Releases

Release notes for the Go reimplementation of Bongo, newest first. Each version links to its GitHub release, where the Linux amd64 and arm64 zips are published. Container images are published at ghcr.io/bongo-project/bongo.

Versions 0.1 to 0.6 of the original C project (2007-2010) have their own release notes under History. The version numbers below belong to the Go project.

v0.15.0calendar

Latest minor

Recurrence and events

  • Recurring events expand in their own time zone, so a weekly 09:00 meeting stays at 09:00 local time across daylight-saving changes. Exceptions and moved occurrences from CalDAV clients are honoured in the web views, free/busy and reminders.
  • Links to a single occurrence work. Old links of the form {guid}_{date} redirect.
  • An occurrence can be edited or deleted on its own, or with all the occurrences that follow it.
  • Calendars created in the web carry their repeat rule, exceptions and time zone over CalDAV and in export.
  • The event form refuses an end that is not after the start, and has Show as Busy or Free. All-day events default to Free, timed events to Busy. All-day events marked Busy block the whole day in free/busy.

Sharing

  • Calendars can be shared at free/busy, read or write level. The recipient accepts or declines at /calendar/shares; an email tells them.
  • Read and write shares appear in the calendar views under Shared with me and as CalDAV collections. With write access the recipient can create, edit and delete events, which are stored under the owner.
  • The event form shows each local attendee’s busy times for the day, for attendees whose free/busy the user may see.

Invitations and scheduling

  • Events with attendees send invitations, updates and cancellations by email (iMIP) from the web calendar.
  • Received invitations show a card in the message view with Accept, Tentative and Decline. Replies update the attendee’s status on the organiser’s event.
  • CalDAV clients send and answer invitations through the server (RFC 6638 scheduling). Apple Calendar and Thunderbird stop sending invitation mail themselves, so each attendee receives one message.
  • Rooms and equipment are accounts that cannot sign in. They accept or decline booking requests from their own domain automatically.

Publishing, subscriptions, tasks and reminders

  • A calendar can be published as a read-only iCalendar feed behind a random link, with full or busy detail and an optional password.
  • External iCalendar feeds can be subscribed to and are refreshed every 6 hours. The fetcher refuses private and internal network addresses.
  • Tasks sync over CalDAV in a Tasks calendar and have a page at /tasks. Tasks with a due date appear in the agenda.
  • Reminders can be sent by email. A reminder is recorded before it is sent and is not sent twice after a restart.

Mail

  • New views: To me, From my contacts and Mailing lists.
  • /summary shows unread mail to me, starred mail, invitations awaiting an answer and the next seven days. It is the default start page; the inbox can be chosen instead in settings.
  • The relay passes DSN and SIZE parameters on and sends SMTPUTF8 when a message needs it. Bounces and delivery reports to senders outside the server are now sent, within the existing limits.

CalDAV and CardDAV setup

  • /.well-known/caldav and /.well-known/carddav answer every request method. Apple Calendar and Contacts could not add an account before.
  • The Apple configuration profile (/.well-known/mobileconfig) also sets up calendar and contacts accounts.
  • docs/caldav-carddav.md covers account details, discovery, client setup and troubleshooting.

Upgrade notes

  • Upgrade every node before anyone shares a calendar. An older node stores sharing commands as plain keys and holds no grants.
  • New key: calendar.subscriptions.allow_private_networks (default false, a restart key).
  • The default start page is now the summary page; set the preference to inbox to keep the old behaviour.
  • Existing shares are migrated once to free/busy grants on the leader.
  • Outbound relay now applies go-smtp’s command and submission timeouts.
  • For discovery by email address, the email domain’s own website must redirect /.well-known/caldav and /.well-known/carddav to the mail host for every request method, or the domain needs the SRV and TXT records the admin DNS guide lists.
  • Browsers keep the old bongo.js until the asset version changes; reload open tabs after upgrading.

Full notes: prds/done/prd-calendar-v0.15.md and prds/verification.md.

Patch releases

v0.15.14 per-component log levels

Logging

  • New setting telemetry.log.components sets the log level for single components, for example imap = “debug”, while telemetry.log.level sets it for the rest of the server. Levels are debug, info, warn and error. A component is the prefix of a log message, such as imap or smtp; deploy/LOGGING.md lists all 65. Errors are always logged. The setting applies on a reload, without a restart. An unknown component or level stops the server from starting, and makes a reload fail without changing anything.
  • Successful IMAP, POP3 and SMTP logins are now logged at info, with the user ID and source IP address. Earlier versions logged them at debug. Failed logins are still logged at warn. Passwords are never logged.

Load test

  • The final load test run uses the same shared-vCPU machine type as the first run, because most installations run on a VM or in a container. A run on dedicated hardware is the last test of the series. docs/load-testing.md and docs/design/load-test-results.md are updated.

Upgrade notes

  • No configuration changes are required. Without telemetry.log.components, logging behaves as before, apart from the login lines at info.

v0.15.14 on GitHub

v0.15.13 load test fixes

SMTP

  • A message with a line longer than the SMTP line limit (2,000 characters) is now refused with a permanent error, 500 5.5.2 “Line too long”. Earlier versions answered with a temporary error, 451 4.3.0, so the sending server retried the message until its queue lifetime ran out. The event is logged as a warning, not an error.

Outbound queue

  • The bongo_queue_oldest_pending_seconds metric and the “oldest pending” figure on the admin queue page now show the age of the oldest queued message. Earlier versions showed the age of the message due to be retried next, which during a backlog could read a few seconds while the oldest message had waited 30 minutes.

Deployment

  • The compose template (deploy/docker-compose.yml) now limits the container to 2 GB of memory instead of 512 MB. In load tests the server used about 1 GB with 820 connected clients. Existing compose files are not changed by an upgrade.
  • docs/install.md has a new Sizing section with CPU, memory and disk figures from the load test.

Load test

  • Results of the first load test run are in docs/design/load-test-results.md. The numbers are preliminary: the test machines had shared CPUs.

Upgrade notes

  • No configuration changes.

v0.15.13 on GitHub

v0.15.12 inbound delivery in arrival order

Delivery

  • The inbound spool now delivers the message that has waited longest first. Earlier versions picked the lowest message ID. IDs are random, so during a delivery backlog a message could wait until the backlog cleared, which in load tests took several hours.
  • With no backlog, delivery is unchanged.

Upgrade notes

  • No configuration changes.
  • Messages already in the spool are delivered in the order they were queued.

v0.15.12 on GitHub

v0.15.11 clean shutdown with web sessions open

Shutdown

  • The server now ends live web UI update streams as soon as shutdown starts. Earlier versions waited 10 seconds for each open browser tab’s stream and then exited with status 1.
  • Under a container runtime’s default 10-second stop timeout, that wait led to a forced kill before the server closed Raft and storage. The derived-indexes.clean file was not written, and the next start rebuilt every index.
  • A single-node server no longer attempts a Raft leadership transfer on shutdown. Earlier versions logged “raft: leadership transfer failed” at WARN on every stop.
  • An error from bongo serve after startup no longer prints the command’s usage text. A mistyped flag still prints it.

