Use the original UID format consistently across API, username, and store validation so UUID-based callers continue to work. Regenerate API documentation and add regression coverage for UUID memo IDs.
Share canonical user and memo query caches across creator, reaction, comment, mention, and relation lookups. Reuse list data for relation snippets and cover overlapping queries with regression tests.
- Cache fingerprinted frontend assets for one month.
- Load media, rich renderers, dialogs, and editors only when needed.
- Simplify seeded demo content and cover deferred-loading behavior.
Validate new user-provided IDs using the AIP-122 format while retaining legacy UID compatibility. Correct resource annotations and canonical names returned by user stats.
Settings drops the all-in-one bordered card for a de-carded layout in
the property-rail design language: a sticky table-of-contents rail at
md+ (Settings wordmark, uppercase group labels, quiet anchor rows) and
a horizontally swipeable chip strip below md, replacing the mobile
section dropdown. Nav items are real anchors with aria-current, and
switching sections scrolls back to the top.
Access Tokens becomes a first-class section with an explainer panel:
what a PAT is and a copyable curl example (real instance origin,
memos_pat_ prefix) beside token-safety guidelines in a two-column band,
with a Learn more docs link and the tokens table beneath. Successful
PAT authentication now records the token's lastUsedAt asynchronously
inside resolveBearer, with a clone-before-mutate cache guard and
monotonic writes in the store, surfaced in a Last used column.
Also localizes the create dialog's 90 Days label, lets the My Account
row wrap instead of clipping on narrow screens, and drops the dead
select-section key from all locales.
Login on the demo instance is SSO-only, so visitors always arrive as a
fresh user and consume seed content through Explore. Rebuild the seed
around that funnel: four personas (steven the maintainer, alice, ben,
zoe), a public feed mixing short captures with long-form anchors, and
protected memos that visibly appear after sign-in to teach the
visibility model.
Every headline feature is now demonstrated: nested tags, memo
references, comments with mentions, reactions, locations, a database-
stored image attachment, and a rendering test memo covering math,
mermaid, code, tables, and task lists.
Rework MemoDetailSidebar into a Linear-style rail: a quiet icon action
cluster (pin, copy link, share image, share links) followed by
label/value property rows for visibility, created/edited times, author,
and location, then hairline-separated sections for tags, attachments,
relations, and the outline. Attachments and reference relations were
previously not surfaced in the sidebar at all.
- Visibility is editable in place, reusing VisibilitySelector with a
new compact size variant that matches the 13px row grammar
- Pin toggle and visibility changes go through useUpdateMemo
- MemoOutline gains active-section tracking (scroll-spy), depth
indentation, and tick-rail styling; drops the animated underline
and native title tooltips
- Widen the detail rail from w-56 to w-60
Remove the zero-padding line override that replaced CodeMirror's built-in horizontal spacing and caused the editor content regression reported in #6093.
Disable completion icons through CodeMirror configuration, preserve placeholder and dark-theme contrast, and remove redundant editor style overrides.
Fixes#6093
Rebuild /about around an identity hero: instance branding with custom
profile fallbacks, a demo badge, and version/commit chips linked to the
GitHub release and commit. Dev builds show a plain version chip and hide
the unknown-commit chip.
Project links become an annotated directory (homepage, docs, API docs,
GitHub repo), followed by an MIT license note. The bird sprite gallery
moves out of the page, and the version/commit block leaves the settings
sidebar now that /about owns build info.
Rebuild the auth pages around a shared AuthPageLayout card with a
three-tier guidance hierarchy: the form (or provider buttons) on top,
a centered switch sentence below it, and a quiet explore band anchored
to the card foot on public instances.
- add AuthPageLayout (brand row, chip, empty state, link prompt,
explore band gated by instance URL with hideExplore opt-out)
- extract CredentialFields shared by sign-in and sign-up forms
- extract IdentityProviderButtons; fetch the provider list via a
cached TanStack Query hook with an enabled flag
- cover every instance config: SSO-only, password-only, registrations
closed, private instance, nothing enabled, first-run setup, admin
- promote /auth/signup, /auth/admin, /auth/callback to ROUTES and add
appendSearchParams for param-preserving auth links
- move the explore link out of AuthFooter (selectors only now)
- refresh en and zh-Hans auth strings; drop dead keys
Rework ListUsers to match the ListMemos pagination contract: opaque
PageToken, normalizePageSize, DB-level limit+1 look-ahead, and a
next_page_token. Adds Offset to store.FindUser with an OFFSET clause in
all three dialects, and an `id DESC` ORDER BY tiebreaker so offset pages
stay stable when created_ts ties.
