Security audit package

From mnema-talk v3.3.13, rendered on this site.

This is the single entry point for an external security review of Mnema Talk. It collects scope, architecture, threat model, cryptographic design, the attack surface, the controls, what the project has already found, what it knowingly accepts and where it would like a second pair of eyes. It summarises and links; the normative description of each control stays in SECURITY.md and the other documents linked below. Nothing here is a claim of an external audit: none has taken place yet (product status).

Contents: 1 Scope and versions · 2 Architecture and data flows · 3 Threat model · 4 Cryptography · 5 Attack surface · 6 Security controls · 7 Results of internal reviews · 8 Known limitations and accepted risks · 9 Building and running a test instance · 10 Suggested test focus

1. Scope and versions

ItemValue
ProductMnema Talk, a self-hosted, invite-only chat and voice server for small communities
Audit targetThe latest release of the current line, 3.3.x (at the time of writing v3.3.10); the tag is the reference
Previous releases3.0 to 3.2 (bot platform with manifests and interactions, timed messages, finer auto-delete), 2.x (built-in TURN relay, wiki, events, raid protection), 1.2 (the large feature release described below), 1.1.0 (direct messages, blocks, reports, forwarding), 1.0.x (first stable line)
Repositoryhttps://github.com/mnema-xyz/talk (public, source available to the reviewers)
ServerGo (module go 1.27), chi 5.5, pgx 5.13, Pion WebRTC 4.2, golang.org/x/crypto, webpush-go 1.4
Web appVue 3 single-page app (TypeScript), ts-mls 1.6.4 for MLS, Vite build, embedded in the server binary
Data storesPostgreSQL 18, S3-compatible object storage (SeaweedFS in the compose stack)
Optional servicescoturn (TURN), update sidecars (updater, recreator, off by default)
Desktop clientSeparate repository (https://github.com/mnema-xyz/talk-desktop, a Tauri shell on Windows and macOS and an Electron client on Linux, Beta, installers unsigned at OS level) running a signed web bundle built from this repository

In scope: the server (cmd/server, internal/), the web app (web/), the signed web bundle format and its verification (internal/webbundle, desktop-trust/), the compose deployment (docker-compose.yml, Dockerfile, deploy/), the release pipeline (.github/workflows/, signing, provenance), the self-update sidecar (cmd/updater, internal/updaterd).

Out of scope, or only on request: the desktop shell's Rust code and its installer signing (separate repository, see section 8), third-party services (push services, GitHub, GIF provider), the operator's reverse proxy and host, social engineering of the Owner, and denial of service by resource exhaustion on an instance you do not operate.

2. Architecture and data flows

One Go binary serves the REST API, a WebSocket hub, a WebRTC selective forwarding unit (SFU) and the embedded web app. Component list: architecture.

 Browser / desktop shell            Operator's reverse proxy (TLS)
        |  HTTPS, WSS                      |
        v                                  v
 +-----------------------------------------------------------+
 | mnema-talk                                                |
 |  chi router -> middleware -> REST handlers                |
 |  ws hub  (events, presence, voice signalling, MLS relay)  |
 |  sfu     (Pion, forwards sealed RTP, UDP)                 |
 |  web     (embedded SPA)                                   |
 +--------+---------------------+----------------+-----------+
          v                     v                v
     PostgreSQL            object storage     outbound: Web Push services,
   (all state, audit)      (uploads, avatars)  webhooks, link previews,
                                               GitHub release check
FlowPathProtectionServer sees
RESTBrowser to /api/...TLS (operator), HttpOnly session cookie, origin check (CSRF), per-user and per-IP rate limits, body limitsEverything (text is server-readable by design)
WebSocket/api/ws, one per sessionCookie auth on upgrade, origin check, re-checked every two minutes, per-event payload caps and flood limits, per-recipient permission filterEverything except media payloads and MLS contents
Voice/video transportUDP between browser and SFUDTLS-SRTPEncrypted RTP, plus metadata (who, when, sizes, layers)
Media E2EEInside the RTP payload, in the browserMLS-derived AES-128-GCM key, Ed25519 signature per frame, replay windowCiphertext and codec headers (authenticated, readable)
MLS handshakeOver the WebSocket as opaque messagesMLS signatures; server checks only framing and sizeKey packages, welcomes, commits (public), who sent them
Web PushServer to browser vendor push serviceRFC 8291 encryption, VAPID; endpoint guarded by the SSRF resolverSubscription endpoints and push contents before encryption
StorageApp to S3 API; media only through the app, never directContent sniffing, avatar/banner/emoji re-encode, metadata of JPEG/PNG/WebP attachments removed (Owner switch, on by default), per-account quotas; access check on every downloadPlain files (other attachments are stored as uploaded)
OutboundWebhooks, link previews, update check, pushinternal/ssrf resolve-vet-connect guard, no redirects, size capsTargets chosen by members or admins
Desktop shellTauri app to the instance through its own proxySigned web bundle (Ed25519, pinned key, security floor), session cookie kept out of the page, strict CSPSame as a browser
Update sidecarAdmin console to updater to recreatorOff by default, shared token, cosign signature and provenance check of the new imagen/a

Further reading: chat data flow, direct messages, search and forwarding, Talks and Stages, moderation, audit and automod, roles and permissions, webhooks, bots, desktop trust model.

3. Threat model

3.1 Actors

ActorCapabilities assumed
Anonymous internet clientReach the public routes (health, legal, invite preview, sign-in, registration with an invite code, webhook execute with a secret)
Invited memberValid session; default @everyone permissions; can upload, search, call, create DMs, create webhooks only if an admin allows
Member with rolesAdditional permissions below their highest role; may not grant what they lack or touch members above them
Administrator / role with AdministratorEverything except Owner protections
OwnerThe one ADMIN_USERNAME account; holds every permission, cannot be restricted by anybody
Bot / webhook holderToken or URL secret; scoped to what a bot role or one channel allows
Other member as call peerInside the trust boundary for media they can read; limited by the group rules (4.3)
Network attackerCan observe and tamper with traffic between client and proxy only if TLS is broken or absent
Malicious or compromised serverOut of scope for media confidentiality and code integrity in browsers (the server serves the web app); in scope for what the desktop client and the safety code can still show
Malicious operator of the hostOut of scope (owns the database and keys)
Supply-chain attackerDependencies, CI, release signing, container images, update path

3.2 Assets

Account credentials and sessions; the permission model's integrity (who may see or do what); message and file confidentiality against non-members (text is not hidden from the server or the Owner); direct-message privacy against moderators and other members; media confidentiality against an honest server and anyone on the path (E2EE); audit-log integrity; instance secrets (JWT_SECRET, S3 and database credentials, VAPID key, TOTP encryption key derived from DATA_KEY, or from JWT_SECRET without one); the signing keys of web bundles, desktop updates and release images.

