Changelog

What changed in each release, newest first. A release is one pull request that publishes several packages at once: the table says which version each reached, and each change names the packages it went into. Read a release's notes before updating (Updating).

5 November 2026 · Release #62

The release's pull request

New

• Switching workspaces and notifications per person (Phase 13, part 2).

  • Switching. Someone who can open more than one workspace switches in the dashboard's sidebar and in the app's header, which opens a bottom sheet. The two session routes are POST /admin-api/session/workspace and POST /app-api/session/workspace (a session cookie only: agent keys can't switch); state and me now carry workspace and workspaces. POST /admin-api/workspaces/:id/open stays as an alias until the next minor version.
  • Pushes for every workspace. A device's subscription belongs to its person and gets the notifications of every workspace they're a member of. Links carry ?w=<id> and open in that workspace; a person no longer in it sees "You're no longer in that workspace." Tags are <workspace>:<target>, and when you can open more than one workspace the title starts with Workspace name · . Instance admins get pushes only from the workspaces they're members of (and the prefix, since they can open them all).
  • Notification groups. Five groups per person and workspace: Needs you, Problems, Published and results, New drafts and comments and Daily digest, set in the app's Settings → Notifications. Admins and creators get all five by default; editors get all but Needs you. Setting a group back to its default follows the role again. The workspace's ntfy topic and the audit log still get everything. Removing a member deletes their settings for that workspace, and their devices stop getting its pushes.

  Updating an existing instance (migration 0021 runs on its own):

  • Existing devices now get pushes from every workspace their person is in, not only the one they subscribed in.
  • Editors no longer get needs_you pushes (posts held past their time, and Shorts comments waiting for a working connection) by default; an editor who wants them turns Needs you on in Settings → Notifications.
  • The Workspaces page's Open still works; the sidebar's switcher does the same for anyone in several workspaces.

  core, api, web, dashboard, cloudflare, node, schema · 3d22c04

• A media folder, media stores and OAuth apps per workspace (Phase 13, part 3a).

  • Media folder. Each workspace has its own: the slug of the workspace's name (demo-travel) for one created since, UPLOADS for a workspace kept from before this release, editable by its admins in Settings → Media (Bongsu never moves files). Inside it: inbox/<project>/, <project>/…, _posted/, and the app's uploads in _uploads/ (they were a top folder, "Bongsu app uploads").
  • Stores. Besides the instance's, a workspace's admins add their own S3 bucket or Google Drive in Settings → Media. A workspace's default is its own store if it has one, else the instance's. A file reference must be in the caller's workspace (its own store, or inside its folder in an instance store), else it's refused with "not in this workspace's media". S3 ids and paths with ., .. or empty segments are refused, and a redirect from an S3 endpoint is an error, never followed. A broken store of a workspace is told to that workspace and links to Settings → Media. Agent API create_draft file ids are in the key's workspace's default store.
  • OAuth apps. A workspace can set its own Meta and Google apps in Settings → OAuth apps; a connection uses it if set, else the instance's, and keeps the one it started with. The redirect URLs are the same.
  • Inbox. Scans each workspace's <folder>/inbox in its stores, at most INBOX_WORKSPACES_PER_RUN workspaces per run, in turn (default 3 on Cloudflare, 20 on Node, at most 100; set at deploy time). The "inbox workspace" picker on the Media page is gone.
  • Scheduler. The Workspaces page has a read-only Scheduler section for instance admins: the three runs and their times, JOBS_PER_TICK, INBOX_WORKSPACES_PER_RUN, the last inbox run, and whether the schedules are off (SCHEDULES=off on Node). Making the schedule tunable is #64.
  • Dashboard. Get started's media step needs an instance store (a workspace's own bucket doesn't count). Settings has General, Media and OAuth apps tabs (/dashboard/settings, /settings/media, /settings/apps).

  Updating an existing instance (the migrations run on their own):

  • Nothing moves: a workspace kept from before this release keeps UPLOADS, and the app's uploads from before stay in "Bongsu app uploads" where they are; new uploads go to _uploads/.
  • New workspaces get a folder named from their name: create it in the store (or change it in Settings → Media) before dropping files in its inbox.
  • The inbox workspace chosen on a store is no longer used: each workspace's inbox goes to that workspace.
  • Optionally set INBOX_WORKSPACES_PER_RUN (in [vars] of wrangler.cron.toml, or .env on Node).

  core, api, web, dashboard, cloudflare, node, schema · aa8968a

• A workspace icon, and people's names per workspace (Phase 13, part 3b).