Upgrade notes

  • No configuration changes.
  • The server can still wait up to 30 seconds for a message in delivery to finish before it stops. Set the container stop timeout to at least 45 seconds (Docker Compose: stop_grace_period: 45s).

v0.15.11 on GitHub

v0.15.10 Go security fix, mail delivery under load, faster restarts

Security

  • Bongo is built with Go 1.27.4, which fixes GO-2026-6859 (CVE-2026-97054). In earlier builds an HTTP request with a body over the size limit could make the server panic.

Mail delivery

  • A message accepted over SMTP is no longer set aside in spool/error/ when a storage write is refused under heavy load. The write is retried for up to two minutes. Earlier versions kept such a message in spool/error/ and did not deliver it.
  • Blob garbage collection no longer stops a pass when a queued message moves while the pass reads the spool. Under steady mail traffic earlier versions skipped most passes, and disk use only grew.

Restarts

  • A restart no longer restores the Raft snapshot and rebuilds every mailbox index before it accepts connections. On a test store with 75,000 messages that took 28 minutes, and the time grew with the amount of stored mail.
  • After a clean shutdown the server writes derived-indexes.clean in the data directory. The indexes are rebuilt at start only when that file is absent or out of date, which follows a crash or a forced stop.
  • The search index is now included in that rebuild. Earlier versions rebuilt the mailbox index only.

Upgrade notes

  • The first start after upgrading rebuilds all indexes once, because earlier versions do not write derived-indexes.clean. The start takes longer than usual; the time grows with the amount of stored mail.
  • No configuration changes.

v0.15.10 on GitHub

v0.15.9 shortcut help close button

Keyboard shortcuts

  • The Close button in the keyboard shortcut help (?) has the same style as the other secondary buttons. In v0.15.8 it showed as an unstyled square button.

Upgrade notes

  • No configuration changes.

v0.15.9 on GitHub

v0.15.8 keyboard shortcut help

Keyboard shortcuts

  • The keyboard shortcut help (?) opens as a centred dialog over the page again, with every shortcut listed. Since v0.15.4 it appeared without its layout.
  • On a small screen the shortcut list scrolls inside the dialog.

Upgrade notes

  • No configuration changes.

v0.15.8 on GitHub

v0.15.7 All Mail, Unread and Starred fixes, tag view actions, macOS shortcuts

All Mail, Unread and Starred

  • All Mail, Unread and Starred no longer list messages in Junk or Trash, or in any folder below them. Before, All Mail left out only a folder named exactly Trash, and Unread and Starred left out nothing.
  • All Mail still lists Sent and Drafts.
  • Actions in these views act on the message in the row. Before, a message was found by its number alone, and numbers repeat between folders. Deleting or starring a Sent message from All Mail could change the INBOX message with the same number.
  • Bulk mark read, mark unread, star and unstar work in these views. They failed before.
  • A link that names one of these views and a message number, such as an old bookmark to /mail/starred/5, now shows Not Found.

Tag views

  • Star, delete, the bulk actions and opening a message work in a tag view. They failed before.

Keyboard shortcuts

  • ? (shortcut help) and # (move to Trash) work on macOS keyboards.
  • The star in a message row can be toggled again after it changes, by click or by keyboard.

Documentation

  • The load test PRD records the results of the connection test (stage 7.4).

Upgrade notes

  • No configuration changes.

v0.15.7 on GitHub

v0.15.6 notification prompt closes correctly

Web interface

  • The desktop notification prompt now closes after enabling notifications or pressing its close button. Its flex layout had overridden the hidden attribute even though the prompt state changed correctly.

Documentation

  • Record findings from load-test stages 7.1 and 7.3.

Upgrade notes

  • No configuration changes are needed.

v0.15.6 on GitHub

v0.15.5 free the disk space of deleted mail

Mail pipeline

  • The spool no longer keeps an envelope for every delivered message. Before, each message that finished the pipeline left an envelope file in spool/deliver/, and nothing removed it.
  • Those envelopes counted as references to the message body. The blob garbage collector therefore never deleted the body of a message received over SMTP, even after the message was deleted and Trash emptied.
  • On the first start, Bongo deletes the envelopes left in spool/deliver/ and logs “spool: removed delivered envelopes” with the count.
  • The blob garbage collector then frees the bodies of deleted mail. A body is deleted when it has been unreferenced on two passes and is older than one hour. Passes run every 15 minutes.

Documentation

  • The load test PRD records the search index size per message measured in the first load test run, and a plan to reduce it.

Upgrade notes

  • No configuration changes.
  • spool/deliver/ is still created, so a rollback to v0.15.4 works. v0.15.4 writes envelopes there again until the next upgrade.

v0.15.5 on GitHub

v0.15.4 unified web navigation, live mailbox updates, load test tooling
  • Every signed-in page has the same application switcher: Mail, Calendar, Contacts, Tasks, Settings and Admin, in the same order. It stays in view while the sidebar below it scrolls.
  • The mail sidebar is grouped into primary mailboxes, More, Views, Folders and Tags. More and Views collapse, and the sidebar keeps its scroll position and collapsed state between pages.
  • Calendar, Contacts, Tasks, Settings, Security, Rules, Tags and Admin use the shared sidebar. “Back to Mail” is no longer the only way out of a page.
  • On a phone the navigation drawer reports its open state to screen readers.

Live mailbox updates

  • New mail appears in the open message list without a refresh.
  • The update stream follows the folder on screen. It was fixed to INBOX.
  • Messages read, moved or deleted in another tab or client update the open list. Sidebar unread counts update as mail arrives and as messages are read.
  • After a reconnect the list and counts are fetched again.

Keyboard triage

  • j and k move the selection. Enter opens the selected message.
  • e archives, # or Delete moves to Trash, u toggles read and unread. After a message leaves the list the next one is selected.
  • ? lists every shortcut. Shortcuts do nothing while typing in a field or the composer.

Web interface fixes

  • A failed save, move or delete shows an error. Before, some failures were silent.
  • The desktop notification prompt no longer blocks the page. It can be dismissed and does not cover controls on a phone.
  • The desktop notification icon loads.
  • The calendar event editor has a page header and its actions stay visible on a phone.
  • Week and Day views open scrolled to an hour before the current time when they show today, and to 07:00 otherwise.
  • Unpublishing a calendar, issuing a new feed link and removing a feed password ask for confirmation.
  • Settings tabs work with the arrow keys and show when more tabs are off screen.
  • Admin tables scroll sideways on a phone instead of cutting off columns.
  • Focus returns to the control that opened a dialog or menu when it closes.
  • Animation is reduced when the system asks for reduced motion.

