Frequently Asked Questions

Everything you need to know about Deva, security scanning, compliance, and deployment.

General

Deva is write-time security and compliance for AI-assisted development. It ships as the Deva CLI for macOS, Windows, and Linux and as the Deva Security extension for VS Code and other VS Code-compatible editors. Both run the same local DevSecCode Core engine to scan code, map findings to compliance controls, and generate AI-assisted fixes.

Deva has a free tier that includes standard local scans, dependency analysis, the rule catalog, and JSON output, with no account needed. Pro costs $9.99 per user per month and adds compliance mapping for 7 frameworks, signed evidence, SARIF and compliance exports, SBOM export, and AI-assisted validation and fixes with local or bring-your-own-key models. Eligible accounts can start a 14-day Pro trial without a card.

Air-Gapped and Local Operation

Partly. Static scanning runs on-device and needs no network connection or account. Deva has no dedicated air-gapped mode yet: dependency analysis uses OSV.dev or a local cache from earlier online runs, paid entitlements expire and need an online refresh or a vendor-signed entitlement file, and the Deva Coder model is pulled through Ollama rather than bundled with Deva. For air-gapped deployments, contact us.

No, by default. Scans run on your machine. AI features are optional: with a local model, nothing leaves the machine; with a cloud provider (Anthropic, OpenAI, Gemini or OpenRouter), the prompt and the code context for that request are sent to that provider. Dependency analysis looks up package versions in the OSV vulnerability database.

The extension supports your own Anthropic, OpenAI, Gemini or OpenRouter key, plus local models through Ollama, including Deva Coder. The CLI provides OpenRouter configuration with deva providers configure openrouter. Local inference stays on your machine; with a cloud provider, relevant code context goes to that provider, which bills usage separately. AI features require Pro or an eligible trial.

Yes. Scanning is rule-based and runs without any AI model, as do compliance mapping and exports. AI features (finding validation and fix generation) are optional.

Security Scanning

Deva uses a 3-phase scanning pipeline: Phase 1 runs static analysis via a Semgrep-compatible engine with YAML rules mapped to CWEs, including taint-mode rules with source/sink/sanitizer tracking. Phase 2 (optional) adds LLM-augmented analysis to confirm or reject findings and reduce false positives. Phase 3 adds compliance mappings on Pro and data from the bundled advisory catalog. OSV.dev lookups are part of the separate dependency analysis.

Deva's rules cover 14 programming languages, including Python, JavaScript, TypeScript, Go, Java, Ruby, and Rust, plus configuration formats (YAML, JSON, Terraform, Dockerfile, Kubernetes manifests). The scanning engine supports AST-based parsing for each language, enabling taint tracking and data flow analysis rather than regex-only pattern matching.

Deva ships with security rules mapped to CWE categories. The rulepack includes taint-mode rules (tracking data flow from input sources to dangerous sinks) and search-mode rules (pattern matching for dangerous configurations and API usage). Rules are updated with each release.

Yes. Deva includes 4 sensitive-data classifiers: PII (personally identifiable information), PHI (protected health information), PAN (payment card numbers), and CUI (controlled unclassified information). Each classifier uses context-aware heuristics to reduce false positives: a variable named “test_ssn” in a test file is treated differently from a hardcoded SSN in production code.

Deva’s taint-mode rules have significantly lower false positive rates than regex-only scanners because they track data flow through the program. The optional LLM-augmented Phase 2 further reduces false positives by having an AI model confirm or reject findings based on code context. Findings are labeled DEFINITE, LIKELY, or PROBABLY_FP so developers can prioritize effectively.

Yes. Deva uses YAML-based rulepack files compatible with the Semgrep rule format. You can add custom rules to the rulepack directory, and they will be included in subsequent scans.

Agent policy enforcement

Pro includes agent policy enforcement with the supported Claude Code hook integration. The CLI compiles rules files such as AGENTS.md and CLAUDE.md, asks you to review and confirm policy, and installs the hook when you run deva guard install --agent claude-code. Decisions allow, deny or request approval, with a local audit log. This requires setup; it does not intercept every agent or tool automatically. Run deva guard status to inspect integration health and limitations. The extension provides policy and activity views and waiting approvals.