Also align pagination across list endpoints:
- Bump DefaultPageSize 10 -> 50 to match the documented default; use
normalizePageSize in ListAttachments.
- Remove the never-implemented total_size field from all six list
responses (ListUsers, ListAttachments, ListMemoComments,
ListMemoReactions, ListUserSettings, ListPersonalAccessTokens) and
regenerate.
- useListUsers now pages through next_page_token so the admin members
view still loads every user past the default page size.
- preserve grid layout while the editor is in focus mode\n- extract save orchestration into a dedicated lifecycle hook\n- simplify code formatting controls in the toolbar
- intercept file paste and drop before CodeMirror inserts file text
- reconfigure localized placeholders without remounting the editor
- add regression coverage and document the CodeMirror review
Keep caret and placeholder behavior on CodeMirror's standard extensions, and move height and scrolling to the editor elements so empty space stays interactive across browsers.\n\n- disable unused multi-cursor selection\n- upgrade CodeMirror state and view patches
The CodeMirror editor only grew to its content height, so in focus mode
(where the host is tall) the empty area below the text was bare host
markup outside the editor. Clicking there could not focus the editor or
place the caret, so it appeared to vanish in "empty" space. Give
.cm-editor height:100% — it collapses to auto in normal (auto-height)
mode and only fills when the host has a definite height.
Fixes#6076
Newly-created memos were pinned to column one and never released, and cards
that grew after placement (late images/comments) never rebalanced, so the
columns drifted badly out of balance.
- Leave the priority (just-created) memo's column assignment transient, so a
superseded memo rebalances instead of piling up in column one forever.
- Self-heal: re-pack from a clean slate when a late height change leaves the
columns lopsided, adopting it only when it meaningfully shrinks the spread.
- Animate only the adopted rebalance; resizes, column-count changes, growth
reflows and first paint stay instant so widths and positions stay in lockstep.
The card-shaped skeleton never matched the loaded memos (heights vary
with markdown, images, and tags), making the swap feel jarring. Use the
same centered spinner the grid layout already used, still gated behind
the 250ms delay so fast loads render no indicator at all.
Ref https://github.com/orgs/usememos/discussions/6066
Reorganize the toolbar into inline marks (bold, italic, strikethrough,
inline code) and blocks (lists, fenced code block), with the Type glyph
for the paragraph/heading picker. Marks derive from the shared MARKS
table; code block toggling reuses the same selection-probe lookup so
toggle-off always agrees with the button's active state. A fresh empty
strikethrough pair (~~~~) parses as a tilde code fence, so empty-pair
removal and code-block detection recognize doubled mark tokens.
Compact cards previously had two truncation systems taped together: a
text-only clamp inside MemoContent (144px, no re-measure on image load,
fade tinted to the wrong surface) and a whole-tile cap in ColumnGrid
(hard clip mid-image, no affordance), later joined by a content-shape
rule choosing an owner per card — which could still stack two Show-more
buttons and silently broke footnote navigation in clamped cards.
Now there is exactly one mechanism: ClampedSection measures its content
and folds anything taller than 420px to a 360px preview with a fade and
a Show more/less toggle. MemoBody wraps the whole body in it (reactions
stay outside, never hidden); ColumnGrid is pure packing again and never
clips; MemoContent is a stateless renderer whose `compact` prop only
informs footnote-link behavior; MemoPreview uses a static CSS bound in
place of its previously inert toggle.
Deltas: text-only compact previews grow from ~6 rows to the same 360px
image cards get, and pinned memos clamp like any card (pinned means
ordering and a badge, not size).
Replace the layout Select with a lean 28px segmented control: Rows3 /
Columns2 / Columns3 / Infinity icons on a quiet muted track, only the
active option filled. Each option carries a tooltip with a label and an
honest description of the ceiling semantics ("up to N columns when the
width allows").
The control is a proper radiogroup — roving tabindex, arrow-key
movement, aria-checked — restoring the single-choice semantics the
Select used to provide. Options derive from ViewContext's canonical
value list, and one exhaustive record maps each value to its icon and
wording, so a future column option fails to compile until both exist.
The Compact mode row now stays visible in multi-column layouts, shown
as on and locked (multi-column always renders compact tiles); the
stored preference is untouched and resurfaces at a single column.
Replace the single-column-only feed with a max-columns model: one
Columns setting (1 / 2 / 3 / ∞) where 1 is the reading list and
anything wider packs into a Google-Keep-style column grid. There is
deliberately no separate list/grid mode — the setting is a ceiling,
and widths that only fit one column fall back to the flow list.
- ColumnGrid: absolute-positioned packing that only translates cards,
so appends and reorders never remount them; sticky column assignment
keeps existing cards in place when new memos arrive; tiles are capped
at 360px with a fade; columns clamp to 420px and center.
- Column one is the action column: the composer and active filters
stack as its first tile, and a just-created memo is pinned directly
beneath them.
- Multi-column always renders compact cards (policy centralized in
PagedMemoList and threaded through renderer(memo, { compact })).
- Setting persists in ViewContext localStorage; the settings menu
derives its options from the context's canonical value list.
Run the instance in private mode when instance_url is not configured: the API rejects anonymous requests except the auth-bootstrap set (sign-in, token refresh, instance profile/settings, SSO providers, share-link access) plus first-run user creation, and the web UI redirects anonymous visitors to /auth instead of /explore. Setting instance_url keeps the current public behavior. Access tokens and personal access tokens are never gated.
Enforcement lives in a shared Authorizer used by both the Connect interceptor and the gRPC-gateway middleware; the file server applies the same rule to public-memo attachments and avatars. Also merges the duplicated Authenticate/AuthenticateToUser token dispatch behind resolveBearer, dedups the AuthContext unauthenticated state, extracts the redirect decision into a pure shouldGatePrivateInstance helper, and prints the access mode at startup.
- Insert menu: replace the nested "More" submenu with a flat dropdown —
insert actions plus inline Focus Mode and Formatting-toolbar toggles.
- Formatting toolbar: remove the link button and its window.prompt flow;
the link command stays in the shared catalog for future surfaces.
Bold (and italic/code) now remove an empty just-inserted delimiter pair instead of nesting more asterisks, and strip real marks whichever way the selection was made; clicking Link inside a link unwraps it instead of inserting [](); the three list modes convert between each other (bullet on a task line no longer leaves '[ ] ...' behind), toggle off any ordered number, respect indentation, and apply across multi-line selections; list/heading edits keep the cursor in place (task insert on an empty line lands after the marker).
Structurally, highlight and toggle now share one line-mode detector so they cannot disagree; the heading regex is shared with headingDecorations, so indented headings render styled; one MARKS table drives all inline-mark behavior; run("link") is total (URL doubles as label on empty selection, now inserted inline rather than as its own block); the unused getSelectedText contract method and the dead markdown-list-detection.ts util are removed.
Known divergence, not addressed here: markdown-task-actions.ts TASK_LINE_REGEXP accepts ordered task items ('1. [ ]') which the editor's TASK_LINE does not; the viewer toggles such checkboxes but the editor toolbar won't detect them.
Add a localStorage-backed insert-menu toggle (default off) that surfaces the formatting toolbar outside focus mode, and redesign it as a lean inline row: ghost icon buttons grouped by thin dividers, a level-reflecting heading glyph (pilcrow / H1-H3), and the active command as the only filled control.
Also keep the editor focused when using the toolbar: command buttons preventDefault on mousedown, and the dropdown menus return focus to the editor on close.
Footnote definitions previously vanished on display. Two underlying issues:
- rehype-sanitize re-clobbered `id` attributes with a second `user-content-`
prefix while leaving hrefs untouched, so footnote refs/backrefs pointed at
ids that didn't exist. Disable clobbering (ids are already namespaced by
remark-rehype) so anchors match their targets.
- Footnote anchors went through the external Link (target="_blank"), opening a
blank tab instead of navigating. Route in-page `#` anchors through a new
AnchorLink: scroll within the memo when shown in full, else navigate to the
memo detail page with the hash, where MemoDetail scrolls it into view.
Also style the footnotes section GitHub-style: thin separator, smaller muted
text, and un-underlined ref/backref links.
The location chip used the kit Button (shrink-0 + whitespace-nowrap),
which grew unbounded so its inner truncate span never fired and long
addresses overflowed the memo card. Replace with a raw button that
caps width (max-w-full min-w-0) and truncates the address text,
matching the existing LocationDisplayEditor pattern.
Rebuild the memo editor as a single CodeMirror 6 "decorated source"
editor. The document is the raw markdown, stored verbatim and styled in
place (markers stay visible), so the editor never serializes a tree back
to markdown — removing the round-trip fidelity bug class (inline images,
setext headings, ordered-list indentation, HTML entities) that the old
editor needed per-case patches for.
- MemoEditor/Editor: CodeMirror 6 + lang-markdown (GFM). Tokens, heading
lines, #tag/@mention, and the autocomplete popover are styled in plain
CSS (Editor/editor.css) with theme tokens, not a CSS-in-JS theme.
- Tab/Shift-Tab nest/outdent list items (marker-aware, ordered items
renumbered so nesting is CommonMark-valid); Escape blurs; #tag
autocomplete sourced from useTagCounts.
- Focus-mode toolbar reimplemented as markdown-text ops; active state read
from the Lezer tree via the backend-agnostic formatting/commands catalog.
- Remove the old serialize-back-to-markdown editor (its Editor dir,
PlainEditor, the editor-mode system) and its now-unused dependencies
(marked, textarea-caret, and the rich-text editor packages).
- Consolidate toolbar components under Toolbar/. Read-only MemoContent
rendering is unchanged.
Make the OpenAPI-driven MCP surface more usable by agents, following the
mcp-builder guidance.
- Enrich proto descriptions (single source of truth, flows to OpenAPI + MCP
tool descriptions): document the memo `filter` CEL grammar with fields and
examples (replacing the dangling "Refer to Shortcut.filter"), clarify the
created_ts/updated_ts vs create_time/update_time naming, the visibility
enum, the declarative replace semantics of Set* ops, and steer tag filters
to `"x" in tags` (not the unsupported `tag == "x"`).
- Mark SetMemoAttachments / SetMemoRelations idempotent via a per-operation
override the HTTP-method heuristic can't express.
- Curate two read-only orientation tools: shortcut_list_shortcuts (surfaces
reusable CEL filters) and auth_get_current_user (the single allowed
auth/identity op, for resolving the current user); guard test updated to
keep the rest of the auth/user surface excluded.
- Add a task-level evaluation suite (server/router/mcp/evals) with 10
verified questions, pinned to the deterministic demo seed.
Mirror the #tag mark so the TipTap editor styles @username while typing
and on load, matching the read-only view. A shared mention grammar
(utils/mention-grammar.ts) feeds both the editor tokenizer and the
read-only remark renderer so they can't drift.
Styling only -- no autocomplete dropdown (ListUsers is admin-only and
there is no user-search RPC). Bare emails still autolink to mailto: and
are never treated as mentions: the tokenizer's start() skips email-glued
@s so marked's GFM autolinker keeps seeing the whole address.
Generate webhook signing secrets server-side and let users reveal them on
demand, replacing the create-dialog secret controls that surfaced internal
mask state (Status / Generate & Copy / Clear / Pending) to users.
- Add owner-gated GetUserWebhookSigningSecret RPC — the only path that
returns the secret; list/create/update responses still omit it.
- Generate the secret server-side on create (webhook.GenerateSigningSecret),
so validity no longer depends on the client.
- Rename UserWebhook.has_signing_secret -> signing_secret_set for parity
with the existing api_key_set field.
- Create dialog drops the secret section to a one-line note; the generated
secret is shown once right after create and revealable from Edit later.