3.3 Trust boundaries

  1. Internet to reverse proxy to app (TLS ends at the operator's proxy; client IPs are taken from X-Forwarded-For only from listed proxies).
  2. Browser page to the server it was loaded from (same-origin; the server is trusted to serve code, except in the desktop client).
  3. Member to member inside a call (MLS group; membership is asserted by the server, see 4.4).
  4. App to PostgreSQL and object storage (private compose network; their ports are not published).
  5. App to the outside (outbound requests pass the SSRF guard).
  6. Release pipeline to operators (cosign signatures, provenance, pinned digests in compose).
  7. Web bundle signer to desktop clients (offline root key, pinned bundle and updater keys).

4. Cryptography

4.1 What is end-to-end encrypted, and what is not

DataProtection
Microphone, camera, screen video, screen audioEnd-to-end encrypted in the browser with MLS-derived keys; signed per frame
Text messages, attachments, avatars, DMs, polls, forum postsNot end-to-end encrypted, by design. Readable by the server and the Owner's database access. TLS in transit; DMs are hidden from moderators by the permission model, not by cryptography. Decision is final: no text E2EE is planned
Metadata (who speaks when, frame sizes, layers, membership, timing)Visible to the server
Web Push payloadsEncrypted for the browser (RFC 8291); the server knows the plaintext
Passwordsbcrypt; sessions are signed JWTs with a per-user version and per-token id; TOTP secrets AES-256-GCM encrypted under a key derived from DATA_KEY (from JWT_SECRET without one), each blob naming its key; recovery codes keyed HMAC; both secrets rotate with a verify-only previous value

4.2 Media encryption

  • Group protocol. MLS (RFC 9420) in the browser with ts-mls 1.6.4, one cipher suite: MLS_128_DHKEMX25519_AES128GCM_SHA256_Ed25519. One group per voice channel; the group id is the channel's UUID. The server relays handshakes as opaque messages and checks only framing and size.

  • Key derivation. Every epoch, the media key is MLS-Exporter("mnema-media-v1", groupId, 16), an AES-128-GCM key held in a non-extractable CryptoKey. The safety code is a hash (domain mnema-safety-v1) of the epoch authenticator, shown as 20 digits so members can compare it over another channel.

  • Frame format version 3 (the only version in 1.2):

    'M' '3' | epoch u32 | stream u32 | counter u32 | line u8 | nonce (12) |
    AES-GCM ciphertext + tag (16) | Ed25519 signature (64)
    

    AAD = channelId 00 senderId, epoch, stream, counter, line byte, clear bytes (codec headers that stay readable). Signed bytes = domain mnema-media-frame-v3 || len(AAD) u32 || AAD || header and ciphertext. The line byte (source in the high nibble, screen layer in the low bits) binds a unit to the line it was made for; a receiver opens a unit only if it names the source the arriving line carries. The nonce is 96 fresh random bits per frame. The signature is verified before decryption, with the sender's MLS leaf signature key (an id with two leaves verifies nothing). The replay window (64 units per (epoch, stream)) is marked only after verification and decryption.

  • Other signed messages over the handshake channel, each with its own domain string and the sender's leaf key: heartbeat (mnema-media-heartbeat-v1, epoch and increasing time, about every 5 minutes) and capability (mnema-media-capability-v1, being phased out with version 2).

  • Key rotation. An epoch lasts at most 30 minutes (random jitter of two minutes per epoch): the committer (lowest member id) advances it with an update-path commit; every other member renews its own leaf on the same schedule. A member silent for 20 minutes is removed. The previous epoch's key opens frames for 30 seconds after a switch within the same group, then its keys and verifiers are deleted. After a removal the sender seals under the new epoch at once; after an add or rotation it keeps the old epoch for 1.5 seconds so receivers have the commit.

  • Commit rules checked by every receiver (after the MLS signature verifies): only the current committer may add or remove; an add must name an identity in the room whose key package the relay delivered and that holds no leaf; a remove must target a member the roster no longer lists, one silent for the absence limit, or one the same commit replaces; no commit may change another leaf's identity, signature key or encryption key. Details: SECURITY.md, Voice and screenshare.

4.3 What a group member can do

A member of a call can read the media of that call (it holds the key). It cannot impersonate another member (frames are signed per leaf), remove a present member, add a second leaf for an identity, or move units between a sender's lines. A malicious committer keeps the powers of the role: removing a member that looks silent to it, delaying joins, stopping the call.

4.4 Authentication of group members

MLS credentials are basic credentials holding the user id; there is no certificate authority and no binding to a long-term identity key. Membership is decided by the roster the server asserts plus a key package the server relays. An active or compromised server can therefore insert itself as a member, swap a present member's leaf for its own, or steer a newcomer into a group of its own, and read the media. The safety code and the "who holds the keys" list exist so members can notice this; they do not help against a server that also serves modified web app code. This is the documented boundary, not a bug: see 8.

4.5 Other cryptographic components

ComponentDesign
Web bundle signingEd25519 over the bundle manifest; key pinned in the desktop client; offline root key can revoke a leaked CI key; security floor refuses older bundles (desktop trust model)
Desktop self-updateTauri updater (the Electron client for Linux applies the same rules), minisign signature against a pinned key distinct from the bundle key, HTTPS from github.com
Release imagescosign signature and SLSA provenance, verified by the update sidecar against the pinned workflow identity before anything is pulled
Outgoing webhooksHMAC-SHA256 over t.body, secret stored separately from the queue
Web PushRFC 8291 aes128gcm, VAPID (RFC 8292), private key in the database
Session cookiesHttpOnly, SameSite=Lax, Secure and __Host- prefix on HTTPS

5. Attack surface

Everything below is generated by scripts/gen-attack-surface.py from api/openapi.json (kept complete by TestOpenAPICoversRoutes) and the event registry (internal/events/testdata/events.golden.json). Run the script again after the routes change; --check fails when this section is stale. Only the authentication gate is listed per route. Which permission a route needs (for example Manage channels or Manage reports) is in its OpenAPI description and in roles and permissions; the permission checks live in the handlers and middleware, never in the client.

Public (unauthenticated) surface at a glance: GET /api/health, GET /api/health/live, GET /api/legal, GET /api/invites/{code}, sign-in and registration routes, password-reset request and confirmation, and POST /api/webhooks/{id}/{token} (secret in the URL, no cookie). The bot gateway and /api/bot/v1 authenticate with a bot token. /api/metrics needs a bearer token and exists only when METRICS_TOKEN is set; /api/docs and /api/openapi.json exist only when API_DOCS_ENABLED is set. Media is served only through the app, after a permission check on the channel.

REST routes

340 operations on 269 paths, grouped by API tag.

