New posts:89 (spread across active threads like the new mobile auth UX component read more, an SEO canonical/hreflang deep-dive read more, and a composer mentions edge-case bug report read more).
New topics:12 (including a new webinar announcement read more, a new theme component to “show” hidden categories read more, and an API support question on deleting drafts read more).
Terminal client “termcourse” gained hi‑res image expansion support read more
Reactions-disabled sites want a simpler “Like” UI again read more
A self-hosting admin hunted for the update UI location (/admin/update) read more
Interesting Topics
#Announcements
lindsey announced an improvement to permanently deleting posts: the can permanently delete setting is now visible to admins (with additional confirmations and a safety delay), removing the need for console work or contacting staff read more.
Related: this continues the long-running “permanent delete” feature evolution under new-feature / permanently-deleteread more.
Data Explorer got more charting options, and hosted customers asked when it would land; staff clarified it should appear on the next deploy and even confirmed a deployment in-thread read more.
Related (Data Explorer maintenance): handling deprecated columns via ignored_columns was discussed separately read more.
A new webinar was announced: “Community Retention: How to Define and Drive Return Behavior” (May 14). The outline frames retention as returning at the right time for the right reasons, not just maximizing activity read more.
#Theme-component
Lilly shipped Discourse Mobile Auth Button (loginsignup): a theme component that replaces the mobile Log In button with a combined Sign Up / Log In entry point, and then iterated with a modal UX improvement for smoother onboarding read more.
Handy docs referenced in the release: install guide read more and the themes beginner guide read more.
pfaffman posted Show hidden categories, a theme component that adds a “fake” category entry that links to a group page so users can request access—plus a candid comparison to existing solutions like Category Previews read more.
Related: Category Previews came up as the likely default choice for most sites read more.
A confusing Onebox behavior: Google Photos album links can render like a single image with no hint it’s an album link. The thread includes redirect/OG tag sleuthing and a call for contributions under pr-welcomeread more.
Related: older context suggests behavior changed over time read more.
Sites that have reactions disabled (i.e., Likes-only) are asking to restore a single Like button UI because the current layout can look like two hearts competing and confuse newcomers read more.
Extra context: users specifically want the prior “nest under Like” experience while keeping the newer popup behavior read more.
A nuanced SEO question for multilingual communities: translated pages served via Content Localization’s crawler parameter (?tl=) appear to canonicalize to the base language, raising concerns that Google may treat translations as duplicates and ignore hreflang. Staff agreed: each language variant should be self-canonical and said they’ll investigate read more.
Tag context: seo and content-localizationread more.
A nasty editor/data growth issue: underscores in image alt text may get double-escaped on each edit, exponentially inflating raw post size until the post becomes effectively uneditable. Staff suspected a regression and started investigating read more.
AI translation quality control: a Portuguese report showed auto-translation sometimes breaks markdown tables (missing |), and a fix involved retriggering translation + pointing reporters to the “inaccurate translation” flag option and the meta gardening hub read more.
Related: the collaborative content gardening hub was suggested as a reporting channel read more.
A sharp edge in composer reachability warnings: uppercase letters in usernames can break the “mention is not reachable” checks in restricted categories, because the backend lookup returns lowercased keys. A concrete code pointer + downcasing fix suggestion was provided read more.
Related symptom surfaced while debugging Private Topics behavior read more.
On iOS PWA, the Assign UI reportedly does not allow typing into the assignee search field (users can only pick from a partial dropdown), making assignment painful for larger sites read more.
An AI translation investigation concluded what looked like one issue was actually three overlapping problems (locale detection, translation pipeline failures, and category targeting quirks). Staff clarified how to override locale via composer controls to ensure correct translation behavior read more.
API usability feedback from a “headless Discourse” implementer: requests for more consistent pagination, cleaner separation between topic metadata and post streams, and generally better field-level documentation for modern frontend architectures read more.
