Principles
Four invariants the codebase enforces — not aspirations.
Read-only repository access
GitHub OAuth scopes are limited to repo read. Attestly never holds write, admin, or workflow scopes. We cannot push commits, open PRs, or modify settings on your repos.
Ephemeral scan environment
Each scan downloads a tarball of the requested ref into an isolated worker, runs static analysis, persists only structured findings (file paths and line numbers — not source), and discards the working copy.
Tenant isolation at the row level
Every row in the database is scoped by tenant_id. Server actions and API handlers resolve the caller's tenant from the verified Clerk session before any read or write.
No free-form LLM prose in published documents
All document generation uses strictly typed schemas. The model fills structured fields; the renderer produces deterministic Markdown. Prompt-injection in your repo cannot exfiltrate text into a published policy.
Data we store
The shortest list of data that lets the product function. Nothing more.
- GitHub OAuth access token (encrypted at rest, scoped read-only)
- Repository metadata (owner, name, default branch, last scanned commit)
- Structured scan findings (kind, source path, line, optional snippet)
- Generated document versions (typed JSON + rendered Markdown)
- Tenant facts you provided in onboarding (legal entity, jurisdiction, contact email)
- Stripe customer / subscription IDs (no card data — Stripe holds it)
Data we never store
If a fact is missing, generators emit “Customer-defined” rather than guess.
- Cloned source code after a scan completes
- Customer end-user PII from your application database
- Production traffic, logs, or telemetry from your services
- Any data outside the repositories you explicitly connect
- Payment card numbers, CVCs, or bank credentials
Compliance copilot & Trust Center Q&A
The in-product assistant and the public Q&A widget are held to the same evidence bar as the rest of Attestly. Specific guarantees, not aspirations.
PII redaction before any LLM call
Every prompt and every response passes through a redactor that strips emails, phones, US SSNs, Luhn-validated cards, common API token prefixes, PEM private keys, and IPv4 / IPv6 addresses — both before the model is invoked and before the message is persisted.
Grounded answers, citation-required
The assistant only sees a per-tenant context built per-turn: workspace counts, your scan findings, your published document sections, your subprocessor inventory, your drift alerts. Framework references in expert mode are anchored — fabricated citations are explicitly disallowed.
Every turn lands in the audit chain
User messages, tool proposals, confirmed mutating actions, and Trust Q&A questions all write entries into the same SHA-256 hash chain. Conversation contents are committed via redacted-content hashes; the export is Ed25519-signed and verifiable offline with @attestly/audit-cli.
Hard budgets, no surprises
Per-tenant monthly message budgets and per-user 12/min rate limits cap the assistant. Trust Q&A counts against a separate budget. Owners see remaining budget under Settings → Compliance copilot. BYO-LLM honored end-to-end on Scale and above.
Infrastructure
Database
Neon Postgres in a single primary region. Encryption at rest (AES-256). Daily automated backups with point-in-time recovery up to 7 days.
Secrets
Stored in the hosting provider's encrypted secret manager. No secrets in the repository. Rotated on personnel change.
Application
Next.js on a serverless edge runtime; background jobs on Inngest with signed event payloads. TLS 1.2+ end to end.
Sub-processors of Attestly
The third-party services Attestly itself uses to deliver the product. We notify Growth+ customers in writing before this list changes.
| Vendor | Purpose | Region |
|---|---|---|
| Vercel | Application hosting | US / EU |
| Neon | Postgres database | US |
| Clerk | Authentication | US |
| Inngest | Background job orchestration | US |
| OpenAI | Document generation + compliance copilot (BYO-LLM available on Scale+) | US |
| Stripe | Billing | US |
| Resend | Transactional email | US / EU |
| GitHub | Source code access (read-only) | US |
Reporting a vulnerability
Email security@attestly.dev with steps to reproduce. We acknowledge within one business day and aim to resolve critical issues within seven. We do not pursue legal action against good-faith research conducted under a coordinated disclosure timeline.
Out of scope: denial-of-service, social engineering of staff, physical access to offices.