Server metrics

  • New Prometheus metrics: open per-user databases and evictions, search index backlog and open indexes, Raft apply time and commit lag, FSM apply time, Pebble compactions, write stalls and WAL sync time, blob count and bytes, blob garbage collection passes, snapshot time, and per-protocol connections and command time.
  • Process CPU, memory and open files, and Go runtime metrics, are exported.

Configuration

  • storage.max_open_user_dbs sets how many per-user SQLite databases stay open (default 256, 1 to 16384). Read at startup.
  • telemetry.pprof serves CPU and heap profiles on the loopback control listener only. Off by default. Read at startup.

Load testing

  • scripts/loadgen generates SMTP, IMAP, POP3, CalDAV and CardDAV traffic, bulk-loads a fresh data directory, samples server metrics and writes a report. It refuses production domains.
  • docs/load-testing.md is the runbook. deploy/loadtest.example.hcl is the test server configuration.
  • The first test run is in progress. Results will be published separately.

Upgrade notes

  • No configuration changes are needed. Both new settings are optional.

v0.15.4 on GitHub

v0.15.3 app passwords

App passwords

  • An app password is a per-device credential for IMAP, POP3, SMTP submission, CalDAV and CardDAV. It never signs in to the web interface.
  • Each password has 24 letters. Bongo shows it once and stores only a hash.
  • Settings, Security lists each password with its label, creation date and last used date. A user can hold up to 10.
  • Creating a password needs the account password, and a code with 2FA on.
  • A successful app password login forgives no failed guesses at the account password.
  • A user name with the shape of an app password is not logged.
  • Revoking a password refuses its next login and ends the IMAP, POP3 and SMTP sessions it opened on their next command.
  • Change password has a box, ticked by default, that also revokes all app passwords.
  • An admin can revoke all of a user’s app passwords.

Add a device

  • Add a device shows a new app password beside the server settings. These match the autoconfig documents.
  • Apple devices can download a personal profile with Mail, Calendar, Contacts and Reminders set up. The profile is unsigned, so the device shows Not Verified.
  • A QR code holds a one-time link for a phone. The link expires after 5 minutes and works once.

Require app passwords

  • An account with 2FA and at least one app password can turn on Require app passwords. IMAP, POP3, SMTP submission, CalDAV and CardDAV then refuse the account password.
  • Turning it on ends sessions opened with the account password.
  • The switch is offered at the end of 2FA enrolment and on the Security page.

Documentation

  • docs/caldav-carddav.md, docs/sendmail.md, docs/walkthrough.md and SECURITY.md cover app passwords and the switch.
  • docs/design/from-nmap-to-one-binary.md records the move from NMAP agents to one binary.

Upgrade notes

  • No configuration changes.
  • Nothing changes for an account until its owner creates an app password or turns on the switch.
  • Upgrade every node together. An earlier binary stores the app password touch and link commands as literal keys, ignores the switch and refuses app passwords.

Known limits

  • The Apple profile and QR link have not been checked on real devices.

Full notes: prds/done/prd-app-passwords-v0.15.3.md and prds/verification.md.

v0.15.3 on GitHub

v0.15.2 web UI fixes

Calendar

  • The calendar settings page uses the shared page, card, field and button styles, in light, dark and mobile.
  • The calendar sidebar shows coloured dots and themed checkboxes. Import is one button. The month title in the mini calendar is readable in the light theme.
  • The week and day views show the weekday and date above each column, with today highlighted. A header links to that day.
  • The event form follows the theme. Its controls were white in dark mode.
  • The calendar invitations page uses the shared page and card styles.
  • The result of an import is shown after the import.

Tasks and Apple Reminders

  • An account with no calendar that holds tasks gets the Tasks calendar the next time it opens a calendar page or a CalDAV client lists its calendars. This calendar sits outside the limit of 5 calendars per user.
  • The placeholder name DEFAULT_TASK_CALENDAR_NAME from Apple Reminders is stored as “Tasks”. A list already created with that name keeps it.

Admin

  • The product name is readable on the admin header. The admin footer and the link back to mail are styled.

Development

  • make seed creates tasks.
  • make ui-shots also covers calendar invitations, the event form and tasks.

Upgrade notes

  • No configuration changes.
  • An account with 4 event calendars and no Tasks calendar has one calendar fewer available after the upgrade, because Tasks is added.

Full notes: prds/done/prd-ui-fixes-v0.15.2.md and prds/verification.md.

v0.15.2 on GitHub

v0.15.1 calendar fixes

CalDAV and CardDAV

  • CalDAV clients can create calendars and lists (MKCALENDAR and extended MKCOL). A list created in Apple Reminders or Thunderbird holds tasks.
  • A calendar can be deleted in the web or from a CalDAV client, with its events, tasks, shares and published feed. Tasks, the default calendar and the last event calendar cannot be deleted.
  • A missing CalDAV object answers 404, not 500.
  • CalDAV and CardDAV ETags are quoted once, so updates with If-Match are accepted. Events created in the web have a content ETag over CalDAV. Clients fetch every contact, and every event created in the web, once after the upgrade.
  • Windows time-zone names, as Outlook sends them, resolve through the CLDR table. Such events were stored at 1970 before.
  • Events are filed only in calendars that exist.

Calendars

  • A calendar can be created in the web, with a name, a colour and optionally tasks. A user owns at most 5 calendars, Personal and Tasks included; the limit applies in the web, over CalDAV and to subscriptions.
  • “This and following” no longer leaves an empty series behind when the first occurrences were deleted.

Tasks

  • Tasks in calendars shared with the user appear, read-only, in /tasks and the agenda.
  • Recurring tasks appear at every occurrence. Overdue open tasks appear on today’s row.
  • A due date before the start is refused.

Invitations and scheduling

  • Updates carry a higher SEQUENCE. Title and description changes are sent to attendees.
  • No invitations or cancellations are sent for events entirely in the past. An all-day event is past from midnight in the user’s time zone.
  • CalDAV clients schedule single occurrences and attendees invited to one occurrence only.
  • Deleting an invitation in a CalDAV client declines it.
  • A reply covering several occurrences is applied to all of them.
  • The invitation card shows the user’s own answer. A user invited only at an alias they cannot send from is told the organiser will not be told.
  • Rooms and equipment answer bookings for single occurrences. Booking mail is no longer kept in the resource’s inbox.

Sharing and reminders

  • Pending calendar shares expire after 30 days. A user can leave a calendar shared with them.
  • Browser reminders fire once.
  • Dates in reminder mail and the agenda follow the user’s language.

Mail and web

  • The To me, From my contacts and Mailing lists views moved to /mail/view/…, so folders with those names are reachable.
  • “Unread mail to me” on the summary page lists unread mail only.
  • Display names containing a comma, including encoded non-ASCII names, are no longer split.
  • Status messages show only fixed, translated texts; a link can no longer put its own text in a message.
  • The recovery-code warning is shown on the inbox.
  • The Apple configuration profile has per-user identifiers, so two users of one domain can install theirs on one device.

