We protect customers the way we hope they protect their users.
LaunchGuard Solana is Solana-first, Web2-ready — and a security product first. The bar isn't "industry standard," it's "the principles we wish every Web3 / Web2 SaaS would follow." This page describes the controls we enforce at the system level.
Core trust pillars
Read-only by default
All scans start in read-only mode. We never modify your code, deploy anything, or interact with your production runtime without explicit authorization.
No raw secrets stored
Secret scanners (TruffleHog and the LaunchGuard core engine) only persist secret type, file path, and a salted fingerprint. The actual secret value never reaches our database.
No model training on your code
AI prompts are scoped to a single scan. We do not train any base or fine-tuned model on customer source. DeepSec can run fully locally for the most sensitive repos.
Shannon requires explicit authorization
Active validation only runs after you tick every consent box and provide an approved deployed URL. Shannon never tests third-party domains, brute forces, or DDoSes.
GitHub App, least privilege
Our GitHub App requests the minimum scopes needed for each tier. Tokens are encrypted at rest and never exposed to the browser.
Fix Mode is opt-in and reversible
Fix Mode can create branches and pull requests but never auto-merges and never pushes to your default branch. You always see the diff before it leaves your repo.
Token encryption
Integration tokens (GitHub, providers, webhooks) are encrypted using AES-256-GCM with keys held outside the database. The frontend cannot select ciphertext columns.
Data deletion
Workspace owners can delete projects, scans, and reports at any time. Deleted artifacts are removed from primary storage immediately and from backups within 30 days.
Audit logs
All sensitive actions — Shannon authorization, PR creation, finding status changes, integration changes, report exports — are recorded in append-only audit logs.
Operational policies
Data retention
Scan metadata and findings are retained for the lifetime of the workspace. Source code excerpts persist only as redacted snippets sized for context, never full files.
Subprocessors
Vercel (hosting), Supabase (Postgres + auth), and the AI provider you configure (Anthropic / OpenAI / Gemini). All others are opt-in.
Responsible disclosure
Found a security issue in LaunchGuard itself? Email security@launchguard.dev (placeholder). We commit to acknowledging within 48 hours and patching critical issues within 7 days.
Private CI mode
For regulated teams, we are building a self-hosted runner that performs every scan inside your network. Findings sync back to your LaunchGuard workspace as pre-redacted summaries, never raw code.