All blog posts
Updates 6 min read

Commonveil Tests an Experimental Secure Net Layer

Commonveil's 0.23.0 code introduced an optional experimental Secure Net layer for direct messages and a community forum for protocol proposals.

By Comutability Published Update from

Project update: 14 November 2026. GitHub pull request #6: merged into GitHub source, not necessarily deployed. Commonveil has not been independently security audited.

Commonveil's 0.23.0 code introduced an optional experimental Secure Net layer for direct messages and a community forum for protocol proposals.

Why this matters

The decision was to integrate cautiously rather than switch everyone's cryptography at once. The experimental control requires conscious opt-in, and the GitHub-linked proposal forum aims to make standards discussions visible. A protocol being present in the repository does not mean it has been deployed or audited.

Detailed change record

The public engineering notes below describe what was implemented or proposed and the evidence recorded with the change. Historical instructions in the notes describe the pull request at the time it was raised.

What this does

Commonveil 0.23.0 takes its first step onto the Secure Net protocol. It is off unless a person turns it on and is labelled experimental. Nothing here has been published to Base44.

For people using the site

  • New switch: Settings, Devices and keys, "Experimental: Secure Net protocol".
  • What it does: with it on for both people, a direct message is sealed a second time under a one-off key. That key goes only to the devices linked at that moment, over post-quantum sessions (X25519 with ML-KEM-768).
  • Why it matters: someone who later obtains an account's keys or recovery phrase still can't read those messages.
  • Never weaker: it adds to the existing encryption and doesn't replace it. Messages to people without it on go exactly as before, and everything already sent stays readable.
  • Identity: turning it on gives the account a Secure Net actor, a quantum-resistant key with the handle @username@site.

A forum for protocol proposals

  • What it is: a new forum, snip-proposals, in the community server, with one thread for every Secure Net Improvement Proposal (SNIP) on GitHub.
  • Opening: a thread is started when a SNIP appears on the default branch or in an open pull request.
  • Closing: a thread gets a closing note and is locked when the SNIP's status becomes Accepted, Rejected, Implemented, Released or Withdrawn, or its pull request is closed without merging. A merged SNIP that is still a Draft stays open.
  • Who posts: posts are end-to-end encrypted, so the site can't write them. The app of whichever community admin is online does it, about every ten minutes. Changes on GitHub appear the next time an admin has Commonveil open.
  • Safeguards: fixed thread ids stop duplicates, members can reply but not start threads, stale GitHub data is never acted on, and a thread locked by hand is never unlocked automatically.
  • Privacy: only the site reads GitHub, and only the public list of proposals. Non-admins never trigger a read.
  • Settings: on for the hosted site (SNIP_FORUM=off turns it off). Off on self-hosted sites unless SNIP_FORUM=on. An optional GITHUB_TOKEN makes checks every 3 minutes instead of every 20. See docs/SNIP_FORUM.md.
  • Tested: one end-to-end test with a pretend GitHub covers every rule above. I removed 13 guards one at a time: 11 made it fail, and the other 2 are deliberate double safeguards.
  • Not tested: against the real GitHub from a deployed function, or in a browser.

How it is built

  • Library: packages/secure-net is now Secure Net 0.5.0.
  • Client: src/lib/secure/pq/ holds the engine (pure functions), the layer (wiring) and the browser store. protocol.js gains an optional second seal on build and open.
  • Backend: a new protocol function, the same code on Base44 and self-hosted sites, with tables ProtoActor, ProtoPrekey and ProtoMail.
  • Base44 compatibility: the server stores and routes sealed envelopes and checks shapes, permissions and rate. Devices verify every post-quantum signature. No post-quantum code runs inside Base44 functions, where it hasn't been proven to work.
  • Clean-up: account deletion and the maintenance sweep cover the new tables.
  • Docs: docs/SECURE_NET_LAYER.md says exactly what it does, its limits, and what the site can see.
  • Standing rule: AGENTS.md now says Commonveil follows each Secure Net update.

Limits (all stated in the app and docs)

  • Coverage: direct messages only. Groups, servers, edits, reactions, polls, calls, scheduled and offline messages are unchanged.
  • New devices: a device linked later can't read messages sent with the layer before it joined. History copying isn't wired up.
  • Passcode lock: the layer can't be on while a passcode lock is set, because the lock doesn't cover its keys yet.
  • One site: it works within one site. Cross-site delivery, discovery, repositories and lexicons come later.
  • First sight: a person's protocol identity is trusted the first time it is seen and pinned after that. Safety numbers aren't shown in the app yet.

Needs review before it is relied on

This builds on Secure Net parts that are Draft and unreviewed (Secure Net PRs #3 and #4, still open). The copy here is taken from the social-protocol branch. Please don't publish this to the live site as anything but an experiment until a qualified reviewer has looked at CVP-016.

Found along the way

The integration caught a protocol bug, fixed in Secure Net PR #4: the social endpoints had been added to the default endpoint list that existing sites publish, so sites would have advertised endpoints they don't serve.

Tested

  • npm run verify passes: lint, type checks, 237 tests and the production build.
  • tests/protocolLayer.test.mjs has five end-to-end tests that drive the real client code against the real backend handlers. They cover both layers being needed, fallback when the other side is off, a second device joining, server permissions, and a changed identity being refused.
  • I removed 11 security checks one at a time. Ten made a test fail. The eleventh is a duplicate of a check that did.
  • Not tested: the feature in a real browser against Base44, and the Docker images locally (Docker Hub is blocked where I work, so CI builds them).

Before publishing

  1. Sync entities and functions to Base44 (four new tables, two new functions).
  2. Try the switch with two test accounts on the live site.
  3. Tag v0.23.0 after merging (tag pushes are blocked for me).

Decisions and principles behind the work

Comutability is building towards privacy, user control and genuinely self-hostable software. That does not mean every proposal is already deployed. Cryptography and security improvements need qualified human review and meaningful tests before any strong safety promises are made.

What comes next

Test against real hosted accounts, gather feedback on identity and device linking, and keep experimental encryption separate from the project's safety guarantees.

Check the sources

Sources and further reading

Explore the evidence and original project records behind this article.