Built to observe the agreement, not to take control of it.

Your agreement stays yours

SentientScope is not a document vault and does not want to be. The signed agreement stays in your own storage. What SentientScope keeps is the structured, human-verified scope — each commitment with the exact clause excerpt and section reference it came from — so every alert can show its evidence.

Uploaded files are transient processing inputs: parsed, extracted, then deleted, with an enforced cleanup deadline for anything that fails mid-processing.

The model is a stateless worker

No feature depends on the AI provider remembering anything. There are no provider-side threads, assistants, or stored files acting as memory. Every AI request is reconstructed from SentientScope's own persisted state, and the application — not the model — chooses which verified scope items are relevant to send.

This is tested: our suite includes a check that wiping all provider-side history changes nothing about what the product can do.

Humans make every commercial decision

The model can extract, compare, and summarize. It cannot set a price, approve scope, send a paid change, or declare a contract updated. Those acts require a signed-in person with the right role, and each one is recorded.

Client approval never silently modifies the agreed baseline. The active scope advances only after someone on your team confirms that whatever change process your agreement requires was completed — and that boundary is enforced in the database itself, not just the interface.

Mailbox monitoring with the least access it needs

Each team member connects their own Microsoft 365 mailbox with delegated, read-only consent. SentientScope watches that person's Inbox and Sent Items from the moment they connect — it never imports historical mail, never holds organization-wide mailbox permission, and never sends mail on anyone's behalf. Nobody can connect or reconnect a colleague's mailbox, and administrators see connection health and coverage, never message contents.

Ordinary messages are discarded after evaluation. For a message that may change a project, SentientScope keeps the minimum excerpt, the message identifiers and the normalized event it needs to show its evidence; the full body is transient processing material and attachments are never fetched. Excerpts from observations that never became a decision are redacted after 90 days; evidence behind a decision is kept with the decision. Teams that prefer not to connect a mailbox can forward or CC client mail to a per-project address instead. Your mailbox remains the system of record for correspondence.

When SentientScope cannot tell which project a new sender belongs to, it asks a person on your team to confirm instead of guessing. A pending question that nobody answers expires after 14 days and its transient content is dropped.

Connected systems with the least access they need

Delivery and CRM systems are connected per workspace with the least access each needs: Jira, HubSpot and Salesforce with delegated OAuth consent through a managed integration provider that holds and refreshes the provider credentials — SentientScope stores no tokens for those connections and revokes them at disconnect — and ClickUp with an API token you supply. Credentials SentientScope does hold, for Microsoft 365, Google Calendar and ClickUp, are encrypted at rest.

Raw provider payloads are never stored. Each event is reduced to a normalized observation — the record it concerns, what changed, who changed it and a link back — before any commercial logic runs, and only the excerpt and provenance needed to explain a decision are kept. Connected systems supply context for a project; they are never authority over its scope.

On the Scale plan, API keys are stored only as a SHA-256 hash with a display prefix, inbound observations are accepted only from a valid key scoped to one workspace, and outbound webhooks are signed with HMAC-SHA256 over a timestamp and the body so receivers can verify every delivery.

Tenancy, access, and audit

Every workspace is isolated at the database layer with row-level security, and every consequential API mutation is authorized server-side against the caller's role — a workspace ID from a browser is never sufficient. Provider tokens are encrypted at rest.

Consequential actions — decisions, approvals, reassignments, contract-update confirmations, baseline activations — are written to an append-only audit history, so the record always shows who did what, when. The platform operations console is a separate realm: its accounts require an enrolled authenticator (TOTP) and it shows connection health and account state, never client documents or message contents.

What we don't claim

SentientScope does not currently hold SOC 2 or ISO certification, and this page will not pretend otherwise. What you have read are engineering boundaries we test in CI on every change. Questions, or something you'd want to verify before trusting us with client work — write to team@sentientscope.io.