Pull requestsGitHub App after Review approval
Optional pushSingle-run GitHub PAT on Agent setup

DAST

Dynamic testing never uses a free-form URL in Security Agent.

  1. 1Add and verify targets under Services → DAST (/dashboard/dast).
  2. 2On Scan, enable DAST and select the verified target.
  3. 3Findings appear under Dynamic Testing in Results.

Full workflow: DAST.

Organizations

Organization Security policy can require SAST, SCA, container scan, or DAST evidence before production deploys. Security Agent supplies the scan evidence; Deploy enforces policy at apply time.

Details: Organizations.

Deploy

After a clean GitHub & verify pass, use Continue to deployment to move into the Deploy pipeline with security context carried forward.

Details: Deploy.

BYOK and other model features

BYOK → Keys, Catalog, Compare, and profile routing are available for the product features that support them. They do not apply to Security Agent remediation, which remains on the platform free-model route.

Details: BYOK models · Security and data.

---

Sessions

Every Security Agent run creates a Session under Services → Sessions (/dashboard/sessions).

FieldUse
Statusqueued, running, completed, failed, needs review
ServiceSecurity Agent
LogsScanner output and pipeline events

Important: Reopening a session shows the pipeline rail as history. It does not resume a live scan. Start a new scan from Security Agent to continue interactive work.

Copy the session id when contacting support.

Details: Sessions.

---

Troubleshooting

Scan stage

SymptomLikely causeWhat to do
Scan Error / Scan FailedScanner timeout, repo access, or tool failureRead the scanner log; retry with fewer modules
Project Access RequiredGitHub App not installed or repo not grantedIntegrations → reinstall or add repository
No scan results yetScan not started or still runningEnable modules → Run first scan; keep tab open
Module shows SKIPPEDNo matching files or module not selectedExpected—enable relevant modules or add files
DAST skippedNo verified targetOpen DAST → verify host → select on Scan
Progress lost after closing tabLive scan is tab-scopedCheck Sessions for logs; start new scan

Results stage

SymptomLikely causeWhat to do
Results UnavailableTransient load or incomplete scanRetry Loading Results
Empty findings but scan “completed”Clean repo or modules skippedConfirm modules; check skipped reasons
Unauthorized target (DAST)Verification expired or wrong hostnameRe-verify in DAST

Agent setup & remediation

SymptomLikely causeWhat to do
Cannot start remediationNo findings or scan incompleteReturn to Results
No remediation model availableNo eligible free platform model is currently available or the selected model is staleRefresh the model list or retry later; do not add a BYOK key for this workflow
Remediation FailedPlatform availability, malformed model output, or patch validation failureRetry from the stage; select another eligible free model when offered; review the failure detail
BYOK call failedInvalid or revoked keyBYOK → Keys → validate; check provider quota

Review & GitHub

SymptomLikely causeWhat to do
PR not createdZIP project or approval missingUse GitHub project; complete Review
Push failedBranch protection or missing PATAdd optional GitHub PAT on Agent setup
Verify still shows critical/highFix incomplete or new issuesReview PR diff; run another remediation round

Organization policy

SymptomLikely causeWhat to do
Deploy blocked after security passOrg policy requires additional scan typesRun required modules; check Organizations → Security policy
DAST required but missingPolicy on, no verified targetComplete DAST setup

---

Best practices

  1. 1Run Full Scan on first connect, then tune modules per repo type.
  2. 2Verify staging URLs in DAST before enabling dynamic tests on production.
  3. 3Use major scope for first remediation pass; widen only when needed.
  4. 4Always Review diffs—treat AI patches like any other contributor’s PR.
  5. 5Select an eligible free remediation model and retry later if the platform reports temporary availability limits; BYOK and paid-model settings cannot bypass this policy.
  6. 6Export PDF from Results for audit trails before remediation changes the picture.
  7. 7Link GitHub early if you want PR-based workflow; ZIP is fine for evaluation only.

---

Related documentation

  • DAST — verify targets and scan profiles
  • Deploy — infrastructure after security sign-off
  • Organizations — roles and security policy
  • BYOK models — model catalog and recommendations
  • Sessions — run history and logs
  • Security and data — GitHub scopes and key handling
  • Glossary and FAQ — quick answers

On this page

At a glancePipeline overviewStage 1 — ScanQuick-start presetsPipeline modulesSeverity modelRunning the scanStage 2 — ResultsResults surfacesSaved viewsKPIsHow findings are groupedExportContinue to remediationStage 3 — Agent setupRemediation model policyChoose a modelGitHub PAT (optional)Organization contextStage 4 — RemediationWhat happensStates you may seeAvailability and usageStage 5 — ReviewYour optionsStage 6 — GitHub & verifyVerification rescanIntegrationsGitHubDASTOrganizationsDeployBYOK and other model featuresSessionsTroubleshootingScan stageResults stageAgent setup & remediationReview & GitHubOrganization policyBest practicesRelated documentation