Admin (61)

MethodPathAuthenticationSummary
GET/api/admin/audit-logSession + admin-area permissionList audit log entries
GET/api/admin/bansSession + admin-area permissionList bans
POST/api/admin/categoriesSession + admin-area permissionCreate a category
PATCH/api/admin/categories/{id}Session + admin-area permissionRename a category
DELETE/api/admin/categories/{id}Session + admin-area permissionDelete a category
POST/api/admin/channelsSession + admin-area permissionCreate a channel
PATCH/api/admin/channels/{id}Session + admin-area permissionRename a channel, change its topic or member limit
DELETE/api/admin/channels/{id}Session + admin-area permissionDelete a channel
PUT/api/admin/channels/{id}/auto-deleteSession + admin-area permissionSet a channel's auto-delete
POST/api/admin/channels/{id}/auto-delete/previewSession + admin-area permissionPreview a channel's auto-delete
POST/api/admin/channels/{id}/duplicateSession + admin-area permissionDuplicate a channel
GET/api/admin/invitesSession + admin-area permissionList invites
POST/api/admin/invitesSession + admin-area permissionCreate an invite
GET/api/admin/invites/limitsSession + admin-area permissionRead the limits of member invites
PUT/api/admin/invites/limitsSession + admin-area permissionSet the limits of member invites
PATCH/api/admin/invites/{id}Session + admin-area permissionChange an invite's expiry or use limit
DELETE/api/admin/invites/{id}Session + admin-area permissionDelete an invite
PUT/api/admin/layoutSession + admin-area permissionApply category order and channel placement
GET/api/admin/mediaSession + admin-area permissionList all media, newest first
GET/api/admin/media/orphansSession + admin-area permissionCount orphaned media objects
POST/api/admin/media/orphans/cleanupSession + admin-area permissionDelete orphaned media objects
POST/api/admin/media/pruneSession + admin-area permissionDelete old chat attachments
GET/api/admin/media/statsSession + admin-area permissionStorage totals
DELETE/api/admin/media/{id}Session + admin-area permissionDelete a media object
GET/api/admin/message-settingsSession + admin-area permissionMessage settings
PUT/api/admin/message-settings/auto-deleteSession + admin-area permissionSet the auto-delete default
POST/api/admin/message-settings/auto-delete/previewSession + admin-area permissionPreview the auto-delete default
PUT/api/admin/message-settings/lifetimesSession + admin-area permissionSet the lifetime settings
GET/api/admin/statusSession + admin-area permissionOperator status
GET/api/admin/status/capacitySession + admin-area permissionServer capacity
POST/api/admin/status/websocketSession + admin-area permissionReport the admin's WebSocket
GET/api/admin/systemSession + admin-area permissionSystem status
POST/api/admin/system/checkSession + admin-area permissionCheck for a new release now
POST/api/admin/system/mail/testSession + admin-area permissionSend a test e-mail
GET/api/admin/system/opsSession + admin-area permissionOperator metrics
GET/api/admin/system/updateSession + admin-area permissionUpdate status
POST/api/admin/system/updateSession + admin-area permissionUpdate the server now
GET/api/admin/usersSession + admin-area permissionList all accounts
POST/api/admin/users/{id}/banSession + admin-area permissionBan a member
DELETE/api/admin/users/{id}/banSession + admin-area permissionLift a ban
POST/api/admin/users/{id}/disableSession + admin-area permissionDisable an account
POST/api/admin/users/{id}/enableSession + admin-area permissionRe-enable an account
POST/api/admin/users/{id}/kickSession + admin-area permissionRemove a user from voice
POST/api/admin/users/{id}/login-lockout/clearSession + admin-area permissionClear an account's login lockout
POST/api/admin/users/{id}/passwordSession + admin-area permissionSet a user's password
POST/api/admin/users/{id}/password-resetSession + admin-area permissionGenerate a temporary password
POST/api/admin/users/{id}/sessions/revokeSession + admin-area permissionRevoke all sessions of a user
PUT/api/admin/users/{id}/statusSession + admin-area permissionEdit or clear a member's status line
POST/api/admin/users/{id}/timeoutSession + admin-area permissionTime out a member
DELETE/api/admin/users/{id}/timeoutSession + admin-area permissionEnd a member's timeout
POST/api/admin/users/{id}/two-factor/resetSession + admin-area permissionReset a member's two-factor sign-in
PATCH/api/channels/{channelID}/forum-settingsSessionUpdate a forum's settings
POST/api/channels/{channelID}/tagsSessionCreate a forum tag
PUT/api/channels/{channelID}/tags/orderSessionReorder a forum's tags
PATCH/api/channels/{channelID}/tags/{tagID}SessionUpdate a forum tag
DELETE/api/channels/{channelID}/tags/{tagID}SessionDelete a forum tag

Auth (16)

MethodPathAuthenticationSummary
POST/api/auth/email/verifyNone (public)Confirm an e-mail address
POST/api/auth/loginNone (public)Sign in
POST/api/auth/login/2faNone (public)Finish a sign-in with the second factor
POST/api/auth/logoutNone (public)Sign out
POST/api/auth/logout-allSessionSign out everywhere
GET/api/auth/meSessionCurrent user
GET/api/auth/optionsNone (public)Sign-in options
PUT/api/auth/passwordSessionChange own password
POST/api/auth/password-reset/confirmNone (public)Set a new password with a reset link
POST/api/auth/password-reset/requestNone (public)Ask for a password reset mail
POST/api/auth/registerNone (public)Register with an invite code
GET/api/auth/sessionsSessionList my sessions
POST/api/auth/sessions/revoke-othersSessionSign out all my other sessions
DELETE/api/auth/sessions/{id}SessionSign out one of my sessions
GET/api/invites/{code}None (public)Preview an invite
GET/api/join/{slug}None (public)Preview a vanity invite link

Bot API (22)

MethodPathAuthenticationSummary
GET/api/bot/v1/channelsBot tokenList the bot's channels
GET/api/bot/v1/channels/{channelID}/messagesBot tokenList channel messages
POST/api/bot/v1/channels/{channelID}/messagesBot tokenPost a message
GET/api/bot/v1/channels/{channelID}/messages/{messageID}Bot tokenGet a message
PATCH/api/bot/v1/channels/{channelID}/messages/{messageID}Bot tokenEdit the bot's own message
DELETE/api/bot/v1/channels/{channelID}/messages/{messageID}Bot tokenDelete a message
PUT/api/bot/v1/channels/{channelID}/messages/{messageID}/reactions/{emoji}/@meBot tokenAdd the bot's reaction
DELETE/api/bot/v1/channels/{channelID}/messages/{messageID}/reactions/{emoji}/@meBot tokenRemove the bot's reaction
POST/api/bot/v1/channels/{channelID}/uploadBot tokenPost a message with an attachment
GET/api/bot/v1/commandsBot tokenList the bot's commands
POST/api/bot/v1/commandsBot tokenRegister a command
PUT/api/bot/v1/commandsBot tokenReplace the bot's commands
DELETE/api/bot/v1/commands/{commandID}Bot tokenDelete a command
GET/api/bot/v1/gatewayBot tokenGet the gateway address
POST/api/bot/v1/interactions/{interactionID}/{token}/callbackBot tokenRespond to an interaction
POST/api/bot/v1/interactions/{interactionID}/{token}/followupsBot tokenSend a follow-up to an interaction
PATCH/api/bot/v1/interactions/{interactionID}/{token}/messages/@originalBot tokenEdit the original response
DELETE/api/bot/v1/interactions/{interactionID}/{token}/messages/@originalBot tokenDelete the original response
GET/api/bot/v1/media/{id}Bot tokenDownload an attachment
GET/api/bot/v1/membersBot tokenList members
GET/api/bot/v1/members/{userID}Bot tokenGet a member
GET/api/bot/v1/users/@meBot tokenGet the bot itself