Draft deletion via API key auth: a developer hit not_found deleting drafts via the API despite it working in-browser; staff suggested a workaround and noted an inconsistency, linking to a fix in progress around falling back to Api-Username when ?username= is missing read more.
lindsey shipped a safety-and-admin-usability upgrade to permanent deletion controls, making can permanently delete visible in-admin with extra confirmations and safeguards read more.
nat agreed that Content Localization language variants should be self-canonical (SEO/hreflang correctness) and committed to taking a look read more.
nat also confirmed a hosted deployment for the new Data Explorer charting changes (“Your site has been deployed!”) read more.
Danielle announced the upcoming retention webinar and outlined practical retention frameworks and engagement levers read more.
Danielle also introduced the purpose of the new Enterprise category, framing it as a home for community building in large organizations read more.
pmusaraj encouraged contributions on improving Google Photos Onebox behavior by tagging the thread with pr-welcomeread more.
pmusaraj helped clarify AI translation behavior and highlighted how to override locale correctly via composer controls read more.
zogstrip acknowledged the underscore double-escaping report as likely a regression and began investigating read more.
zogstrip also identified an API consistency gap in draft deletion and referenced a fix to align username handling read more.
Falco pointed AMA planners at Discourse’s newer nested-replies mode as a cleaner fit for Q&A-style conversations, and recommended contacting support if they want it enabled read more.
Falco expanded emoji alias translation work by pointing to an upstream PR in the emoji repo read more.
kris.kotlarek merged a fix related to “clearing advice” not actually clearing, crediting the contributing PR read more.
mcwumbly answered when new Data Explorer chart features would appear on hosted instances (“next time your site is deployed”) read more.
markvanlan added guidance on where nested replies make sense (very large, fast-moving communities) and emphasized it won’t be recommended for most sites read more.
That’s a wrap: may your canonicals be self-referencing, your mention checks be properly downcased, and your mobile login buttons be delightfully obvious.
lindsey shipped the April roundup in Current Projects, covering recent wins like full-app embedding (read more)—with pointers to improvements like embedding as a comment system (read more), simplified solved setup (read more), and easier topic-voting category creation (read more).
nicolsdennis previewed Domniq, a Discourse-native Android + iOS client with account integration, notification controls, and moderation actions (read more); follow-up discussion also referenced UI inspirations like Chat bubbles (read more) and the Horizon theme (read more).
nicolsdennis also shared a companion “premium landing page” plugin for unauthenticated users—hero variants, stats blocks, trending topics, contributor highlights, and more (read more).
A community member floated “MNN / Meta Monthly News”, proposing a member-written newsletter and asking Meta if there’s appetite for it (read more).
#Site feedback
After multiple “I made a thing” posts landed, mcwumbly suggested a lightweight convention: keep them in General and tag them show-and-tell—at least until a clearer pattern emerges (read more).
#Feature
ganncamp asked for an “Obnoxious” notification level for incident threads—something more forceful than Watching, ideally enabling desktop notifications per-topic rather than site-wide (read more).
Canapin requested a way to make “view posts by @user” filtered topic views shareable/linkable, so useful author-only views can be passed around without hand-editing URLs (read more).
Backup performance returned to the spotlight: a proposal to migrate from gzip to zstd argued for significantly faster compression and smaller archives, with follow-up notes about docker image tooling requirements (read more); related compression/backup threads also saw renewed attention (read more).
A thorny seo + localization issue: translated pages weren’t ranking due to canonical URL behavior, and nat reported a fix shipping in the newest build while the thread stays under watch for regressions (read more).
A composer regression report claimed default_composition_mode was being ignored (markdown sites defaulting users into rich text), and sam walked through repro checks, safe-mode testing, and likely plugin culprits (read more).
Moderation workflow pain point: automation-driven flags generate bland “reviewable item #123” notifications; the request is to include more context like topic titles—similar to standard user flag notifications (read more).
