No. 05 The audit trail
Proofs.
Every claim on this site, indexed with its evidence. Where the evidence is internal, we say so; where a number is a target and not a result, we say that too. This page holds us to our own constitution: verified, not vibes.
| Claim | Evidence | Kind | Status |
|---|---|---|---|
| "No trackers on this site" | View source. No analytics, no cookies, no fingerprinting. Zero third-party requests — fonts are self-hosted, so the page loads from exactly one origin. | PUBLIC | VERIFIED |
| "Our agents do everything except hit send" | Approval gates are architectural: agent frameworks interrupt before any send, post or payment; drafts wait in an approval queue surfaced on dashboard, Telegram and terminal, with an audit log. | INTERNAL | AUDITED |
| "In pilot across four business functions" | Work OS, HR OS, Sales OS and Media OS run as working builds on live accounts (email, calendar, real lead pipelines) with human-approval gates on. Pilot means pilot — not GA, not "coming soon." | INTERNAL | AUDITED |
| "Drafts grounded in your CRM, your policies, and traced quotes" | Sales drafts are grounded in crawled business data and CRM history; HR drafts carry policy citations; media drafts trace quotes to sources. Grounding artifacts are stored with each draft. | INTERNAL | AUDITED |
| "Model-agnostic routing" | Every OS ships a multi-provider router with fallback chains and per-provider quota tracking across frontier and open-weight models. No product hard-codes a single vendor. | INTERNAL | AUDITED |
| "Evals before production" | Pilot systems carry dedicated eval suites and regression tests for agent behavior; speech pipelines gate releases on word-error-rate benchmarks against public corpora. | INTERNAL | AUDITED |
| "Runs in your cloud — or fully on-premises" | Working local-first builds exist across the lab: fully on-device transcription and summarization, phone-local agents with tool use and retrieval, offline-capable coding agents. Nothing phones home. | INTERNAL | AUDITED |
| "Hard per-run budget caps" (Media OS) | Agent runs enforce a per-run USD ceiling with cache-aware token pricing; runs halt at the cap. Enforced in code, not in a prompt. | INTERNAL | AUDITED |
| "No-send by design" (Sales OS) | The lead pipeline verifies deliverability and drafts emails but ships no sender: results export to CSV or your own tools. The send button is architecturally absent, not disabled. | INTERNAL | AUDITED |
| "DPDP- and GDPR-aware architecture" | India-market designs specify India-jurisdiction storage, delete-my-data flows and consent screens per the DPDP Act 2023. Design-stage: not certified deployments, and this row will say so until they are. | TARGET | IN DESIGN |
| "Cited, or silent — under 1% confident-wrong" (Knowledge OS) | A design target, not a measured result — Knowledge OS is in development. The spec pins the metric, a cross-vendor grading protocol, and an abstention floor. Numbers publish when they exist. | TARGET | NOT YET MEASURED |
| Certifications (SOC 2, ISO) | None held today. No badges appear on this site until they are real. This row exists so the absence is a statement, not an omission. | PUBLIC | NONE HELD |
Spot a claim anywhere on this site that isn't on this page? That's a bug. Report it and we'll index it or retract it.