Bots (18)

MethodPathAuthenticationSummary
GET/api/admin/botsSession + admin-area permissionList bots
POST/api/admin/botsSession + admin-area permissionCreate a bot
POST/api/admin/bots/installSession + admin-area permissionAdd a bot from a manifest
POST/api/admin/bots/manifests/checkSession + admin-area permissionCheck a bot manifest
GET/api/admin/bots/{id}Session + admin-area permissionGet a bot
PATCH/api/admin/bots/{id}Session + admin-area permissionChange a bot
DELETE/api/admin/bots/{id}Session + admin-area permissionDelete a bot
GET/api/admin/bots/{id}/accessSession + admin-area permissionShow what a bot can reach
PUT/api/admin/bots/{id}/accessSession + admin-area permissionChoose which channels a bot works in
POST/api/admin/bots/{id}/avatarSession + admin-area permissionSet a bot's avatar
GET/api/admin/bots/{id}/commandsSession + admin-area permissionList a bot's commands
PATCH/api/admin/bots/{id}/commands/{commandID}Session + admin-area permissionEnable or disable a command
DELETE/api/admin/bots/{id}/commands/{commandID}Session + admin-area permissionDelete a command
PUT/api/admin/bots/{id}/rolesSession + admin-area permissionSet a bot's roles
GET/api/admin/bots/{id}/tokensSession + admin-area permissionList a bot's tokens
POST/api/admin/bots/{id}/tokensSession + admin-area permissionIssue a token
DELETE/api/admin/bots/{id}/tokens/{tokenID}Session + admin-area permissionRevoke a token
POST/api/admin/bots/{id}/tokens/{tokenID}/regenerateSession + admin-area permissionRegenerate a token

Channels (19)

MethodPathAuthenticationSummary
PUT/api/categories/{id}/notificationsSessionSet category notification level
GET/api/channelsSessionList categories and channels
GET/api/channels/{channelID}/first-unreadSessionFirst unread message of a channel
GET/api/channels/{channelID}/forum-settingsSessionGet a forum's settings
PUT/api/channels/{channelID}/notificationsSessionSet channel notification level
GET/api/channels/{channelID}/postsSessionList forum posts
GET/api/channels/{channelID}/posts/unreadSessionList a forum's unread posts
PUT/api/channels/{channelID}/posts/{messageID}/readSessionMark a forum post read
POST/api/channels/{channelID}/readSessionMark a channel read
GET/api/channels/{channelID}/tagsSessionList forum tags
POST/api/channels/{channelID}/unreadSessionMark unread from a message
POST/api/channels/{id}/voice/{userID}/disconnectSessionDisconnect a member from a Talk
POST/api/dms/{id}/talkSessionStart a Talk from a direct message conversation
GET/api/notification-settingsSessionNotification settings
PUT/api/notification-settings/serverSessionSet server notification default
GET/api/read-stateSessionUnread summary of all channels
POST/api/read-state/read-allSessionMark everything read
PATCH/api/talks/{id}SessionRename a temporary Talk or set its member limit
POST/api/talks/{id}/invitesSessionInvite a conversation to a Talk

Commands (5)

MethodPathAuthenticationSummary
GET/api/channels/{channelID}/commandsSessionList the slash commands of a channel
POST/api/channels/{channelID}/ephemeral/{noteID}/componentsSessionUse a component of an ephemeral note
POST/api/channels/{channelID}/interactionsSessionRun a slash command
POST/api/channels/{channelID}/messages/{messageID}/componentsSessionUse a component of a bot's message
POST/api/channels/{channelID}/modals/{modalID}SessionSubmit a modal

Direct messages (9)

MethodPathAuthenticationSummary
GET/api/dmsSessionList direct messages
POST/api/dmsSessionStart a direct message
PUT/api/dms/favouritesSessionSet the caller's favourite direct messages
GET/api/dms/{id}SessionGet a direct message
PUT/api/dms/{id}/hiddenSessionClose or reopen a direct message in the caller's list
POST/api/dms/{id}/membersSessionAdd a member to a group direct message
DELETE/api/dms/{id}/members/{user}SessionLeave a group direct message
GET/api/users/me/dm-settingsSessionRead the direct message settings
PUT/api/users/me/dm-settingsSessionSet the direct message settings

Emoji (6)

MethodPathAuthenticationSummary
POST/api/admin/emojiSession + admin-area permissionAdd a custom emoji
PATCH/api/admin/emoji/{id}Session + admin-area permissionRename a custom emoji
DELETE/api/admin/emoji/{id}Session + admin-area permissionDelete a custom emoji
GET/api/emojiSessionList custom emoji
GET/api/emoji/{id}/imageSessionCustom emoji image
GET/api/emoji/{id}/stillSessionCustom emoji still frame

Events (24)

MethodPathAuthenticationSummary
GET/api/event-role-templatesSessionList role templates
POST/api/event-role-templatesSessionCreate a role template
PATCH/api/event-role-templates/{templateID}SessionUpdate a role template
DELETE/api/event-role-templates/{templateID}SessionDelete a role template
GET/api/eventsSessionList community events
POST/api/eventsSessionCreate a community event
GET/api/events/{id}SessionGet a community event
PATCH/api/events/{id}SessionUpdate a community event
POST/api/events/{id}/cancelSessionCancel a community event
GET/api/events/{id}/commentsSessionList the comments of an event
POST/api/events/{id}/commentsSessionComment on an event
PUT/api/events/{id}/comments/muteSessionMute or unmute the notices of an event's comments
PATCH/api/events/{id}/comments/{commentID}SessionEdit a comment
DELETE/api/events/{id}/comments/{commentID}SessionDelete a comment
POST/api/events/{id}/coverSessionUpload event cover image
DELETE/api/events/{id}/coverSessionRemove event cover image
PUT/api/events/{id}/rolesSessionSet the roles of an event
DELETE/api/events/{id}/roles/{roleID}SessionDelete a role of an event
PUT/api/events/{id}/rsvpSessionSet RSVP response for an event
DELETE/api/events/{id}/rsvpSessionDelete RSVP response for an event
GET/api/events/{id}/rsvpsSessionList RSVPs for an event
PUT/api/events/{id}/signupSessionSign up for a role of an event
DELETE/api/events/{id}/signupSessionLeave the role of an event
DELETE/api/events/{id}/signup/{userID}SessionRemove a member from the roster of an event

Invites (3)

MethodPathAuthenticationSummary
GET/api/me/invitesSessionList my invites
POST/api/me/invitesSessionCreate an invite as a member
DELETE/api/me/invites/{id}SessionRevoke my invite

Media (6)