A support mailbox question explored whether archiving could be per-volunteer (hide items “not mine” without archiving them for the entire group), to prevent important requests from being missed in high-volume group inboxes (read more).
Mention validation: uppercase letters in usernames were reported to break reachable-mention checks in the composer, and zogstrip outlined a safer incremental approach rather than merging a large sweeping fix at once (read more).
#Translations
In German, Moin asked why main-branch translations appeared delayed; gerhard explained the merge slowdown (people away + a temporary breakage) and confirmed things should return to normal after merging the missing updates (read more).
#Theme
A major theme compatibility warning: FKB Pro reportedly crashes on Discourse 2026.4.0-latest due to references to a removed Ember compatibility shim (ember-this-fallback/js-component)—a reminder that older theme code may need explicit this. migration (read more).
lindsey published the April status newsletter (read more) and followed up on translation readiness for blog styling components in the March thread (read more).
zogstrip weighed in on the composer mention bug involving uppercase usernames (read more), posted progress on meta tag cleanup work (read more), and flagged a tricky markdown corruption edge case around underscore escaping in image alt text (read more).
chapoi confirmed the team is reverting to a single like button for sites without Reactions, telling folks they can “stop adding +1s” because it’s actively being handled (read more).
davidb merged the change to restore one like button behavior (and move the count back to the right) on reactions-disabled sites (read more), replied to follow-up observations in the “display all likes/reactions” thread (read more), and posted an update that work is ongoing for the mainEntity.suggestedAnswer structured-data issue (read more).
gerhard responded to concerns about delayed translations, summarizing what went wrong and confirming missing i18n updates were merged (read more).
mcwumbly set expectations on where to post “I made a thing” content (use show-and-tell in General) (read more) and offered ideas for safer posting workflows in the “future self / trusted friends” approval discussion (read more).
nat reported a shipped fix for translated-page canonical behavior impacting Google indexing, advising affected sites to update and continuing to monitor the report thread (read more).
sam investigated reports that default_composition_mode was being ignored, recommended safe mode (no downtime), and helped narrow the issue down toward plugin interactions (read more).
j.jaffeux noted the team plans a better upcoming solution than the “User Field Prompt” theme component, and deprioritized further work on the component itself (read more).
If yesterday taught us anything, it’s that Meta can ship shiny new show-and-tell demos and still find time to un-break canonicals and un-double the like button—efficient chaos at its finest.
Default email subject lines are getting simpler/less technical, removing redundancy like [%{email_prefix}] for “sent from the community” messages, in Simpler email subject lines (email). The discussion also surfaced cases where people still rely on subject “context,” like category indicators and subcategory display needs (see Subcategory display options in email templates).
A thoughtful request: should contributors be notified when a topic disappears (deleted, made private, or otherwise removed), so they don’t lose track of time invested? That idea—and the context of a category accidentally being public—came up in Random thought (read more).
A provocative “show-and-tell” raised eyebrows: using AI bots to repopulate a mostly-dead forum, plus discussion about disclosure and what “community” means when content isn’t organic, in I populated my (mostly-dead) Discourse forum with bots! (show-and-tell). The replies dug into transparency expectations and the “Dead Internet” concern (read more).
A community-driven “newsletter about Meta” concept continued to be debated—especially around incentives and avoiding popularity contests—in MNN Meta Monthly News (read more).
Discourse AI translations behavior under model churn: if you change the LLM provider/model, do translations restart? The answer: old translations are kept, and the new model only fills gaps, in What happens to translations when LLM changes? (ai) (read more).
A practical monetization question: selling access to instructional videos inside a Discourse community. The thread quickly turned to architecture tradeoffs (hosting vs. embedding, storage headaches, and “members-only” options) in Videos and Monetizing my Discourse community (read more).
