Compass connects read-only to the tools your team already uses, takes numbers-only snapshots over time, and deterministically infers what’s shifting — in plain language. No content reading. No black-box guessing.
Deterministic and aggregate. Compass reads login · usage signals and calendar metadata — never the content inside your tools.
Three stages, fully reproducible. No model weights to explain, no confidence intervals to hide.
Start with one real source (Google Calendar OAuth), add more over time. Read-only scopes; tokens encrypted server-side. Compass never requests write access or content-level permissions.
Compass stores immutable, numbers-only aggregate snapshots on a schedule — counts and durations, never content — then diffs them across weeks. Each snapshot is append-only and audit-logged.
A pure, reproducible function reads the diffs and surfaces the pattern. No LLM in the loop, no hidden scoring — the same inputs always give the same answer. You can audit every rule.
Here is what an inference result looks like after six weeks of calendar snapshots. The engine produces a plain-language headline, a trend line, and supporting signals — nothing more.
Weekly deep-focus percentage — 6-week window. Downward trend: −18 pp from W1 to W6.
Deterministic and numbers-only. Compass reads login/usage signals and calendar metadata — never content. Aggregate by default. The engine surfaces the pattern; you decide what to do with it.