MethodPathAuthenticationSummary
POST/api/channels/{channelID}/attachmentsSessionUpload an attachment for a message
DELETE/api/channels/{channelID}/attachments/{id}SessionRemove an unsent attachment
POST/api/channels/{channelID}/uploadSessionPost a message with an attachment
GET/api/link-previewSessionLink card for a URL
GET/api/link-preview/imageSessionProxied preview image
GET/api/media/{id}SessionDownload a media object

Messages (33)

MethodPathAuthenticationSummary
GET/api/channels/{channelID}/messagesSessionList channel messages
POST/api/channels/{channelID}/messagesSessionPost a message
POST/api/channels/{channelID}/messages/bulk-deleteSessionDelete a member's latest messages in a channel
PUT/api/channels/{channelID}/messages/{messageID}SessionEdit own message
DELETE/api/channels/{channelID}/messages/{messageID}SessionDelete a message
GET/api/channels/{channelID}/messages/{messageID}/editsSessionEdit history of a message
PUT/api/channels/{channelID}/messages/{messageID}/pinSessionPin a message
DELETE/api/channels/{channelID}/messages/{messageID}/pinSessionUnpin a message
GET/api/channels/{channelID}/pinsSessionList a channel's pinned messages
POST/api/channels/{channelID}/postsSessionCreate a forum post
PATCH/api/channels/{channelID}/posts/{messageID}SessionUpdate a forum post
POST/api/channels/{channelID}/scheduledSessionSchedule a message
GET/api/me/remindersSessionList my reminders
POST/api/me/remindersSessionRemind me of a message
DELETE/api/me/reminders/{reminderID}SessionCancel or dismiss a reminder
GET/api/me/savedSessionList saved messages
PUT/api/me/saved/{messageID}SessionSave a message
DELETE/api/me/saved/{messageID}SessionRemove a saved message
GET/api/me/scheduledSessionList my scheduled messages
PATCH/api/me/scheduled/{scheduledID}SessionEdit a scheduled message
DELETE/api/me/scheduled/{scheduledID}SessionDelete a scheduled message
GET/api/mentionsSessionMentions inbox
GET/api/message-settingsSessionLifetime settings
GET/api/messages/by-number/{number}SessionGet a message by its link number
GET/api/messages/{messageID}/forwardSessionRead the original of a forwarded message
POST/api/messages/{messageID}/forwardSessionForward a message
POST/api/messages/{messageID}/reactionsSessionToggle a reaction
POST/api/messages/{messageID}/reportSessionReport a message
GET/api/messages/{messageID}/threadSessionGet a thread
GET/api/searchSessionSearch messages
GET/api/threads/followingSessionThreads the caller follows
PUT/api/threads/{threadID}/followSessionFollow a thread
DELETE/api/threads/{threadID}/followSessionUnfollow a thread

Moderation (13)

MethodPathAuthenticationSummary
GET/api/admin/automodSession + admin-area permissionList automod rules
POST/api/admin/automod/rulesSession + admin-area permissionCreate an automod rule
PUT/api/admin/automod/rules/{id}Session + admin-area permissionReplace an automod rule
DELETE/api/admin/automod/rules/{id}Session + admin-area permissionDelete an automod rule
GET/api/admin/reportsSession + admin-area permissionList reports
GET/api/admin/reports/open-countSession + admin-area permissionCount open reports
GET/api/admin/reports/{id}Session + admin-area permissionGet a report
DELETE/api/admin/reports/{id}Session + admin-area permissionDelete a report
POST/api/admin/reports/{id}/resolveSession + admin-area permissionClose a report
GET/api/admin/users/{id}/moderationSession + admin-area permissionA member's moderation history
POST/api/admin/users/{id}/moderation/notesSession + admin-area permissionAdd a moderator note
PUT/api/admin/users/{id}/moderation/notes/{noteID}Session + admin-area permissionEdit your moderator note
DELETE/api/admin/users/{id}/moderation/notes/{noteID}Session + admin-area permissionDelete a moderator note

Onboarding (15)

MethodPathAuthenticationSummary
PUT/api/admin/community/iconSession + admin-area permissionSet the community icon
DELETE/api/admin/community/iconSession + admin-area permissionRemove the community icon
GET/api/admin/first-runSession + admin-area permissionRead the Owner's first-run checklist
PUT/api/admin/first-runSession + admin-area permissionClose, reopen or advance the Owner's first-run checklist
GET/api/admin/onboardingSession + admin-area permissionRead the onboarding settings
PUT/api/admin/onboardingSession + admin-area permissionSet the onboarding settings
GET/api/admin/raidSession + admin-area permissionRead the raid protection settings
PUT/api/admin/raidSession + admin-area permissionSet the raid protection settings
PUT/api/admin/raid/lockdownSession + admin-area permissionSwitch the lockdown on or off
GET/api/admin/vanity-linkSession + admin-area permissionRead the vanity invite link
PUT/api/admin/vanity-linkSession + admin-area permissionSet or remove the vanity invite link
GET/api/community/iconNone (public)Community icon
GET/api/onboardingSessionRead the onboarding state
POST/api/onboarding/acceptSessionAccept the community rules
PUT/api/onboarding/tipsSessionStore dismissed tips

Polls (4)

MethodPathAuthenticationSummary
GET/api/channels/{channelID}/messages/{messageID}/pollSessionGet a poll with my votes
POST/api/channels/{channelID}/messages/{messageID}/poll/endSessionEnd a poll early
PUT/api/channels/{channelID}/messages/{messageID}/poll/votesSessionVote in a poll
POST/api/channels/{channelID}/pollsSessionPost a poll

Push (6)

MethodPathAuthenticationSummary
GET/api/push/configSessionWeb Push settings
PUT/api/push/preferencesSessionSet the push preview
GET/api/push/subscriptionsSessionList my push devices
POST/api/push/subscriptionsSessionTurn push on for this device
DELETE/api/push/subscriptions/{id}SessionTurn push off for a device
POST/api/push/testSessionSend a test push

Realtime (4)

MethodPathAuthenticationSummary
GET/api/webrtc/configSessionICE servers for voice
POST/api/webrtc/probeSessionAnswer a connection test
GET/api/webrtc/speedtestSessionDownload speed test
GET/api/wsSession cookie on upgrade, origin checkWebSocket event stream

Roles (17)

MethodPathAuthenticationSummary
GET/api/admin/categories/{id}/overwritesSession + admin-area permissionPermission overwrites of a category
PUT/api/admin/categories/{id}/overwritesSession + admin-area permissionReplace the permission overwrites of a category
GET/api/admin/channels/{id}/overwritesSession + admin-area permissionPermission overwrites of a channel
PUT/api/admin/channels/{id}/overwritesSession + admin-area permissionReplace the permission overwrites of a channel
POST/api/admin/rolesSession + admin-area permissionCreate a role
PUT/api/admin/roles/orderSession + admin-area permissionReorder roles
PATCH/api/admin/roles/{id}Session + admin-area permissionEdit a role
DELETE/api/admin/roles/{id}Session + admin-area permissionDelete a role
GET/api/admin/roles/{id}/view-asSession + admin-area permissionView the server as a role
GET/api/admin/upload-limitSession + admin-area permissionRead the global upload limit
PUT/api/admin/upload-limitSession + admin-area permissionSet the global upload limit
PUT/api/admin/users/{id}/rolesSession + admin-area permissionSet a member's roles
PUT/api/me/roles/{id}SessionTake a self-assignable role
DELETE/api/me/roles/{id}SessionDrop a self-assignable role
GET/api/permissionsSessionOwn permissions
GET/api/rolesSessionList roles
GET/api/roles/self-assignableSessionList the roles members can choose