Admin workflow papercut: changing a member’s email can produce an “Access Denied” when clicking the verification link while still logged in as admin—leading to talk about impersonation and “admin should be able to change anything” expectations in Admin changing email address of a member is still quirky!. A related “fast path” workaround was referenced via console tooling: Change staff email address from console (read more).
A plugin regression report: a Discourse update seemingly disabled category settings for a plugin’s buttons, in Trading Buttons (read more).
A theme perception issue: someone questioned whether a default pattern element resembles a swastika, and the thread touched on “what users notice first” vs. intent in Graceful Theme (read more).
lindsey published a product-facing walkthrough of the new defaults (and rationale) for tidier notifications in Simpler email subject lines, aiming to make Discourse emails feel less “technical” in members’ inboxes.
That’s a wrap—may your email subjects be simpler, your emoji searches multilingual, and your “show-and-tell” ideas just organic enough to avoid summoning the Dead Internet discourse again.
Coverage anchors (sample of the active window): bots-in-Discourse experimentation (read more), AI triage troubleshooting (read more), and self-hosting install/debugging on HestiaCP (read more).
Top users (posts / likes in this window, with a recent reference):
“I populated my (mostly-dead) Discourse forum with bots!” (show-and-tell) — kirupa shared how clearly-labeled bot accounts can provide instant, high-quality technical replies and curated content sources, plus ideas like “resurrect old posts” with modern commentary (read more; read more). chapoi noted it’s “not exactly community,” but a genuinely interesting way to keep up with interests (read more), while Ed_S suggested a periodic “blast from the past” approach and flagged the risk of hallucinations (read more).
Support bot strategy: Discourse AI vs Chatbot (plus vision experiments) (aiai-bot) — In a long-running thread, 37Rb described experiments using vision capabilities to interpret photos (e.g., alarm panels) and provide wiring advice—promising in tests, but not yet something customers used in practice (read more).
#Feature
“Post approval by ‘future self’ or ‘trusted friends’?” for episodic unwellness — Ginger_Marcus shared a real-world moderation tactic: shrinking the edit window dramatically (e.g., ~60 seconds) to reduce impulsive posting and keep replies anchored to what people actually read at the time (read more), followed by discussion about balancing accountability with compassion for genuine health challenges (read more; read more).
#Site-feedback
Rebuilding the Meta branded theme: icons, custom homepage, and the Blocks API (meta-redesign-2026) — The redesign work continued to spark “site transformation” excitement, with nathank calling out how the tooling could help smaller orgs simplify their overall website stack around Discourse (read more).
Easy way to roll back a wiki page? (Solved) — A quick support exchange clarified that the revert control exists in edit history for admins/mods (read more), but category moderators don’t get “Revert Edit Revisions,” as ondrej pointed out with a permissions reference (read more; read more). The question also ties back into best practices around wiki usage (read more).
AI triage examples not sent properly? (Examples being interpreted as the content to moderate) (automationai) — A practical gotcha: examples were being treated like prior turns, causing every post to be flagged because the example itself contained bank details (read more; read more). Falco recommended leaning into tool calls (e.g., the flag tool) and emphasized that examples must mimic the expected model behavior—including tool-call shaped responses when tools are expected (read more).
“They are sent as previous turns… they need to mimick the exact expected LLM response.” (read more)
What happens to translations when the LLM changes? (Tokens consumed, but 0 translation progress) (ai) — A troubleshooting report described heavy daily token usage with no apparent translation progress; Falco suggested enabling verbose logs and inspecting the AI API audit log table for the latest entries (read more; read more). This also connects back to core Discourse AI capabilities and configuration expectations (read more).
Videos and monetizing a Discourse community — The thread explored monetization complexity, especially when content creators must be paid (raising legal and operational questions beyond just implementation) (read more).
Discourse logs sent to wrong S3 — A debugging angle suggested looking outside Discourse for old sync jobs (cron / rclone / aws s3 sync) that may still be pushing files to a legacy bucket even after configuration changes (read more).
