Release notes and roadmap
Roadmap
Pro on the web
Answers that cite their sources
Chrome extension
Organizations
Sign in with Google and Apple
One-click Notion connection
Second on the App Store
Obsidian community plugin
Action history
Public REST API
Merge duplicates
Direct handoff between AI apps
Per-chat folder access
Releases
Web 1.6.0NewFixes
Highlights
- Add global capture search across metadata, saved content, PARA locations, and Archive with URL-backed filters, pagination, and an unindexed-data fallback.
- Make permanent item and PARA-node deletion discoverable across web and desktop, preserving node captures in Inbox and syncing deletion cleanup.
Fixes and improvements
- Add a standards-compliant Agent Plugin package for Second's production MCP server and capture/search/PARA workflows.
- Use the supplied Second product artwork across web, extension, and desktop branding and packaging surfaces.
- Use the official Chrome Web Store listing for extension installation links.
- Make
secondthe CLI and Homebrew distribution name, and update the public tap through the sanctioned release flow. - Make Start over permanently delete ownership-scoped Second data, revoke access credentials, queue search-tenant cleanup, and converge local-first clients with sync tombstones.
- Move filing rules into a user-facing Automation settings tab and keep Developer focused on API tokens.
- Keep the authenticated app shell within a bounded viewport so the global header stays fixed while route content scrolls in one accessible area.
Web 1.5.1Fixes
Fixes and improvements
- Enforce v1 OAuth scopes and validate sync mutation values.
- Recover interrupted sync item deletions on retry
- Use stable keyset cursors for desktop item discovery.
Web 1.5.0New
Highlights
@second/web adds:
- Migration 0010:
sync_ops(the oplog table —entityType/entityId/field/value/hlc/deviceId, composite(userId, hlc)index for pull, composite(userId, entityType, entityId, field, hlc)index for LWW lookups and compaction). POST /api/v1/sync/push,GET /api/v1/sync/pull?cursor=,GET /api/v1/sync/bootstrap— the standardauthorizeV1+ zod + rate-limited v1 pattern. Push applies per-op LWW via hlc comparison, is idempotent under retry (a resubmitted op with an already-recorded hlc reports accepted without double-applying), and returns the canonical row for losers. Pull is one indexed range query. Bootstrap returns the full current snapshot (items via the same mappingGET /api/v1/itemsuses, para nodes, rules) plus a starting cursor.- Every existing mutation of a client-mutable field (item title/note/paraNodeId/archivedAt; para-node name/goal/deadline/standard/sortOrder/archivedAt/completedAt; rule create/delete) now writes its oplog row in the same
db.batch()as the row mutation, across both the web app's server functions and the v1 REST routes — a device syncing sees edits made from any client, not just ones that go through push. Deletes are tombstones (field: "_deleted") rather than a silent row disappearing; hard-delete-after-tombstone-propagation stays a deferred, separate decision, and this cycle's compaction policy is written to never prune a field's current/latest row, so a tombstone is never pruned by it either. - The capture-ingestion workflow's two terminal steps (materialization failure, revision activation) write a synthetic
_contentStatusoplog row (reserved"server"deviceId) so pull observes ingestion completing without instrumenting every intermediate status transition — the one deliberately-scoped-down part of "server-mastered fields flow down"; seeapps/web/CONTEXT.mdand this change's own documentation for the boundary. - A new daily cron (
0 3 * * *, alongside the existing 15-minute integration sync and daily review digest) compacts superseded oplog history behind a 30-day retention window. No device-registry table: cursors are stateless, client-held opaque hlc strings — a device that hasn't synced in over the retention window falls back tobootstraprather than an incrementalpull, a defined degraded path, not data loss.
@second/api-client gains typed syncBootstrap, syncPull, syncPush methods, response-validated against the new @second/types/sync schemas.
GET /api/v1/transitions?days=— recent transitions (was web-only).GET /api/v1/items/:id/transitions— per-item transition history (was web-only).POST /api/v1/para/:idgains a"complete"action (project-only; archive/restore unchanged).POST /api/v1/items/archive-all— the information-bankruptcy reset (was web-only).POST /api/v1/rules/offer— rule-offer suggestion, including its "already covered by an existing rule" check (was web-only).GET /api/v1/url-metadata?url=— bounded URL preview; now goes through the same public-URL SSRF guard (assertPublicHttpUrl) used by capture extraction, which this endpoint was missing.GET /api/v1/searchgains cursor pagination (cursor=,nextCursorin the response) past the previous silent 20-result cap, bounded by AI Search's own 50-result ceiling; existing no-cursor callers are unaffected other than now correctly honoring alimitabove 20.GET /api/v1/itemsgainssince=(ISO or epoch-millis, filters onupdatedAt) andorder=updatedAt(ascending) for delta sync pulls, alongside the existingcreatedAtDESC default. Fulfils the?since=wish noted atapps/obsidian-plugin/src/sync.ts:146.POST /api/v1/capturesaccepts an optional client-suppliedid(UUIDv7-shaped). A repeat of the same id by the same user returns the existing item as a duplicate-style 200 (idempotent outbox retries); the same id under a different user is rejected with 409.
@second/api-client gains typed methods for all of the above (listRecentTransitions, listItemTransitions, archiveAllItems, offerRule, getUrlMetadata, widened transitionParaNode, and since/order/cursor options on listItems/searchItems). @second/types gains the shared captureIdSchema (UUIDv7 shape) on capture inputs.
Web 1.4.0New
Highlights
- Rebuild URL content extraction on Browser Run quick actions: one
snapshotpass renders the page (HTML + whole-page Markdown + status/title), then an html-modemarkdownpass prunes page chrome in the browser and converts — replacing the HTMLRewriter clean + Workers AI conversion path. Transient Browser Run failures (429/5xx) now retry through the ingestion workflow, exhausted retries mark the revision failed instead of leaving it stuck materializing, bot-walled preflight fetches no longer block the browser render, and rendered page titles replace hostname-placeholder item titles.
Web 1.3.0New
Highlights
-
Captures now store the content instead of the page. Per-host parsers (x/twitter, youtube, reddit, hacker news, github threads) build Markdown from the page's own metadata, and everything else goes through an HTMLRewriter pass that strips nav, footers, cookie walls and login prompts before conversion — applied on both the Browser Rendering and AI-conversion paths. Images are kept as images end to end rather than being rewritten to links at storage time.
Clicking a capture now opens
/app/items/$itemIdwith the ingested content, not the source site, and a signed-in visitor landing on/is sent to the Inbox. Ingestion no longer marks a capture failed when only search indexing fails, so materialized Markdown stays readable. Fixes horizontal page scroll caused by the sidebar inset refusing to shrink, inbox rows overflowing instead of truncating, and inbox lists loading every revision's full Markdown to render a 280-character excerpt. The extension popup lists its keyboard shortcuts, read from the browser so they reflect user remaps.