System (6)

MethodPathAuthenticationSummary
GET/api/docsNone (public)API reference
GET/api/healthNone (public)Health check
GET/api/health/liveNone (public)Liveness check
GET/api/legalNone (public)Operator details and privacy-policy facts
GET/api/metricsBearer METRICS_TOKENPrometheus metrics
GET/api/openapi.jsonSessionThis API description

Users (27)

MethodPathAuthenticationSummary
GET/api/me/blocksSessionList blocked members
PUT/api/me/blocks/{userID}SessionBlock a member
DELETE/api/me/blocks/{userID}SessionUnblock a member
GET/api/membersSessionList members
POST/api/users/me/2fa/disableSessionTurn off two-factor sign-in
POST/api/users/me/2fa/enableSessionTurn on two-factor sign-in
POST/api/users/me/2fa/recovery-codesSessionMake new recovery codes
POST/api/users/me/2fa/setupSessionStart two-factor setup
GET/api/users/me/accountSessionOwn sign-in security
POST/api/users/me/avatarSessionUpload avatar
DELETE/api/users/me/avatarSessionRemove avatar
PUT/api/users/me/bannerSessionSet the profile banner
DELETE/api/users/me/bannerSessionRemove the profile banner
POST/api/users/me/deleteSessionDelete my account
PUT/api/users/me/emailSessionSet the account's e-mail address
POST/api/users/me/email/removeSessionRemove the account's e-mail address
POST/api/users/me/email/resendSessionSend the confirmation mail again
GET/api/users/me/exportSessionStatus of my data export
POST/api/users/me/exportSessionRequest an export of my data
GET/api/users/me/export/downloadSessionDownload my data export
PUT/api/users/me/localeSessionSet UI language
PUT/api/users/me/presenceSessionChoose presence
GET/api/users/me/privacySessionRead the privacy settings
PUT/api/users/me/privacySessionSet the privacy settings
PUT/api/users/me/profileSessionUpdate display name and bio
PUT/api/users/me/statusSessionSet own status line
GET/api/users/{userID}SessionGet a user profile

Webhooks (14)

MethodPathAuthenticationSummary
GET/api/admin/channels/{id}/outgoing-webhooksSession + admin-area permissionList a channel's outgoing webhooks
POST/api/admin/channels/{id}/outgoing-webhooksSession + admin-area permissionCreate an outgoing webhook
GET/api/admin/channels/{id}/webhooksSession + admin-area permissionList a channel's webhooks
POST/api/admin/channels/{id}/webhooksSession + admin-area permissionCreate a webhook
PATCH/api/admin/outgoing-webhooks/{id}Session + admin-area permissionChange an outgoing webhook
DELETE/api/admin/outgoing-webhooks/{id}Session + admin-area permissionDelete an outgoing webhook
GET/api/admin/outgoing-webhooks/{id}/deliveriesSession + admin-area permissionRecent deliveries of an outgoing webhook
POST/api/admin/outgoing-webhooks/{id}/secretSession + admin-area permissionRotate an outgoing webhook's secret
POST/api/admin/outgoing-webhooks/{id}/testSession + admin-area permissionSend a test event
PATCH/api/admin/webhooks/{id}Session + admin-area permissionChange a webhook
DELETE/api/admin/webhooks/{id}Session + admin-area permissionDelete a webhook
POST/api/admin/webhooks/{id}/avatarSession + admin-area permissionSet a webhook's avatar
POST/api/admin/webhooks/{id}/tokenSession + admin-area permissionRegenerate a webhook's token
POST/api/webhooks/{id}/{token}Secret in URLExecute a webhook

Wiki (12)

MethodPathAuthenticationSummary
GET/api/wikiSessionList the wiki pages
POST/api/wiki/imagesSessionUpload a wiki picture
POST/api/wiki/pagesSessionCreate a wiki page
GET/api/wiki/pages/{slug}SessionGet a wiki page
PATCH/api/wiki/pages/{slug}SessionChange a wiki page
DELETE/api/wiki/pages/{slug}SessionDelete a wiki page
PUT/api/wiki/pages/{slug}/lockSessionLock or unlock a wiki page
GET/api/wiki/pages/{slug}/revisionsSessionList the revisions of a wiki page
GET/api/wiki/pages/{slug}/revisions/{rev}SessionGet one revision of a wiki page
POST/api/wiki/pages/{slug}/revisions/{rev}/restoreSessionRestore a revision of a wiki page
POST/api/wiki/pages/{slug}/undeleteSessionRestore a deleted wiki page
GET/api/wiki/searchSessionSearch the wiki

WebSocket events

Server to client (67): block_updated, cameras_capped, channels_changed, community_updated, dm_call_end, dm_call_ring, dm_created, dm_member_added, dm_member_removed, dm_prefs_updated, emoji_changed, event_comment, event_comment_notice, event_delete, event_reminder, event_signup_notice, event_update, forum_tags_changed, interaction_deferred, interaction_modal, interaction_response, kicked, media_handshake, member_joined, member_roles_changed, member_timeout, message_create, message_delete, message_reaction, message_update, moderation_notice, permissions_changed, pins_changed, poll_update, pong, presence_snapshot, presence_update, read_state, reminder_update, reports_open, roles_changed, saved_updated, scheduled_message_update, screen_layers, screen_layers_wanted, screen_quality, screen_viewers, server_info, stage_hand, stage_talk_power, system_update, talk_leader, talk_moved, talk_reaction, thread_follow_updated, typing, user_stats, user_update, voice_kicked, voice_mute_state, voice_rooms, voice_snapshot, voice_speaking, voice_state_update, webrtc_candidate, webrtc_media_state, webrtc_offer

Client to server (20): media_handshake, ping, presence_idle, stage_hand, stage_talk_power, talk_reaction, typing, voice_join, voice_leave, voice_mute_state, voice_speaking, webrtc_answer, webrtc_camera_stop, webrtc_candidate, webrtc_diag, webrtc_layers, webrtc_request_keyframe, webrtc_screenshare_start, webrtc_screenshare_stop, webrtc_subscribe