#Self-hosting
Installing on HestiaCP: domain/port issues + a puzzling GitHub/DNS failure — A self-hosting install attempt hit multiple layers (ports 80/443 contention, proxy templates, and outbound fetch failures). The follow-up revealed the “GitHub issue” was resolved by switching away from Cloudflare/Google DNS to internal DNS in that environment (read more; read more).
Can /var/discourse and /var/lib/docker live on different volumes? (server-resources) — The consensus: Discourse doesn’t care where Docker stores its data, and mounts can go where they make operational sense; the remaining open question focused on whether a split-volume setup would create noticeable performance impact in practice (read more; read more).
Accessing the Discourse DB with DBeaver (safely) — A hands-on guide proposed exposing Postgres via a custom mapped port and restricting access to a private network (e.g., Tailscale), plus adjusting Postgres auth to allow password logins instead of default peer-only behavior (read more).
Connecting Discourse with two OIDC providers (openid-connect) — A reply argued you may not need “two OIDC providers” at all—suggesting one OIDC plus Microsoft Auth, and noting the relevant localization namespaces differ between OIDC and Microsoft login UI (read more). The post linked related references for Microsoft Auth (read more), OIDC button text customization (read more), and Microsoft 365 OAuth2 text issues (read more).
#Extras
Discourse Shortlink Router: “teleport” support URLs into the right admin path — A new utility replaces your forum base URL with a router domain, preserving the path so helpers can share “go here in admin” links faster (without repeatedly describing navigation steps) (read more).
#Plugin
Locations Plugin early-access invites for sponsors (paid-support) — merefield announced early access invites rolling out to current sponsors first, with daily sponsor-list review and a planned cadence for migrating select features to the public repo later (read more).
#Theme
FKB Pro theme: update to v2026.4.0-latest triggers missing Ember module error — A report flagged a production-breaking client error referencing a missing ember-this-fallback/this-fallback-helper module after upgrading, with a request for a fix (read more).
chapoi weighed in on “community by bots” — Reacted positively to the execution/styling and framed bot-populated Discourse as an interesting “stay up to date on your interests” pattern, even if it’s not traditional community interaction (read more).
Falco debugged AI triage behavior and improved prompting strategy — Recommended using the Triage with AI Agent automation type and calling the flag tool rather than returning raw strings (read more), then clarified why “examples” can backfire if they don’t mimic the exact expected output format—especially when tool calls are involved (read more). Separately, Falco advised turning on verbose logging and checking audit logs to diagnose runaway translation token usage (read more).
Yesterday, Meta managed to be both a bot-filled sci‑fi forum and a URL-teleportation lab—with a quick pit stop to teach AI agents how to behave along the way.
Notable: mcwumbly’s clarifications helped land a solution on Discourse ID username changes (read more) and kicked off UX triage on author-change visibility (read more).
Interesting Topics
#Feature
Ethsim2 proposed decoupling passwordless “magic link” email login/signup from enabling local password auth—aiming for local accounts + email verification + no password as a first-class configuration (read more). Follow-up discussion touched on making the UI clearer (and possibly hiding password fields) when admins want the passwordless flow (read more).
Discourse ID users can now change their username/name directly in user settings, even if the account was created via ID—confirmed by mcwumbly in the accepted solution (read more). This resolves long-standing friction around identity-linked profiles (read more).
Ethsim2 asked whether allowed email domains is enforced against the OIDC-returned email during account creation, seeking confirmation of how domain restrictions apply in openid-connect setups (read more). (This pairs closely with ongoing login policy tuning in the passwordless thread above: read more).
A Discourse AI translation debugging thread surfaced repeated rate-limit failures and suspicion of a “stuck” translation loop, with verbose logs showing token/day exhaustion on a specific model (read more). The report includes the exact error payload and timing, useful for diagnosing retry behavior (read more).