  • Icon. A workspace's admins set an optional icon in the dashboard's Settings: a square PNG, 192 to 1024 pixels, at most 200 KB (PUT and DELETE /admin-api/settings/icon). It shows beside the workspace's name in the dashboard's sidebar, in the app's header and workspace picker, and in its push notifications. It is served, public, at /icon/<hash>.png. The installed app's icon stays the instance's. iconUrl is now part of state, me and the settings.
  • Names. You set your own name in the app's Settings → You, and a name each workspace shows for you (POST /app-api/me/name and /app-api/me/display-name); an admin can edit a member's name for their workspace (POST /admin-api/members/:userId/name). Approvals ("Approved by"), pauses ("Paused by"), drafts, moved releases, the member list and the audit log name people this way. Agents see paused_by, approval.approved_by and each draft's by.

  Updating an existing instance (migration 0023 runs on its own): nothing to do. No workspace has an icon and nobody has a display name until someone sets one; people are shown by their name, else the start of their email, as before.

  core, api, web, dashboard, cloudflare, node, schema · 6348d13

• People, roles, invites and workspaces (Phase 13, part 1).

  • Roles. A person is an admin, an editor or a creator in each workspace, and the instance has instance admins. Admins run their workspace (connections, accounts, keys, settings, members); editors work on the plan, texts and drafts and read the rest, with no approving, pausing or connecting; creators use the app, where they plan, approve, pause, resume and connect their own accounts, and are pointed there when they sign in to the dashboard. Instance admins also get an Instance group: Workspaces, Media and OAuth apps.
  • First run. The setup token makes an instance admin, a user who belongs to no workspace, and the dashboard asks for the first workspace (a name, its time zone, its first admin: you or an email). Nothing lands in a workspace nobody chose: an inbox draft goes to the workspace whose plan has its project, else to the store's new Inbox feeds workspace (dashboard, Media), else it waits; a broken store is told to the instance admins.
  • Members and invites. A workspace's admins invite people by email with a role: a one-time link for 7 days, always shown to copy and emailed when a mailer is set. Admins and editors join in the dashboard, creators in the app, with a new passkey or the one they have. Admins change roles and remove members (cut off at once, and the invites they sent as an admin revoked); a workspace always keeps one admin. Instance admins make a member an instance admin, or stop them being one, from Members.
  • Workspaces page. Instance admins list, create, rename, archive and restore workspaces, add an admin and open one to work in it. An archived workspace stops completely (nothing publishes, keys and members are refused, no digest, no comments or metrics read). A workspace's own admins rename it in Settings. An instance admin in no workspace lands on the Instance pages.
  • Audit. A workspace's log shows only its own rows, each naming the person who acted, in the app as in the dashboard; instance admins also read the instance's (new workspaces, media stores, OAuth apps, instance admins).
  • Connect from the app. Admins and creators connect Instagram and YouTube from the app's Settings with the instance's OAuth apps; a connection lands in the workspace the flow started from.
  • Connect flows stay in their browser. Every OAuth flow (Instagram and YouTube from the dashboard or the app, Connect Google Drive) sets a short-lived cookie in the browser that started it, and the callback finishes it only in that browser: a provider link sent to someone else can no longer connect their account into your workspace. A flow someone started before the update is refused; start it again.
  • Pause per workspace. Pause this workspace and Resume with passkey act on one workspace; the others go on publishing. An agent key's pause_scheduler pauses its own workspace, and get_status and the digest say so.

  Updating an existing instance (migrations 0018, 0019 and 0020 run on their own):

  • Roles are converted: owner becomes admin, approver becomes creator and viewer becomes editor. People who were owners also become instance admins, so they keep Media, OAuth apps and the new Workspaces page; make anyone else an instance admin (Make instance admin in Members) only if they should have them.
  • Existing approvers become creators and so use the app, not the dashboard; if one needs the dashboard, change their role to editor or admin in Members.
  • The default workspace stays (it has your data), and so does its first admin. A fresh instance has no default workspace and no owner user: its first user is the one who signs up with the setup token.
  • The old instance-wide pause moves to the default workspace; any other workspace is not paused.
  • Each media store keeps feeding default from its inbox (Inbox feeds), as before; change it in the dashboard, Media, if another workspace should receive them.
  • Signed-in sessions keep working and open the person's first workspace. A person who belongs to several workspaces works in the first until a switcher arrives; an instance admin opens another from the Workspaces page.
  • Agent keys and CLI tokens are unchanged, except that a CLI token carries its person's new role.

  core, api, web, dashboard, cloudflare, node, schema · c6add56

4 November 2026 · Release #55

The release's pull request

New

• The ntfy topic is a setting of each workspace: the dashboard's Settings → Notifications (ntfy) sets, tests and removes it, and only that workspace's notifications go there. It's stored encrypted (it needs ENCRYPTION_KEY) and the page shows only its server. Bongsu posts to it without following redirects and with a time limit.

  To update: NOTIFY_URL is gone. If you used it, paste the same topic in the dashboard's Settings (migration 0016 adds the column) and remove the variable or secret.

  core, api, dashboard, cloudflare, node, schema · 057bff6

• Media stores are set up in the dashboard (Media), and there can be several side by side.

  • S3-compatible buckets: AWS S3, Cloudflare R2, Backblaze B2, MinIO. Instagram fetches the files from the bucket through presigned links and the app uploads straight to it (S3-compatible buckets).
  • Connect Google Drive with your Google app (the one YouTube uses): no service account any more (Google Drive).
  • One store of each kind, one of them the default; each file is read from the store it's in, the inbox is read in every store, and the app's upload page lets you pick the store. A weekly check marks a store that stopped answering.
  • Get started has a fifth step for it; the app's Settings and /health list the stores.

  To update: the MEDIA_STORE, DRIVE_SA_JSON and DRIVE_OWNER_OAUTH variables are gone. Set the store up again on the dashboard's Media page (Drive with Connect Google Drive; on Docker, This server's disk for the local store), and remove those secrets. Migration 0017 adds the media_store table. Signed media links now name their store (/m/<store>/<id>), so links made before the update stop working; they only last a few hours.

  core, api, web, dashboard, cloudflare, node, schema · de54272

Fixes

• Push notifications go only to the members of the workspace they're about: a device belongs to the user who turned notifications on and the workspace they did it in, and stops getting them when that user leaves it. Send a test reaches only your own devices. Push subscriptions are only taken for the browsers' push services, and Bongsu sends to them without following redirects and with a time limit. A device subscribed before this update keeps working for the default workspace's owner; others turn notifications on again in Settings.

  core, api · 18c6031

3 November 2026 · Release #47

The release's pull request

New

• pnpm brand <mark.svg> makes the app's seven icons from one SVG on BONGSU_THEME_COLOR, and the app's build stops with the name of any icon that is missing or at the wrong size. The project site reads its brand files from BRAND_DIR.

  web · 67038ba

Fixes

• @bongsu/core@0.3.0

  api, schema

• Every string in the app goes through the translation catalogue (i18n.ts), with no wording change; a test fails on English text left in the app's JSX or on a key the catalogue doesn't have.

  web · ff958db

• @bongsu/api@0.3.0

  • @bongsu/core@0.3.0

  cloudflare

• The Docker guide can run the published image: set BONGSU_IMAGE=ghcr.io/nyt87/bongsu:<version> in .env, then docker compose pull bongsu and up -d. Without it, compose builds from the checkout as before.

  node · eb0fd38

• @bongsu/api@0.3.0

  • @bongsu/core@0.3.0
  • @bongsu/schema@0.2.1

  node

3 November 2026 · Release #7

The release's pull request

New

• Each account's page. The inbox run samples every account's follower (subscriber) count once a day into a new account_metrics series (migration 0013); an instance with imported nightly snapshots gets their history once. An account's page in the app (Stats → an account) charts its followers (Total | Daily) and its posts' views per day over 7, 30 or 90 days, and ranks the posts and videos that brought followers. The Stats page reads followers from the series. The Stats and reviews routes now show only the caller's workspace's snapshots and reviews (they showed every workspace's).

  core, core, core, core, core, core, core, core, core, core, core, core, core, core, core, core, core, core, core, api, api, api, api, api, api, api, api, api, api, api, api, api, api, api, api, api, web, web, web, web, web, web, web, web, web, web, dashboard, dashboard, dashboard, dashboard, dashboard, cloudflare, cloudflare, cloudflare, cloudflare, cloudflare, cloudflare, cloudflare, cloudflare, node, node, node, node, schema, schema, schema, schema, schema, schema, bongsu (command line), bongsu (command line) · 839cc12

• How many publishing jobs a run advances is now a setting, JOBS_PER_TICK: Cloudflare keeps 4 (the free plan's subrequest limit), Node and Docker default to 20 since they have no such limit, and either can set it (up to 100). The configuration guide explains how items due at the same minute go out.

  core, node · 6b61b80

• The Stats tab reads only what Bongsu collects or receives: each account's followers from the account series, then the last 7 days' posts and videos, most viewed first, each opening its Results or video page. The snapshot Performance page and the review pages are gone with their routes (/app-api/performance, /app-api/report), and migration 0015 drops the reports table (export it first if an instance imported reviews). The snapshots table stays: it backfills the account series once.

  api, web, schema · 0af344f

Fixes

• A failed post (three failed tries, or a video the platform never finished processing) comes back when you approve it again: an approval made after the failure starts it over with a new container. A job whose publishing call was made stays failed, since the post may be live: check the account, then record it as posted by hand.

  core, core, core, core, core, api, api, api, api, api, api, api, api, api, api, api, api, api, web, web, web, web, web, web, web, dashboard, dashboard, dashboard, dashboard, dashboard, dashboard, dashboard, cloudflare, cloudflare, cloudflare, cloudflare, node, node, schema, bongsu (command line) · 839cc12