6. Security controls

ControlWhereSummary
Permission engineinternal/perm, roles and permissionsIn-memory snapshot resolving @everyone, roles, category and channel overwrites and the hierarchy; Owner shortcut; timeouts and rules acceptance folded in; fails closed when the snapshot cannot load. Enforced in REST, the WebSocket hub (per-recipient delivery), search, push, mention inbox, export, media download and the SFU (speak/video)
DM isolationperm.Snapshot.Channel, dm_membersMembership is the only rule, evaluated before the Owner shortcut; search re-tests membership in SQL; DM events never reach webhooks or the bot gateway
Sessionsinternal/authPer-user token_version and per-token id revocation; own device list and inactivity sign-out (1.2); sockets re-checked every two minutes
Rate limitsinternal/httpx, per featurePer-IP and per-user limits on REST, uploads, previews, search (two concurrent per user, 5 s statement timeout), login lockout tiers, WebSocket frame/event/handshake budgets per connection and per account, voice and Talk creation gates
Origin / CSRFhttpx.CSRFExceptState-changing requests and upgrades need an allowed Origin; exceptions are the webhook execute route and the bot API, which ignore cookies
CSP and headershttpx.SecurityHeadersNo inline script or style (style-src-attr 'none'), no eval, frame-ancestors 'none', nosniff, referrer policy; e2e specs fail on any violation
Content handlinginternal/media, imgprocContent sniffing, risky types served only as downloads, filename sanitising; avatars, banners and emoji decoded, bounded, re-encoded; pixel and frame budgets before decode
Markdown renderingweb/src/shared/lib/markdownTree of plain nodes, one HTML-producing module with attribute allow-lists; single v-html sink
Audit loginternal/auditAppend-only (database trigger), visibility classes per action (VisAuditors, VisChannel, VisPrivileged), no secrets stored, scrub-only update path for account erasure
Account erasure and exportinternal/auth/accountTransactional, foreign-key-walking test, tombstone row guarded by a trigger
Automod (1.2)internal/automodWord, link, raid and new-account rules, hits go to the report queue; DM hits store no text
Secretsinternal/configEnvironment only; production refuses weak secrets and wide trusted-proxy ranges; gitleaks in CI and pre-commit
Supply chain.github/workflows, scripts/Actions pinned by commit SHA, compose images pinned by digest, cosign and provenance, govulncheck, npm audit, third-party notices
Fuzzingmake fuzz-smokeGo fuzz targets for the search builder, mentions, durations, URL guard and version parser run in CI for seconds each; seed corpora run with every go test

7. Results of internal reviews

All reviews so far are internal. They are listed so you can see what was looked at and what was found, and so you can spend your time elsewhere.

7.1 1.0 six-lane review (2026-10-22)

Lanes: authentication and sessions, permissions, content and media, integrations (webhooks, bots, link previews), realtime and voice, supply chain. Result: 0 Critical, 0 High; every Medium was fixed with a test that fails without the fix. Deferred past 1.0: frame version 2 (now removed in 1.2) and regroup collusion (Info). Closed in 1.1: image signature check in the updater, binding of legacy bot tokens. Source: SECURITY.md. Not covered by that review (reviewed per change by automated security review and a verification pass): e-mail/SMTP, two-factor sign-in, account deletion and export, Web Push and the installable app.

7.2 1.1 review

Read-only review of the 1.1.0 diff (direct messages, blocks, reports, forwarding, saved messages, GIF proxy, avatar and banner re-encoding, migrations 0049 to 0058) in five areas: authorization/IDOR, information leaks, media processing and GDPR, migrations, input validation/rate limits/SQL. Result: no exploitable finding. It was done by an automated model-based reviewer and spot-checked by the maintainers, so treat it as lower assurance than a human review.

7.3 1.2 review (diff v1.1.0 to a 1.2 integration snapshot)

Read-only review of the new routes, WebSocket events, stores and migrations, with one targeted test run through a build overlay. No Critical or High findings. The status column follows the 1.2 security-fix pull request. Items are described by class, without exploit detail.

IDSeverityAreaClassStatus at time of writing
M1MediumAutomod linksBlock-list mode can be bypassed by input the renderer links but the rule does not parseFixed in the 1.2 security-fix pull request, with tests
M2MediumEventsBroadcast payload carries a viewer-specific field of the acting memberFixed in the 1.2 security-fix pull request, with tests
M3MediumErasure/exportA new messages column is missing from account erasure and data exportFixed in the 1.2 security-fix pull request, with tests
L1LowAutomodSome user-supplied text fields (forum titles and tags, poll options) are not screenedFixed in the 1.2 security-fix pull request, with tests
L2LowAutomodWord matching does not normalise all Unicode look-alikes and invisible charactersFixed in the 1.2 security-fix pull request, with tests
L3LowScheduled sendBurst of due messages bypasses the request limiterAccepted, documented in SECURITY.md
L4LowPollsLive tallies allow inference in very small audiences (by design, documented hardening)Broadcast tally fixed; counts shown by REST documented as accepted
L5LowForumAuthor edit rights on title and tags are looser than send rightsFixed in the 1.2 security-fix pull request, with tests
L6LowEventsRetraction of hidden events can be missed on two quick permission changes (REST stays correct)Fixed in the 1.2 security-fix pull request, with tests
L7LowTalksPresence queue overflow or timer race can mis-remove or keep a TalkFixed in the 1.2 security-fix pull request, with tests
L10LowPrivacyMinor leftovers after erasure/export (names in import records, notes, filenames of removed attachments in search)Fixed in the 1.2 security-fix pull request, with tests
L11LowModerationTimeline readable with Manage reports only, wider than audit classesAccepted, documented in SECURITY.md
L12InfoWebSocketPer-user gates reset on reconnect, bounded by the connect limiterAccepted, documented in SECURITY.md

7.4 Fuzzing and malformed-input review of the media paths (1.2, work package L)

  • Web property tests (fixed seed): random, truncated, extended, bit-flipped and old-format media units, capability and heartbeat messages and MLS handshake wires are refused without throwing.
  • Finding W1 (fixed). decodeMlsMessage throws on some malformed input and MediaGroup.receive did not catch it, so a bad handshake wire rejected the call's inbox drain and skipped the sync after it. Any member (or the relay) could make a call's key exchange stall by repeating this. Availability only; no confidentiality impact. Fixed with a regression test.
  • Review of MLS, rotation and signature-domain paths (read-only, by the maintainers; no crypto change made):
IDResult
R1Signature domains are distinct (frame-v2, frame-v3, heartbeat-v1, capability-v1, safety-v1), none is a prefix of another, and the version magic byte is inside the signed bytes. No cross-protocol or cross-version signature confusion found. A version 2 and a version 3 AAD can coincide byte for byte for suitable clear bytes, but the signature domain and the signed magic keep the units apart.
R2Heartbeat signed bytes are domain || groupId || userId || body without length prefixes. Safe today because the group id is 16 bytes and a user id is validated to a fixed 36-character form; a future variable-length field would need prefixes. Recommended hardening, no current exploit.
R3Downgrade: while version 2 existed, a server withholding capability messages could keep an epoch on version 2 (no line binding). Version 2 is no longer opened in 1.2 (such units are only counted); no acceptV2 path remains in the frame code.
R4A removed member's frames under the previous epoch still open at receivers for up to 30 seconds (documented; trade-off for media continuity). The removed member also keeps the old key; the server must stop forwarding.
R5An unauthenticated "commit for a newer epoch" from one of the two lowest members makes a receiver start over (rate limited with backoff). A server can do this anyway; availability only. Already listed as RT-6.
R6Rival-welcome handling can pull a member out of a group into one built on a key package it offered; rate limited per sender with doubling backoff; requires holding that key package (the relay can). Documented.
R7The relay-asserted sender id of a handshake wire is not itself authenticated. Commits and welcomes are MLS-signed; checks bind the signer leaf to the claimed sender for commits. For welcomes the code accepts the group if the claimed sender is the expected committer and the welcome decrypts under the offered key package; it does not additionally require the welcome's signer to equal the claimed sender. Auditors should confirm this cannot admit a member into a group the claimed committer did not create (the group's own signatures cover this inside ts-mls; worth a targeted test).
R8AES-GCM with random 96-bit nonces under a per-epoch key shared by all members stays below the NIST 2^32 limit by a wide margin at the sizes documented in SECURITY.md; this relies on rotation, which needs at least two members (a lone sender keeps its key until someone joins; nobody else holds it).
R9Heartbeats and capability messages are bound to the group and user id; a captured heartbeat can only be replayed if its time exceeds the last accepted one, so withholding and later delivering the newest heartbeat can delay a removal by the withheld interval. No privilege gain.
R10Key lifetime hooks (__MNEMA_MEDIA_TIMING__) are compiled out of release builds (__MNEMA_E2E_HOOKS__ is false), so a page cannot shorten lifetimes in production.