Upgrade notes

  • No configuration changes.
  • Links to /mail/to-me, /mail/from-contacts and /mail/lists change to /mail/view/…; the old paths are not redirected.
  • Status links carrying free text in ?flash= no longer show it.

Full notes: prds/active/prd-calendar-fixes-v0.15.1.md and prds/verification.md. docs/caldav-carddav.md covers creating and deleting calendars.

v0.15.1 on GitHub

v0.14.0mail delivery and rules

Postmaster and submission

  • postmaster@ and abuse@ resolve on every hosted domain: to the domain owner, else to smtp.postmaster_account, else to the catch-all. Without a target, RCPT gets 550 5.1.1 and startup lists the domains concerned.
  • queue.bounce_copy_postmaster (default off) sends a headers-only copy of each bounce to postmaster@ the sender’s domain.
  • Implicit-TLS submission on port 465 is on when smtp.tls is set. Autoconfig, Autodiscover, the Apple profile and the admin DNS guide list 465 first, only when the SMTP listeners start.
  • An RCPT to an unknown local part on a hosted domain, alias domains included, is refused with 550 5.1.1. An authenticated session used to relay it and the sender received a bounce later.

Vacation and rules

  • Vacation replies at /settings/vacation, with RFC 3834 loop protection, one reply per sender per interval (1 to 30 days, default 7) and at most 100 replies per user per hour.
  • A rule can have up to 10 actions. flag, mark read and copy now work as named; copy no longer moves. Spam stays in Junk.
  • New rules continue to the next rule after a match. Existing rules are migrated to stop after a match, so they behave as before. The migration also runs on a node that becomes leader later.
  • Rules can match any condition instead of all. New conditions: MIME type of any part, and any header by name.

Queue, DSN and storage

  • queue.max_lifetime (default 5 days, was about 17 hours), queue.max_retry_interval (1 hour) and queue.concurrency (5).
  • SMTP DSN parameters RET, ENVID, NOTIFY and ORCPT are accepted and honoured for local senders, including delay reports (queue.delay_warning_after, default 4 hours). Reports to external senders are not sent yet.
  • Below storage.min_free_mb (default 1024) the server refuses new mail, IMAP APPEND, uploads and DAV writes with a temporary error, and the admin page shows a banner.

Domains, web mail and CLI

  • Domain aliasing: mail to user@alias.example is delivered to user@example.com. A domain with alias domains cannot be deleted.
  • Archive and Unarchive in the message view, the bulk toolbar, the e key and a left swipe on a phone, including from Starred and other virtual views. A left swipe used to move the message to Trash. IMAP clients see \Archive on the Archive mailbox.
  • Reply to List. Replies follow Reply-To. Your own addresses are removed from To and Cc; Reply on your own sent message addresses its original recipients.
  • bongo sendmail, a sendmail-compatible submission client for cron and scripts (docs/sendmail.md).
  • bongo user list, bongo domain list and bongo alias list, as a table or JSON.

Upgrade notes

  • Upgrade every node before setting any new key (smtp.postmaster_account, smtp.submissions_port, queue.*, storage.min_free_mb) or creating an alias domain, a multi-action rule or a match-any rule. An older node refuses the new keys and ignores the new rule fields.
  • Publish port 465 before deploying, or set smtp.submissions_port = 0.
  • Check the 1 GiB storage.min_free_mb default against the free space on the data volume.
  • A configured smtp.tls certificate that fails to load now stops startup.
  • Browsers keep the old bongo.js until the asset version changes; reload open tabs after upgrading.

Full notes: prds/done/prd-mail-delivery-v0.14.md and prds/verification.md.

Patch releases

v0.14.1 postmaster startup report

Fixes

  • The startup warning about postmaster@ and abuse@ now names a domain only when one of those addresses does not resolve. Explicit addresses, legacy aliases and catch-alls count. In v0.14.0 it checked only the domain owner and smtp.postmaster_account, so it named domains whose postmaster@ was delivered by an alias and said their mail was refused.
  • The warning now reads “domains where postmaster@ or abuse@ does not resolve”. Log filters that matched the v0.14.0 text need updating.

Upgrade notes

  • No configuration change. Delivery is unchanged; only the warning differs.

Full notes: prds/done/prd-mail-delivery-v0.14.md.

v0.14.1 on GitHub

v0.13.0web UI rebuild

Web UI

  • Every screen is rebuilt to the Bongo Mail design: new shell with top bar, account menu and grouped sidebar; design tokens, Geist fonts and one icon set; light and dark mode, including the login and other public pages.
  • Mailbox list shows previews, attachment markers, header counts and a bulk toolbar. Message and thread view has a quick reply. Compose uses recipient pills and a drop zone for attachments.
  • Calendar has an agenda view and shows times in the user’s time zone throughout. Contacts, rules, settings, security, tags, Junk and admin (KPI cards, tables, status pills) follow the same layout.
  • No emoji remain in the templates.

Fixes from operator testing

  • Tagging works from the message view, the bulk toolbar and the context menu. Typed tag names are folded into stored form (“My Tag” becomes “my-tag”); a rejected tag shows an error instead of closing the picker.
  • The rule editor, the contact editor and the calendar show/hide checkbox work under the Alpine CSP build.
  • Sidebar unread counts update live when messages are read or arrive.
  • “Create filter” in a message’s More and context menus opens a new rule matching the sender.
  • Tag colour swatches are announced by colour name.

Correctness

  • SQLite busy_timeout and WAL mode apply to every pooled connection, not only the first; a contended index write now waits instead of failing with SQLITE_BUSY.
  • An agent cancelled during registration shuts down cleanly.

Operations

  • golang.org/x/net updated to v0.62.0 for two html advisories.
  • Templates are generated with templ v0.3.1001.
  • Assets are cache-busted by commit hash; reload open tabs after upgrading.

Full notes: prds/active/prd-ui-rebuild-v0.13.md.

Patch releases

v0.13.4 reminder checker crash fix

Stability

  • The server no longer panics in the calendar reminder checker. The checker closed a storage iterator twice. Storage reuses closed iterators, so the second close could break another scan in progress and stop the server with “index out of range” in mergingIter.Close. The fault was present since v0.1.0 and occurred after several hours of running. A repeated close of a storage iterator is now a no-op.

v0.13.4 on GitHub

v0.13.3 the app is named bongo

Naming

  • The server binary and CLI are named bongo. The release zips are named bongo_<version>_linux_amd64.zip and bongo_<version>_linux_arm64.zip.
  • The container image runs /usr/local/bin/bongo.
  • The systemd unit is bongo.service.
  • The default data and log directories are /var/lib/bongo and /var/log/bongo.
  • Protocol greetings, the Received-SPF header, log messages, the self-signed certificate and the calendar PRODID use the name Bongo.

v0.13.3 on GitHub

v0.13.2 domain administration fixes, Linux binaries, documentation

Security

  • Only global administrators can delete a domain. A domain administrator could previously delete a domain they administer with a direct request, although the web UI did not offer it.

