Skip to content

docs(try): note hosted data is encrypted with the user's access key - #70

Merged
sanity merged 1 commit into
mainfrom
feat/try-callout-encryption
Jul 7, 2026
Merged

docs(try): note hosted data is encrypted with the user's access key#70
sanity merged 1 commit into
mainfrom
feat/try-callout-encryption

Conversation

@sanity

@sanity sanity commented Jul 7, 2026

Copy link
Copy Markdown
Contributor

DRAFT until review passes (merge auto-deploys freenet.org).

Problem

The /try/ callout says "your data lives with us" but omits a real, reassuring fact: hosted per-user data is encrypted at rest with a key derived SOLELY from the user access token, so the operator cannot read it when the user is not actively connected.

Approach

One added sentence in the callout: "It's encrypted with your access key, so while you're not using it even we can't read it."

Backing (verified, not assumed)

Confirmed against freenet-core crates/core/src/server/client_api/hosted_export.rs ("Threat model: the token is the entire secret"): the per-user DEK and bundle key are derived solely from the user token and are node-KEK-independent by design; there is no node-side second factor. The token is header-only, zeroized, and lives in the user's browser. Honest nuance: while the user IS actively connected the node holds the token and can decrypt to run the app; the claim is scoped to data at rest ("while you're not using it"), which the wording states.

Testing

Rendered locally; callout reads correctly, no em/en-dashes. Copy-only change.

[AI-assisted - Claude]

🤖 Generated with Claude Code

https://claude.ai/code/session_012jmCd1tvDH1EsVTtaDVnxJ

Add a plain-language reassurance to the /try/ hosted-demo callout: hosted data is
encrypted at rest with a key derived solely from the user's access token
(node-KEK-independent per hosted_export.rs 'the token is the entire secret'), so
when a user isn't actively connected the operator has no token and can't decrypt
it. Verified against the freenet-core hosted secrets model before claiming it.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012jmCd1tvDH1EsVTtaDVnxJ
@sanity
sanity merged commit b8952b8 into main Jul 7, 2026
3 checks passed
@sanity
sanity deleted the feat/try-callout-encryption branch July 7, 2026 21:22
sanity added a commit that referenced this pull request Aug 9, 2026
* docs(quickstart): stop competing with /try/, and link to it

/quickstart/ was titled "Try Freenet: Join River", which collided with
/try/ ("Try Freenet in your browser"). The two were near-indistinguishable
in a tab strip or a search result, while the nav called the page a third
thing, "Quickstart".

The collision is a leftover of one-directional editing. /quickstart/ got
that title on 2026-05-04 (7e3cb08), two months before /try/ existed. The
/try/ work on 2026-07-07 (#68/#69/#70) added the new page and rewired the
homepage, but never came back to the older one.

The link was one-way for the same reason: /try/ pointed at /quickstart/
twice, /quickstart/ pointed at /try/ not at all. So anyone arriving from
the nav, a search result, /apps/, /river/ or the manual never learned the
no-install option existed, which is the direction that matters more since
the install page is the one carrying the friction.

- Title and <title> become "Install Freenet: Join River", matching the
  homepage's primary CTA and freeing "Try Freenet" for /try/. The URL is
  unchanged, so no redirects.
- A pointer to /try/ sits above Step 1, before the install friction rather
  than buried in "What's Next?", and describes the demo the way /try/
  describes itself ("hosted demo", runs on our server, a preview).

Both pages now name each other in the other's own vocabulary: /quickstart/
links "Try River in your browser", /try/ links "Install Freenet →".

Verified with cargo make check-links (self-test plus 222 pages, no broken
references) and by screenshotting both pages at 1280px and 390px.

[AI-assisted - Claude]

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KBzsn4BUH4dHQ9gMHSGj2P

* docs(quickstart): fix an inaccurate claim about the hosted demo

Review found the new paragraph described /try/ as running "on our server
rather than peer-to-peer". That is wrong, and it made /quickstart/ claim
more centralization than /try/ claims for itself.

try.freenet.org is a Freenet peer we run, not a non-P2P reimplementation.
It serves the same node-gateway path shape a local peer serves, for the
same contract key: /v1/contract/web/raAqMhMG7KUpXBU2SxgCQ3Vh4PYjttxdSWd9ftV7RLv/
against try.freenet.org, versus localhost:7509 for an installed peer (the
default base in themes/freenet/layouts/shortcodes/river-invite-button.html).
What is centralized is whose peer holds your session, not whether the
network is involved. Reworded to say that, matching how /try/ puts it
("your data lives with us").

Two more review findings:

- "fastest way to see the network in action" was contradicted one
  paragraph later by a no-download option. Now "best way".
- Troubleshooting's network-requirements note was a dead end for exactly
  the reader /try/ serves: someone whose firewall blocks UDP hole
  punching, who has already tried to install and failed. It now points
  there, since the browser demo reaches the network over HTTPS.

Not adopted: propagating /try/'s "export your data in one click" claim to
this page. The reviewer flagged it as unverified and I did not verify it,
so it stays on the one page that already asserts it.

Verified again with a clean build: check-links self-test passes, no broken
internal links across 222 pages.

[AI-assisted - Claude]

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KBzsn4BUH4dHQ9gMHSGj2P

* docs(quickstart): balance the hosted-demo caveats

Second review pass. Three copy findings, all in the text this PR added:

- "your data lives with us" carried the alarming half of /try/'s framing
  without the mitigation that follows it there ("encrypted with your access
  key, so while you're not using it even we can't read it"). A reader who
  never clicks through got only the scary part. Dropped rather than
  duplicated: "a preview rather than the real thing" already states the
  tradeoff, and /try/ carries the full data story where the reader actually
  decides to use it.

- The firewall note promised the demo "still works", unconditional. The
  transport claim holds for the parts this repo can show (the page and the
  invite API are both HTTPS/443), but River in the browser also needs a
  WebSocket to the hosted node, which lives in another repo and was not
  verified, and a DPI proxy can block WS upgrades while allowing HTTPS. Now
  "usually still works", and says "over HTTPS" so the reason is legible.

- That same sentence called it "River in your browser" with no indication it
  is hosted. Someone deep-linking to #troubleshooting from a search would
  read it as the real thing. Now "the hosted demo".

- "best way to see the network in action" was still an unsupported
  superlative sitting above a paragraph that offers a different path first.
  Now "for real", which is the distinction that actually holds.

Clean rebuild: check-links self-test passes, no broken internal links across
222 pages, and both /try/ links plus the new wording confirmed present in the
built HTML. Re-screenshotted at 1280px.

[AI-assisted - Claude]

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KBzsn4BUH4dHQ9gMHSGj2P

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
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