8. Known limitations and accepted risks

  • Text is server-readable. By design and final. Do not report as a finding.
  • Active or compromised server versus media E2EE (section 4.4): can insert a member, swap a leaf, split a call, drop messages, see metadata. The safety code is the only in-product detection. Browsers receive their code from the server, so a malicious server can serve a modified app.
  • Desktop client: removes the server-served-code risk by running a signed bundle, but installers are unsigned at OS level (no Authenticode, notarisation or package signature); the bundle and updater signing keys are the trust anchors; Beta status.
  • Frame metadata: codec headers (H.264 parameter sets and slice headers, VP8 payload header) are readable but authenticated; sizes and timing are visible to the SFU.
  • Replay by an active SFU: a layer stream a receiver has not seen yet starts a fresh replay window, so an active SFU can feed old recorded frames of that layer.
  • Rate limiter and lockout state is in memory and cleared on restart.
  • Hub state is in memory (presence, voice rooms and the 30 second voice grace, bot gateway sessions and their replay buffers): a restart drops it and clients reconnect and rejoin.
  • Regroup nuisance (RT-6): two colluding members or the server can make a quiet listener regroup; backoff-limited.
  • Audit log keeps deleted and edited message text for moderation; operators must name it in their privacy notice. Direct messages are not audited.
  • Attachments are not re-encoded. JPEG, PNG and WebP attachments lose their metadata (EXIF with GPS, XMP, IPTC, comments; internal/imgstrip, an Owner switch that is on by default since migration 0101); other files, and pictures stored before that, keep theirs. Only avatars, banners and emoji are re-encoded.
  • Self-update sidecar: the Docker socket is root on the host; the sidecar is off by default and described in SECURITY.md.
  • No external audit, no bug bounty. Reporting: SECURITY.md.

9. Building and running a test instance

Use a throwaway machine. Please do not test against instances you do not run.

git clone https://github.com/mnema-xyz/talk && cd talk
git checkout v3.3.10                  # or the latest release tag
cp .env.example .env
# Set at least: PUBLIC_URL (for example http://127.0.0.1:8080), JWT_SECRET (32+ random bytes),
# POSTGRES_PASSWORD, S3_ACCESS_KEY, S3_SECRET_KEY, ADMIN_INITIAL_PASSWORD.
# For a local plain-HTTP test use APP_ENV=development (listens on loopback only);
# set API_DOCS_ENABLED=true to get /api/docs and /api/openapi.json,
# and METRICS_TOKEN to get /api/metrics.
make docker                           # builds mnema-talk:local
make up                               # docker compose up -d (app, postgres, seaweedfs; coturn and the update sidecars are optional)
curl -s http://127.0.0.1:8080/api/health

Sign in as the Owner (admin by default, password from ADMIN_INITIAL_PASSWORD), create invite codes and register more accounts to test roles and cross-account cases. A published image can be used instead of building (MNEMA_IMAGE=ghcr.io/mnema-xyz/mnema-talk:<version>); deployment lists the steps CI runs. For development without the image: make dev starts PostgreSQL and SeaweedFS in Docker and runs the server with go run; the web app builds with make web (or the Vite dev server in web/).

Useful commands: make test (unit), make test-integration (needs Docker or TEST_DATABASE_URL), make test-web, make fuzz-smoke (FUZZTIME=60s for a longer run), make lint, make vuln, make openapi-check. Call and camera end-to-end specs run in CI only (they need media devices); they are not required to review the code.

The mail path needs an SMTP server (SMTP_HOST); without it no mail is sent. Web Push needs a browser and HTTPS. Roles and overwrites for permission tests are set up in the admin console after the first sign-in.

10. Suggested test focus

Ordered by where a mistake would cost most.

  1. Permission enforcement parity. For each new 1.2 feature (Talks and leaders, Stages, events on hidden Talks, forum posts, reminders, scheduled send, sessions, view-as-role), check the same decision is made in REST, the WebSocket hub (events must not reach members who cannot view), search, push, export and the SFU. Try IDOR on every {id} route with an account that lacks view access, and compare 403 versus 404 answers. 2. Direct-message isolation. Owner and Administrators must not read or learn existence of conversations they are not in; check search, mentions, unread counts, sequence numbers, push, reports, export and webhooks/bots. 3. Media E2EE. Frame parser and replay window (web/src/lib/media/frame.ts, replay.ts, mediaFrameWorker.ts); MLS commit vetting (groupVet.ts, group.ts, groupTree.ts); confusion between handshake message types; behaviour with hostile members (second leaf, forged remove/add, early removal), reloads and races; signature domains (R1, R2); welcome sender binding (R7). The tests under web/src/lib/media/*.test.ts are a map of what is already pinned. 4. Authentication. Sign-in lockout and the known-device cookie, TOTP brute-force budget, reset and confirmation tokens, session revocation paths (password change, log out everywhere, admin revoke, inactivity), bot token binding to the issuer. 5. Uploads and storage. Content sniffing versus served type, image decoders (avatar, banner, emoji, event covers: bomb limits), quotas, path handling, download authorisation after a permission change. 8. Desktop trust chain. Bundle manifest verification, security floor and rollback protection, proxy cookie handling, the Tauri updater, revocation statements (desktop trust model); the shell itself is in the separate repository. 9. Supply chain and release. Workflow permissions and pinning, image signing and provenance verification in the update sidecar, compose defaults (published ports, network separation). 10. Data lifecycle. Account erasure and export completeness for every new column and table (M3, L10), retention jobs, auto-delete, audit scrubbing.

Something wrong on this page? Tell us.