Mail

  • A disabled domain refuses all inbound mail. SMTP answers 550 5.1.1 for every address on it, including aliases and the catch-all. Accounts and stored mail are kept; re-enabling the domain restores delivery. Previously disabling a domain only removed it from client autoconfiguration.

Releases

  • Each version tag now publishes a GitHub release with Linux binaries for amd64 and arm64, a SHA256SUMS file, and these notes. The zips include LICENSE and THIRD_PARTY_NOTICES.md.

Documentation

  • The walkthrough and development guide describe the rebuilt web UI.
  • Deployment, installation, configuration, import, clustering and multi-domain documentation was checked against the code and corrected.
  • The edge and agent server roles are documented as incomplete. Use the server role.
  • Dev mode serves the web UI over plain HTTP. make seed, make validate and make ui-test-setup now work against a dev server.

Full notes: prds/active/prd-ui-rebuild-v0.13.md (section 8.2).

v0.13.2 on GitHub

v0.13.1 web UI fixes from operator testing

Mail view

  • Two-pane shows the message list above the reading pane; three-pane shows them side by side. The two layouts previously looked the same.
  • In both pane layouts a message opens in the reading pane instead of replacing the whole view. A click anywhere on a message row opens it.
  • In the pane layouts the list and the reading pane scroll separately.

Keyboard

  • In the pane layouts j and k open the highlighted message after a 300 ms pause. Holding the key moves through the list without marking every message read.
  • d moves the highlighted message to Trash. In Trash, d does nothing.
  • d, r and drag-and-drop act on the mailbox on screen after sidebar navigation. Previously they could act on a message with the same UID in the previously loaded mailbox.

Fixes

  • Deleting, moving or marking spam in a subfolder such as INBOX/Important returns to that subfolder instead of showing “Invalid UID”.
  • Pages load in dark mode without a light flash when the OS or the saved preference is dark.

Branding

  • The top bar and login page show the swoosh from the original Bongo project logo.

Assets are cache-busted by commit hash; reload open tabs after upgrading.

Full notes: prds/active/prd-ui-rebuild-v0.13.md (US-024 to US-029).

v0.13.1 on GitHub

v0.12.0hardening release

Security

  • /internal/* control routes move to a loopback control listener, server.control_addr (default 127.0.0.1:8081); the web port answers 404. The bongo-go CLI defaults –http-addr to it.
  • Bounces of forwarded mail are relayed only for SPF-pass or authenticated senders.
  • SMTP DATA and BDAT refuse a body more than a tenth plus 1 KiB over the declared SIZE (552 5.3.4).
  • CalDAV and CardDAV share a 64 MiB budget for request bodies in flight.
  • Send-as verification no longer reveals whether an address is taken; at most 20 pending claims per user; claims are removed with the address or domain; concurrent correct codes yield one owner.
  • IMAP, SMTP and POP3 fail closed when the account state cannot be read.
  • A redirect-listener bind failure stops startup; a web Serve error after ready shuts the node down.
  • Uploads spool to disk above 8 MiB; the mailbox import streams messages and checks quota first.

Correctness

  • SMTP writes one envelope per local recipient, so one recipient’s rules no longer file or drop another’s copy.
  • A retried web send no longer duplicates (send ID; 409 on changed content). Compose tabs opened before the upgrade must be reloaded.
  • Calendar export keeps its time zones; all-day reminders use the owner’s zone; the month grid shows neighbouring months’ all-day events.
  • Compose messages are translated (de, es, fr).

Operations

  • Config validation refuses colliding or out-of-range listener ports and a server with a wildcard bind and no advertise or cluster bind address, at startup and on hot reload.
  • Built with Go 1.27.3; raft 1.8.0 and memberlist 0.7.0 (protocol versions unchanged); chi updated past three advisories.
  • The repository is licensed AGPL-3.0; the web UI links to the source (web.source_url).

Full notes: prds/active/prd-hardening-v0.12.md, section 8.2.

v0.11.0

Phase B of the hardening PRD: agent role and cluster transport security.

  • Internal requests sign with the effective session secret; agent-role SMTP verifies SRS recipients; agent antispam no longer skips DMARC on unverifiable trust flags.
  • New cluster.bind_addr; Raft and gossip never bind server.bind_addr.
  • Raft with peers requires cert, key and CA, verifies peers, no plaintext.
  • Agent gRPC servers and clients use mutual TLS and bind the cluster address.
  • Blob sync and cluster join move to a mutual TLS peer listener (cluster.peer_port, default 8302) on nodes with peers.

Upgrade notes: a server with a non-loopback advertise_addr needs cluster cert, key and ca_file; server.bind_addr 0.0.0.0 without advertise_addr no longer starts; role agent needs the agent_server.tls files; nodes with peers must be upgraded together. Production (role server, single node, loopback advertise) is unaffected.

v0.10.0

No release notes were published with this tag.

v0.9.0

No release notes were published with this tag.

v0.8.0

No release notes were published with this tag.

Patch releases

v0.8.2

No release notes were published with this tag.

v0.8.2 on GitHub

v0.8.1

No release notes were published with this tag.

v0.8.1 on GitHub

v0.7.0

No release notes were published with this tag.

Patch releases

v0.7.1

No release notes were published with this tag.

v0.7.1 on GitHub

v0.6.0

No release notes were published with this tag.

Patch releases

v0.6.31

No release notes were published with this tag.

v0.6.31 on GitHub

v0.6.30

IMAP subscriptions, correct sequence numbering, and a working cluster status.

feat(imap): implement folder subscriptions

Subscribe and Unsubscribe were no-ops and List read only ReturnStatus, so LSUB returned every mailbox rather than the subscribed subset, \Subscribed was never emitted, and unsubscribing a folder did not stick.

State lives on MailboxMeta as “unsubscribed”, inverted deliberately. Every mailbox already on disk has no such field and decodes as the zero value; with “subscribed” that value is false, which would unsubscribe every folder for every user on upgrade. Inverted, it means still-subscribed, so this release changes nothing for existing mailboxes until someone explicitly unsubscribes.

SUBSCRIBE applies through a synthetic Raft command rather than writing back a whole MailboxMeta, which would race a delivery and reset uid_next, handing the next message a duplicate UID.

fix(imap): number EXPUNGE and SEARCH from the client’s view

Sequence numbers were computed from the server’s message list. They are only meaningful relative to what the client has been told, and the two diverge as soon as another connection removes a message between the client’s last update and the command – at which point EXPUNGE tells the client to drop the wrong message.

fix(cmd): authenticate bongo-go cluster status

It called http.Get unsigned against an endpoint behind RequireInternalAuth, so it returned 401 on every node with web.session_secret set. It worked only on a dev node with no secret, which is where nobody needs it.

v0.6.30 on GitHub

v0.6.29

Outbound mail fails fast when its body is gone.

fix(queue): bounce immediately when the message body is gone

handleFailure treats every error as worth retrying and only bounces once maxAttempts is spent. A message enqueued before the 2026-08-30 blob-GC incident therefore spent seven days cycling through attempts that each failed identically: its body blob no longer exists, so the ninth attempt fails exactly as the first did.

handlePermanentFailure bounces without consuming the remaining retries. It fires only on store.ErrNotFound, because the distinction that matters is “the body is gone” against “the blob store is unwell” – an I/O error, a permissions problem, or a store still starting are transient and keep their backoff. Both read paths are covered, including RawMessageBytes for split messages where a missing attachment is equally unrecoverable.

Nine wasted attempts is the smaller cost. The larger one is that the DSN the sender received blamed delivery, when the body is gone and resending is the only remedy. The bounce now says so.

v0.6.29 on GitHub

v0.6.28

IMAP clients are told about new mail outside IDLE.

fix(imap): tell clients about new mail outside IDLE

session.Poll returned nil with the comment “no-op, we use IDLE”. go-imap runs Conn.poll after every command in the authenticated or selected state, and that is how a server delivers untagged EXISTS and EXPUNGE to a client polling with NOOP. Returning nil meant the only way any client learned about new messages was IDLE, and only while idling on that exact mailbox.

Observed: a message appended to a user’s Sent folder stored correctly, but Thunderbird did not display it until the client was restarted, which forced a fresh SELECT. Thunderbird idles on INBOX and polls other folders with NOOP.

The fix needs two baselines. cachedMsgs is the sequence-number working set that handlers refresh freely – Fetch, Store, Search, Copy and Move all overwrite it at the top of the handler, and the poll runs after the handler returns, so diffing against it destroys the pending EXISTS rather than deferring it. reportedUIDs records what the client has actually been sent.

The baseline moves wherever the client is genuinely told something: Select, Poll, Idle, Expunge and Move. Expunge and Move write untagged EXPUNGEs themselves, and omitting them would leave the next poll re-sending a sequence number the client has already renumbered against, dropping a message it should have kept.

Expunges are written highest sequence number first, since each renumbers those after it, and are withheld when allowExpunge is false because FETCH, STORE and SEARCH must not emit them mid-command.

Known and unchanged: Subscribe and Unsubscribe are still no-ops, and List ignores SelectSubscribed, so LSUB returns every mailbox rather than the subscribed subset. Every folder is effectively always subscribed.

v0.6.28 on GitHub

v0.6.27

Raft snapshots actually happen, and the log finally truncates.

fix(cluster): apply every raft tuning value, and stop the log growing forever

The production node had not snapshotted since 2026-08-23. Every restart restored a two-week-old snapshot, rebuilt derived indexes for the state as of that date, and warned for each message whose blob had since been collected – warnings that accumulate permanently, one per delete-then-GC cycle. It also feeds the accepted first-delivery stall (B12), which scales with log size.

Three causes, only one of them tuning.

snapshot_interval was parsed, defaulted, documented and set in the production config, and never assigned to raft.Config. A settable key with no destination is invisible: nothing downstream changes, so no behavioural test can see it.

TrailingLogs was not configurable, and raft’s own default of 10240 exceeds this project’s snapshot_threshold default of 8192. Neither number is wrong alone, but together the threshold is inert – a snapshot fires and then retains more entries than triggered it, so the log never truncates. It is now configurable, and an inverted pairing is refused at startup with the reason in the error. A node on the defaults is warned rather than refused, because changing those defaults would alter behaviour for existing deployments.

The threshold was also far too high for a node delivering roughly five messages a day. Code defaults are unchanged; only the production config is tuned, to 256 with 128 trailing.

Adds “bongo-go cluster snapshot” to force one on the leader, since a stale snapshot cannot otherwise be cleared without waiting for the threshold.

Known issue, not fixed here: “bongo-go cluster status” cannot authenticate against a node with web.session_secret set. It calls http.Get unsigned against an endpoint behind RequireInternalAuth and returns 401.

v0.6.27 on GitHub

v0.6.26

Tooling to see and act on mail whose body blob was lost.

feat(admin): report and purge mail whose body blob is gone

Eight messages carry MailMeta records whose raw_message blob no longer exists, left by the 2026-08-30 blob-GC incident. They appear in no per-user SQLite index, which proves nothing: that row is written during indexing, and indexing is what fails on a missing body.

Adds FindDanglingMail and DeleteDanglingMail, the /internal/mail/dangling and /internal/mail/purge-dangling endpoints behind internal auth, and the “bongo-go mail dangling” / “mail purge-dangling –confirm” commands, mirroring the existing orphaned-address repair.

The report lists every MailboxMessage entry referencing each GUID, which is what distinguishes orphaned metadata from a live message sitting in a mailbox. Deletion removes those entries too, and before the MailMeta: the sequence is several Raft commands and is not atomic, and this order leaves orphaned MailMeta on an interruption – which sweepOrphanedMail already reclaims – rather than a mailbox entry pointing at nothing, which nothing cleans up.

Each blob is re-checked immediately before its record is removed, since blob sync can pull one back from a peer between the scan and the delete.

Also corrects a false claim in v0.6.25’s 14.5 work. It said a MailboxMessage decoded as a MailMeta fails proto.Unmarshal on a wire-type clash, and that a user could therefore disable blob GC by naming a mailbox “mail”. The decode does not fail; protobuf-go skips a wire-type mismatch on a known field number as an unknown field. That denial-of-service never existed. The structural key parse remains correct for the plainer reason that unmarshal success is not evidence a value is the type you expected.

This release ships the report. No purge has been run.

v0.6.26 on GitHub

v0.6.25

Blob GC reference integrity: the fixes for the 2026-08-30 mail-body loss.

blob-gc is the only component that deletes blobs, and blobs are content-addressed and shared, so one missed reference loses that content for every holder with no undo. On 2026-08-30 a sweep deleted 17 blobs, two of which were still referenced. An audit found eight gaps. This release fixes all of them.

blob_gc must stay “dryrun” until the candidate lists produced by this build have been reviewed on real data. This release makes that review possible; it does not authorise re-enabling deletion.

fix(agent): re-validate blob GC candidates immediately before deleting

The minAge grace assumed a blob’s mtime tracks its most recent write, but BlobStore.Put early-returns without touching the file when the hash already exists. A deduplicated write therefore keeps the ORIGINAL mtime, so a blob that becomes referenced again during the reference scan is simultaneously absent from the snapshot, older than the cutoff, and a two-pass candidate. Both existing guards miss it, and the stamped-message redelivery path makes it deterministic. Candidates are now re-checked against a fresh reference set immediately before deletion, inside deleteSweptBlobs so nothing can reach a delete without it.

fix(agent): count queued outbound attachments as referenced blobs

Attachment-dedup splits an outbound message into a skeleton plus separate attachment blobs, which the queue worker reassembles before relaying. Only the skeleton was referenced, so the GC reclaimed the attachments and the message relayed with them missing. The queue entry stayed healthy throughout, so this damaged outbound mail without leaving any evidence.

fix(agent): parse mail keys structurally, then fail loudly on bad records

The mail scan matched a “/mail/” substring, which a user could trip by creating an IMAP mailbox named “mail”. proto.Unmarshal failures were then silently skipped, dropping references. Fixed in that order deliberately: making unmarshal fatal first would have let any user disable the GC node-wide.

fix(agent): do not carry blob GC candidates across a failed pass

Both error paths returned the previous candidate set, so it survived an arbitrary run of failed passes and “two consecutive passes” stopped meaning two consecutive good ones.

fix(agent): scan the spool twice so a released message keeps its body

An admin releasing a message from quarantine moves it backward through the directory order the scan reads, so a release landing mid-scan appeared in no listing at all and its body looked unreferenced.

feat(agent): log every blob a sweep reclaims, with size and mtime

The incident left “candidates:17 deleted:17” and nothing else, so the hashes were unrecoverable and attribution meant decoding spool envelopes by hand. Both modes now log per-blob detail at Info: a deleting pass is the one whose record cannot be reconstructed afterwards.

test(agent): fail when a blob-bearing proto field escapes the reference set

Twice a field that pins a blob was missing from the scan. A descriptor walk now requires every blob-bearing proto field to be explicitly classified, so a third recurrence fails the build instead of production.

v0.6.25 on GitHub

v0.6.24

Two data races in the Raft FSM, and the per-user data leak behind the ghost-mailbox incident.

fix(cluster): synchronise FSM dependency injection against the apply loop

Agent.Run injects the search indexer, blob syncer, notifier and derived-index queue into the FSM after handing it to Raft, so the apply goroutine read those fields while Run was still writing them. This fires on any restart that applies Raft entries during startup - i.e. every restart with a non-empty log. StartIndexer’s check-then-assign was also a double-start window: two callers could each create a queue, orphaning the first worker. All four fields are now guarded and read through accessors.

fix(admin): reclaim per-user data left behind by account deletion

DeleteUser removed /user/{id}/* and the users row but neither the per-user metadata database nor the search index, so every deleted account leaked both. That leak is what let delivery keep succeeding into a deleted account’s mailbox. Adds the cascade, plus an orphan-GC sweep that reclaims /user/{id}/ trees with no users row and no profile - which also unpins the blobs those keys were holding.

chore(ci): stop running every workflow twice per pull request

v0.6.24 on GitHub

v0.6.23

fix(admin): delete a user’s address-index records when the user is deleted

Deleting a user left its /system/address/ records behind. Because RecipientResolver consults the address index ahead of aliases, the users table and the catch-all, a stale record outranked any alias later created for the same address and resolved mail to the deleted account - delivered into an unreachable mailbox, neither bounced nor visible.

Adds the DeleteUser cascade, plus ‘bongo-go address orphans’ and ‘bongo-go address repair’ to find and remove existing stale records.

v0.6.23 on GitHub

v0.6.22
  • Blob/orphan GC now announces its configured mode at startup, so a blob_gc change is verifiable from the logs instead of being silent.
  • Composite antispam processor merges the losing backend’s reasons rather than discarding them (B14).
  • deploy: blob_gc returned to “on”; grafana dashboard removed with the observability stack.
  • ci: golangci-lint pinned to v2.13.2.

v0.6.22 on GitHub

v0.6.21

Returns blob_gc to “on” now that collectReferencedBlobs scans all five spool directories, and is the first release published under the bongo-project/bongo name.

v0.6.21 on GitHub

v0.6.20 - data-loss fixes, DAV correctness, tenant isolation

Contains fixes for three ways live mail could be destroyed, one of which was observed in production on 2026-08-27. Deploying this is what allows storage.blob_gc to be returned to “on”; until then the running config must stay at “dryrun”, and note that dryrun does NOT protect the quarantine deletion paths.

Data loss

  • Blob GC never counted the on-disk spool. A blob referenced only by a spool envelope – quarantined mail awaiting review, or anything deferred past the one-hour grace – was reclaimed while the envelope survived. Observed symptom: an outbound forward failing with “blob not found”, because deliver.go reuses the spool envelope’s BlobHash when it enqueues a forward. Content in that state is unrecoverable.
  • Quarantine deletion destroyed other users’ mail. Blobs are content-addressed and shared, so quarantined spam has the same hash as any byte-identical copy delivered elsewhere – the normal case for a broadcast message. All three paths did it, including the unattended hourly auto-purge, which no configuration setting disables.
  • Queue retry, bounce and retry-now were two Raft commands each: delete the old key, put the new one. A crash or rejection between them left the message at neither key. Now one composite command applied in a single Pebble batch.

Security and correctness

  • The Global Address List could not be scoped: any authenticated user could enumerate every address of every tenant through compose autocomplete. Now restricted to the caller’s own domain.
  • IMAP EXAMINE was not read-only. STORE, EXPUNGE and MOVE all mutated a mailbox the client had been told was read-only; MOVE emptied it.
  • CalDAV/CardDAV ignored If-Match and If-None-Match entirely, so concurrent edits silently last-write-wins. ETags were also second-resolution, which had to be fixed first or the preconditions would have protected nothing.
  • Mixed permanent and transient relay outcomes re-sent to and re-bounced permanently-failed recipients on every attempt, up to ten duplicate DSNs per address.

Other

  • Received mail bodies are now indexed, so body search works for mail you received and not only mail you sent.
  • Tagging a message no longer deletes its indexed body.
  • Calendar events with an unparseable RRULE are shown as a single occurrence rather than vanishing from every view.

Tests

Around 60 tests added, each verified to fail against the defect it covers. Includes the DAV cross-account scoping suite, which had no coverage at all, and replaces two tests that asserted nothing.

v0.6.20 on GitHub

v0.6.19 - tester release with corrected dev-mode instructions

Identical server code to v0.6.17. Corrects the tester README’s first instruction, which told testers to open https://localhost:8080 when dev mode serves plain HTTP, and lists the dev-mode mail ports.

v0.6.19 on GitHub

v0.6.18 - tester release README

Identical code to v0.6.17; this tag exists so the shared release artifact corresponds to a real commit. Fills in the tester README’s contact channel (issues on the private repository) and its terms (private evaluation build, no open-source licence applying yet).

v0.6.18 on GitHub

v0.6.17 - delivery-path performance and observability

Derived search/metadata indexing no longer runs inside the Raft apply path, so opening a user’s index handles cannot block mail delivery.

Raft’s LastIndex no longer scans the entire transaction log to find its last entry. On a log the size of production’s this was ~20ms per call against 15.7us for a direct seek, and Raft calls it constantly.

Raft’s own logs are no longer discarded. The consensus layer previously ran completely silent, which made a delivery stall inside it impossible to explain from the outside.

The delivery stage now reports a per-step breakdown when it is slow, and the FSM reports slow applies and slow index work.

KNOWN ISSUE: the first delivery after a restart can still take ~70s. Normal delivery is sub-second (median 0.59s) and warm delivery is ~63ms, so this affects one message per restart. It is measured to be inside hashicorp/raft between the write call and our FSM being invoked; see task B12.

v0.6.17 on GitHub

v0.6.16 - derived indexing no longer blocks mail delivery

The first delivery after a restart took 42-85s, against 63ms once warm, because the Raft apply maintained per-user search and metadata indexes inline: the first write paid the cost of opening that user’s handles while the message waited behind it.

Mail and event indexing now runs on a background worker, so a message is stored and visible immediately and only search catches up fractionally later. Mailbox listings read from Pebble and are unaffected.

Profile indexing stays inline on purpose, because it writes the shared system.db that user creation also writes.

Also documents that the 5s passed to the Raft apply is not an upper bound on the call, sets a SQLite busy timeout, and logs slow index work so a recurrence reports itself.

v0.6.16 on GitHub

v0.6.15 — trusted submission, scanner wiring, quarantine release fix

Fixes mail between accounts on a hosted domain being filed as spam: an authenticated submission (or a connection from a configured trusted network) is no longer judged by SPF, DKIM alignment and DMARC, which cannot succeed from a container or LAN address. Content filtering is unchanged — scoring, Bayes and virus scanning still run for trusted mail.

Fixes a silent mail-loss bug: releasing a message from quarantine moved it to a spool directory nothing reads, so it was never delivered and never surfaced again. Release now re-queues it for delivery without re-running the check that quarantined it.

Wires up scanners that were implemented but never constructed. Enabling antivirus now actually scans; rspamd and SpamAssassin can be selected and are composed with the builtin scorer rather than replacing it. An unreachable antivirus scanner quarantines rather than delivering unscanned (previously it silently passed everything); an unreachable spam scanner degrades to the builtin verdict, because mail should never be held up by a spam scanner.

Adds a bulk rescan action so a quarantined backlog can be drained.

Operator note: quarantine is auto-purged after pipeline.quarantine_retention (30d default) and the purge deletes the message, so the antivirus recovery window is finite. Alert on the ERROR log if antivirus is enabled.

v0.6.15 on GitHub

v0.6.14 test, security and CI repair

Restores a green CI pipeline for the first time since 12 July. Three independent failures, each masking the next:

  • integration: a test scraped a /mail/blob/{hash} URL the app stopped emitting after the Phase 9 attachment-dedup work; one failure cascaded into a 10-minute suite timeout on CI runners.
  • govulncheck: 20 vulnerabilities. 18 stdlib, cleared by moving the go directive to 1.26.6; the rest by bumping grpc and x/text.
  • gosec: never ran until govulncheck passed. Of three pre-existing HIGH findings, the quota-store integer conversions were a real enforcement bypass - a negative usage made every storage-quota check pass.

No delivery-path changes since v0.6.13.

v0.6.14 on GitHub

v0.6.13 outbound deliverability

Fixes outbound SMTP announcing itself as EHLO localhost, which strict receivers (Postfix reject_non_fqdn_helo_hostname and most large providers) permanently rejected - breaking outbound mail from all local users.

Also makes the resulting bounces visible instead of silently junked: DSNs are DKIM-signed so they pass our own DMARC check, conform to RFC 3464/3834, and use the null envelope sender. Delivered messages now carry Authentication-Results and X-Spam-* headers.

v0.6.13 on GitHub

v0.6.12 log attachment dedup outcome (Loki-visible) + Grafana dedup panels

No release notes were published with this tag.

v0.6.12 on GitHub

v0.6.11 attachment dedup metrics (9.8); Phase 9 complete

No release notes were published with this tag.

v0.6.11 on GitHub

v0.6.10 outbound attachment dedup (9.6) with queue reconstruct

No release notes were published with this tag.

v0.6.10 on GitHub

v0.6.9 fix split-attachment download/preview (decode via message route)

No release notes were published with this tag.

v0.6.9 on GitHub

v0.6.8 attachment dedup (Phase 9 inbound) + opt-in blob GC

No release notes were published with this tag.

v0.6.8 on GitHub

v0.6.7 opt-in display for image attachments

No release notes were published with this tag.

v0.6.7 on GitHub

v0.6.6 fix inbound mail attachment downloads (message-scoped attachment route)

No release notes were published with this tag.

v0.6.6 on GitHub

v0.6.5 in-app admin help (8.1) — DNS status legend + DKIM re-publish note

No release notes were published with this tag.

v0.6.5 on GitHub

v0.6.4 backlog cleanup + hardening (B2-B7) + queue/antispam metrics

No release notes were published with this tag.

v0.6.4 on GitHub

v0.6.2

No release notes were published with this tag.

v0.6.2 on GitHub

v0.6.1

No release notes were published with this tag.

v0.6.1 on GitHub

v0.5.0

No release notes were published with this tag.

Patch releases

v0.5.1

No release notes were published with this tag.

v0.5.1 on GitHub

v0.4.0

No release notes were published with this tag.

Patch releases

v0.4.1

No release notes were published with this tag.

v0.4.1 on GitHub

v0.3.0

No release notes were published with this tag.

Patch releases

v0.3.14

No release notes were published with this tag.

v0.3.14 on GitHub

v0.3.13

No release notes were published with this tag.

v0.3.13 on GitHub

v0.3.12

No release notes were published with this tag.

v0.3.12 on GitHub

v0.3.11

No release notes were published with this tag.

v0.3.11 on GitHub

v0.3.10

No release notes were published with this tag.

v0.3.10 on GitHub

v0.3.9

No release notes were published with this tag.

v0.3.9 on GitHub

v0.3.8

No release notes were published with this tag.

v0.3.8 on GitHub

v0.3.7

No release notes were published with this tag.

v0.3.7 on GitHub

v0.3.6

No release notes were published with this tag.

v0.3.6 on GitHub

v0.3.5

No release notes were published with this tag.

v0.3.5 on GitHub

v0.3.4

No release notes were published with this tag.

v0.3.4 on GitHub

v0.3.3

No release notes were published with this tag.

v0.3.3 on GitHub

v0.3.2

No release notes were published with this tag.

v0.3.2 on GitHub

v0.3.1

No release notes were published with this tag.

v0.3.1 on GitHub

v0.2.0

No release notes were published with this tag.

v0.1.0

No release notes were published with this tag.

Patch releases

v0.1.2

No release notes were published with this tag.

v0.1.2 on GitHub

v0.1.1

No release notes were published with this tag.

v0.1.1 on GitHub