#Self-Hosting
An IPBv4 migration report described partial avatar imports, user collision issues (existing admin blocking post migration), and multiple content-rewrite/reaction conversion gaps—highlighting the tricky edge cases in real-world imports (read more). The post also notes link rewriting and missing <a href>-wrapped text as high-priority cleanup items (read more).
A Cloudflare best-practices discussion got a fresh field report: blank login page on first load plus broken embedded PDFs, persisting even when Cloudflare is paused (suggesting something more subtle than cache toggles) (read more). The author also shared what they already tried (free Cloudflare + cache rules bypass attempts), helping narrow troubleshooting paths (read more).
A self-hoster revisited the long-running “SSO bridge alongside discourse_docker” topic, asking whether a 5-year-stale OIDC bridge repo is still viable—because they want Discourse as an identity provider for apps like Nextcloud and Mailcow (read more). This continues to be a common architecture question for communities consolidating auth systems (read more).
A subtle trust/visibility gap: Edit history visible to public appears not to cover changing the author/owner of a topic/first post, which matters for admins who “ghostwrite” by posting on behalf of users then reassigning ownership (read more). mcwumbly flagged it as likely bug or missing feature and moved it into UX contribution space to discuss expected behavior (read more).
A mobile title-wrapping report (long titles overflowing) may already be fixed by updates, with mcwumbly noting they couldn’t reproduce on multiple themes and suggesting affected sites update (read more). The reporter later confirmed Meta had been updated on their end (read more).
A UI consistency nit: the “delete/discard” icon differs between the composer and edit flow, with a pointer that it may relate to a recent change discussed in “Do you want to discard your post?” (read more; and related: read more). The original report includes side-by-side screenshots to illustrate the mismatch (read more).
#Site feedback
The refreshed Meta Branded theme homepage continues to draw “how do I install this?” questions—answered plainly: the Meta theme isn’t packaged for reuse right now, and updates should be done via your own theme config panel instead (read more). The thread also reiterates staff intent that the theme is Meta-specific and evolving, not currently intended for redistribution (read more).
#Extras
A new “Discourse Shortlink Router” sparked a privacy and transparency mini-debate: Jagster pressed on whether the service operator can see destination/site info (read more), and darkpixlz responded with implementation details, logging examples, and what isn’t stored server-side (read more). The exchange underscores how support tooling often needs clear privacy posture to earn trust (read more).
The ICS → Discourse importer thread added a useful nuance: --time-only-dedupe isn’t purely time-based because it still checks location “closeness,” which can be beneficial—room changes surface as new topics while location noise doesn’t spam Latest (read more). The post frames it as a practical “signal vs noise filter” for calendar-driven publishing (read more).
#Plugin
The Community Network Visualisation plugin got “some love,” including a snazzy 3D view option shown via screenshots—an eye-catching upgrade for communities wanting richer relationship mapping (read more). This is a notable iteration on an older plugin thread that’s still evolving with fresh improvements (read more).
#Community Building
A practical community strategy question: how to reconfigure an existing community for a new product launch when moving from one product to a two-product company (read more). mcwumbly prompted for context about product portfolio structure (read more), and noahl suggested working through an “audience venn diagram” to assess overlap and segmentation (read more).
mcwumbly confirmed a recent platform change enabling Discourse ID users to self-serve username/name changes, marking the solution on the support thread (read more).
mcwumbly assessed author-change visibility as likely a bug or missing UX coverage, and moved the discussion into UX contribution for deeper review (read more).
mcwumbly tested and reported inability to reproduce the mobile long-title overflow issue across several themes, suggesting affected sites update if they still see it (read more).
mcwumbly asked clarifying questions about restructuring a community for a second product—probing whether the org is transitioning from single- to multi-product and what that implies for information architecture (read more).
If yesterday’s Meta was any indication, today’s motto is: keep your magic links passwordless, your edit history honest, and your community graphs gloriously 3D (read more, read more, read more).