Skip to content

docs: replace fabricated local-encryption claims on product+security copy#231

Merged
operatoruplift merged 1 commit intomasterfrom
honesty/dataservice-product-and-security
Apr 28, 2026
Merged

docs: replace fabricated local-encryption claims on product+security copy#231
operatoruplift merged 1 commit intomasterfrom
honesty/dataservice-product-and-security

Conversation

@operatoruplift
Copy link
Copy Markdown
Owner

Summary

  • Closes the honesty sweep across src/services/dataService.ts (product card Install Vercel Web Analytics for Next.js #1 + security section) and src/sections/ProductPage.tsx (extendedDescriptions + technicalDetails)
  • Replaces "everything lives on your computer, encrypted" / "AES-256 encrypted local storage" / "your data never leaves your environment" with the real posture: cloud-routed prompts, named third-party providers, approval gate + signed receipt + on-chain audit log
  • technicalDetails now points at shipped files (lib/capabilities.ts, lib/toolSafety.ts, Privy JWTs, ed25519 receipts) so a reviewer can ground-truth each claim

Why this matters

The Vercel-hosted Next.js app routes prompts through Anthropic / OpenAI / Google / xAI / DeepSeek depending on which provider the user picks per turn — it has never been a local-only app. PRs #227#230 already cleaned Privacy / Compliance / ProblemStatement / Pricing / Comparison / FAQ / CloudVsLocal. This finishes the homepage product+security surfaces.

Test plan

  • `pnpm check` (4/4 passed: copy-check, capability-check, trust-gate, fabrication-rot-check)
  • `pnpm exec tsc --noEmit` (clean)
  • Manual inspection of product cards and security section copy

Continues the honesty sweep across the product/security copy that still
told users their data lives encrypted on their machine. The web app does
not run locally and never has — it is a Vercel-hosted Next.js app routing
prompts through Anthropic, OpenAI, Google, xAI, or DeepSeek per the user's
selection per turn.

dataService.ts:
- Product card #1 now describes the real data posture: export/erase from
  Settings → Data; named third-party AI providers see only the prompts
  you send them; we do not resell.
- security section retitled TRUST with a posture grounded in the gate +
  receipt + audit log we actually ship. Subhead names HIPAA-aware
  architecture, GDPR workflows, MIT-licensed open source.
- features.security pivots to "Every action carries a receipt" (ed25519
  signing + Merkle root anchored on Solana via the deployed audit-trail
  Anchor program). linkUrl moved to /docs/receipts.

ProductPage.tsx:
- extendedDescriptions[0]: dropped "private space on your computer,
  encrypted" → describes per-session scope, capability gating, and
  inter-helper isolation in terms that match runtime behavior.
- extendedDescriptions[4]: dropped "taking an action on your computer"
  → describes outbound sends/bookings/API calls with approval and
  request-id logging.
- technicalDetails fully rewritten: each bullet now points at a shipped
  surface (lib/capabilities.ts, lib/toolSafety.ts, Privy JWTs, ed25519
  receipts) so a reviewer can ground-truth the claims.

After this lands, the only "stays on your computer" framing left in
homepage copy is roadmap language explicitly tied to the future desktop
app, which is correct given the Ollama path is documented as roadmap
in the FAQ and Privacy Policy.

Verification:
- pnpm check (4/4 passed: copy-check, capability-check, trust-gate,
  fabrication-rot-check)
- pnpm exec tsc --noEmit (clean)
@vercel
Copy link
Copy Markdown

vercel Bot commented Apr 28, 2026

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
website Ready Ready Preview, Comment Apr 28, 2026 10:39am

operatoruplift added a commit that referenced this pull request Apr 28, 2026
The "Why now?" narrative said "one assistant on your computer can
replace the dozens of productivity apps you juggle today. We're
building it before big tech does, and putting it on your machine
instead of theirs."

The web app is Vercel-hosted and routes prompts through Anthropic /
OpenAI / Google / xAI / DeepSeek per the user's selection per turn.
It is not a local-only product. The desktop+Ollama path is roadmap-
only, documented as such in the FAQ and Privacy Policy.

Rewrite the narrative around what the product actually is:
- One assistant that can sit across many SaaS apps and move work
  between them (true: Gmail+Calendar via Google OAuth, more on the
  roadmap).
- Approval before every action and a signed receipt afterward
  (true: ed25519 receipts anchored on Solana via the audit-trail
  Anchor program).

This continues the honesty positioning sweep that PRs #227-#231 did
across Privacy / Compliance / ProblemStatement / Pricing / Comparison
/ FAQ / CloudVsLocal / dataService / ProductPage. After this lands,
the only remaining "computer/machine" framing on the site is the
explicit roadmap copy in the FAQ and Privacy Policy pointing at the
future desktop app.

Verification:
- pnpm check (4/4: copy-check, capability-check, trust-gate,
  fabrication-rot-check)
- pnpm exec tsc --noEmit (clean)
@operatoruplift operatoruplift merged commit f176009 into master Apr 28, 2026
4 checks passed
operatoruplift added a commit that referenced this pull request Apr 28, 2026
)

Add 4 anchored patterns to scripts/fabrication-rot-check.mjs that fail
the build if a future PR reintroduces the local-machine fabrications
that PRs #231-#232 retired:

- "on your machine instead of theirs" -> retired in #232
  (MarketOpportunity)
- "Everything lives on your computer" -> retired in #231
  (dataService product card #1)
- "AES-256 encrypted local storage" -> retired in #231
  (dataService security section)
- "your data never leaves your environment" -> retired in #231
  (dataService security features.security)

Each pattern is anchored to the exact phrasing that was in the
retired copy. The COMMENT_LINE_RE skip means rule comments and
audit-trail commentary in code are still allowed, only live source
strings fail.

Total rules: 10 -> 14.

Verification:
- pnpm check: 4 passed (4/4: copy-check, capability-check, trust-gate,
  fabrication-rot-check) with 14 rules across 197 files.
- pnpm exec tsc --noEmit (clean - no source changes).
operatoruplift added a commit that referenced this pull request Apr 28, 2026
src/sections/Security.tsx: replace the "AES-256-GCM Encrypted" badge
on Security card 1 with "Ed25519 Signed Receipts". The card is now
about ed25519-signed receipts anchored on Solana (per #231 rewrite of
features.security), so the AES-256-GCM badge no longer matches the
copy underneath it. The badge should describe what the card actually
proves.

app/blog/posts.ts: rewrite the excerpt for "Local AI vs Cloud AI: The
Privacy Case Nobody's Making". The previous excerpt asserted "When
your AI runs locally, you make the decisions" without any web-app
caveat, which read as "Operator Uplift runs locally today." The new
excerpt distinguishes the present-tense web app (cloud-routed per
turn) from the desktop+Ollama path on the roadmap, and points out
both layers share the approval-gate + signed-receipt primitives.

Continues the honesty positioning sweep that PRs #227-#238 landed.

Verification:
- pnpm check (4/4: copy-check, capability-check, trust-gate,
  fabrication-rot-check, all clean with 14 rules across 197 files)
- pnpm exec tsc --noEmit (clean)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant