Controls you can trace
to the product.
A plain-language account of how Vigil isolates workspaces, governs access, preserves evidence, and records sensitive actions. Each statement is limited to current product, infrastructure, or qualified provider evidence.
Workspace isolation
Every piece of content in Vigil — sources, compiled pages, decisions, assumptions, briefs, alerts — belongs to exactly one workspace.
Membership is checked on every request.
Each API request resolves the caller's workspace membership server-side before anything else runs; requests without a valid membership are rejected.
Every read and write is scoped to the workspace.
Application queries carry the workspace as a mandatory scope, reinforced by composite tenant-bound database relationships for tenant-owned records.
Workspaces don't share derived state.
Compiled knowledge, AI-extracted assumptions, and alert history are computed per workspace; one tenant's content never informs another's results.
Access control
Workspace access is owner-controlled.
Only the workspace owner can invite people or manage the subscription; members work within the workspaces they've been invited to.
Invites are controlled.
Invitations are tokenized links that expire and are bound to the invited email address; joining requires signing in as that address.
Sessions are server-side.
Sign-in creates a database-backed session with a 7-day lifetime, carried in an httpOnly, secure, SameSite cookie — session tokens are never exposed to page JavaScript.
Passwords are hashed
With scrypt, a memory-hard algorithm; we never store or log plaintext passwords.
Data handling
Workspace export is controlled and auditable.
Owners can export current pages, source metadata and extracted text, claims, conflicts, briefs, and answers as Markdown and JSON. Each export is recorded in the workspace audit trail.
Vigil does not train models on your content.
Default OpenAI processing follows the applicable API terms; processing through your own endpoint follows your provider agreement.
Deletion is self-serve.
Owners can delete one workspace while keeping their account, session, billing identity, and other workspaces, or separately delete their whole account. Active workspace content is removed after the required checks; identifiable audit/security records, legally required records, de-identified operational records, provider records, logs, and disaster-recovery copies follow their own retention.
Evidence keeps its provenance.
Imported and watched material retains its connector, external reference, version reference, and observation time where the provider supplies them. Vigil does not replace provider-stamped facts with pipeline timestamps.
No third-party analytics SaaS.
The product and marketing site run first-party analytics only — no external trackers or ad pixels.
Auditability & retention
Sensitive actions enter an append-only workspace audit trail.
Membership changes, exports, public-share changes, and deletions are recorded with the actor and workspace context.
Public brief links are revocable and access-logged.
Shared links are rate-limited, excluded from search indexing, and stop working immediately when revoked.
Retention follows explicit data classes.
Audit, analytics, queue, session, and share-access records have separate deployment-level retention policies; evidence and reasoning history are not silently swept by a generic age-based purge.
AI processing
Default:
Relevant excerpts are processed through our shared OpenAI route under the applicable API terms. Full prompt bodies are not stored in Vigil's AI usage ledger.
Enterprise: bring your own endpoint.
An eligible workspace can route structured workspace product calls to an OpenAI-compatible endpoint in your own cloud (for example Azure OpenAI). The support assistant remains on our shared provider. The credential is stored encrypted with AES-256-GCM, is never shown again after entry, and decryption failure stops the workspace call rather than silently changing routes.
Network & infrastructure
Hosted on AWS
The application is hosted in AWS us-west-2 in the United States, served over TLS with HSTS and an enforcing Content-Security-Policy. Bring-your-own AI routing is not application residency or self-hosting.
The URL fetcher is hardened against internal-network access.
Watched-source fetches resolve and pin DNS, reject private and internal network addresses, and re-validate on every redirect — a watched URL cannot be used to probe our infrastructure.
Connector failures fail closed and remain actionable.
Google Drive, Microsoft OneDrive and SharePoint, Dropbox, Box, SEC EDGAR, Companies House, and web sources use a closed connector registry and structured failure categories. Credential and permission failures leave automatic scheduling until a user explicitly resumes the watch.
Payments never touch our servers.
Card details go directly to Stripe; we store only the subscription state Stripe shares with us.
Verify any control on this page.
We work with security and diligence teams directly. For questionnaires, subprocessor detail, data-handling terms, or a walkthrough of any control on this page, questions are answered by the team that builds the product.