Compliance

Deva offers 7 compliance frameworks on Pro. NIST SP 800-53 Rev 5, PCI DSS v4.0, SOC 2, and OWASP Top 10 (2025) are shipped; FedRAMP Rev 5, ISO 27001:2022, and HIPAA are in preview. Each framework maps specific controls to CWE rules, covering the code-detectable subset of the framework.

When Deva finds a vulnerability (e.g., CWE-89 SQL injection), it maps that finding to every active compliance framework’s relevant controls. For HIPAA, CWE-89 maps to the Input Validation control under 164.312(a)(1). For PCI-DSS, it maps to Req 6.2.4. The mapping is bidirectional: you can also view compliance coverage by framework to see which controls have findings, which are passing, and which need attestation.

Deva exports scan reports in 6 formats: SARIF (Static Analysis Results Interchange Format), JUnit XML, agent-json, OSCAL Assessment Results (Open Security Controls Assessment Language), Markdown, and signed evidence packages. Scans can also write JSON directly, and compliance results export as OSCAL and POA&M. SARIF and OSCAL include compliance control metadata, making them suitable for audit evidence packages. Deva evaluates code-level technical controls only. These outputs support your assessment; they do not certify compliance or guarantee an assessor will accept a particular artifact.

Core supports creating, validating, importing, exporting and evaluating custom frameworks on Pro through its authenticated local API. The current CLI and editor extension do not provide a custom-framework authoring wizard. Custom authoring requires a Core integration; contact us to discuss setup.

Deva offers FedRAMP Rev 5 mapping in preview on Pro. It relates code-level findings to relevant controls and exports technical evidence for your own assessment. It does not establish complete control coverage, grant authorization or replace a 3PAO assessment. Scanning runs locally; optional cloud AI sends the selected code context to your provider.

Deva can help identify code-level security issues and export technical evidence, but CMMC framework evaluation and SPRS scoring are not currently offered. CMMC references in our educational guides describe how weaknesses relate to assessment topics, not an available Deva CMMC preset or a certification claim.

Deployment and Integration

Download the Deva CLI from the Deva website for macOS (Apple Silicon), Windows x64, or Linux x64, or install the Deva Security extension from the VS Code Marketplace or Open VSX Registry. Each CLI package includes the signed Core engine. The extension works in VS Code and VS Code-compatible editors such as Cursor and Windsurf.

Yes. The deva command-line tool can be used in CI/CD pipelines, pre-commit hooks, and automated workflows. It supports scanning presets, compliance frameworks, and report formats available to your plan. The CLI runs without a GUI and exits with appropriate codes for pipeline integration.

Yes. The deva CLI integrates into any CI/CD pipeline (GitHub Actions, GitLab CI, Jenkins, Azure DevOps). Run deva scan . --format json --output deva.json to produce scan reports, and set --fail-on high to gate deployments on finding severity. With Pro, add --compliance <frameworks> for compliance-aware scans and use --format sarif for SARIF output.

Yes. Deva Security is itself an extension, so it runs alongside your existing extensions, themes, keybindings, and settings in VS Code and VS Code-compatible editors. The CLI works independently of any editor.

Privacy and Security

Deva has no telemetry. No usage analytics, crash reports, or feature flags phone home. Standard scans make no network calls. Other calls happen only for features you use: sign-in and plan checks if you have an account, cloud AI requests if you choose a cloud model, and OSV.dev and package registry lookups for dependency analysis. The installed extension includes its signed Core engine. Network access is also used for extension updates and local models you choose to download.

Scans and compliance mapping run on-device, with no telemetry. With a local model such as Deva Coder, AI inference stays on the machine. Deva has not been accredited for classified use and has no dedicated air-gapped mode yet. For air-gapped deployments, contact us.

Deva is built by DevSecCode, a security-focused development tools company. The team includes security researchers and software engineers focused on making compliance-aware security tooling accessible to development teams in regulated industries.