2026/08/29 v0.3.0 Latest
Universal Client and content workbench A shared information architecture for Web and Client with explicit local storage, online sync, and agent execution choices.
2026/08/08 v0.2.0
Reviewable multi-channel delivery Derive channel versions from a master draft and deliver them safely through packages, the browser extension, and receipts.
2026/07/12 v0.1.0
From source capture to a traceable master draft The first complete path through capture, knowledge, content memory, topics, and agent edit proposals.
A layered Universal Client with shared console views across desktop and mobile.
Local workspaces stored in device SQLite with offline startup and no automatic upload.
Per-workspace sync cursors and local read-only snapshots for online workspaces.
Independent choices for workspace storage and local or DraPub Cloud agent execution.
Shared navigation contracts, visual tokens, authentication views, and core page geometry across Web and Client.
Client sign-in sessions now live in SQLite; explicit sign-out removes the local token.
Mobile layouts use a single column, 44px touch targets, and safe-area constraints.
The initial mobile release uses DraPub Cloud Agent and does not run desktop-class runtimes inside the phone process. Signing in or syncing an online workspace does not upload local workspace content.
Separate master drafts and channel versions with independent title, body, media, and review state.
Immutable content packages that cannot enter publishing execution before human approval.
A Chrome Manifest V3 extension for current-page capture and approved-package draft delivery.
Publishing receipts with platform content ID, state, time, and failure diagnostics.
Platform cookies, tokens, and tickets are never sent to DraPub Backend.
The extension stores the DraPub sign-in token in chrome.storage.session only.
Delivery tasks persist before execution to reduce duplicate drafts after service worker interruption.
Publishing review now places sources, diffs, media, and destination in one decision context and records accept, reject, and return reasons.
Capture from files, web pages, manual input, Cubox, Feishu, and Obsidian.
Knowledge spaces that organize original sources, editable documents, and provenance.
Searchable pages, chunks, vectors, and facts through GBrain.
Topics and channel-neutral master drafts with source evidence.
Previewable agent edit proposals that write only after confirmation.
Backend provides the public HTTP, tRPC, and Agent SSE gateway. The same codebase consumes background work in work mode. Database and brain access stay inside Backend infrastructure while public SDK, CLI, and Skillpack use safe interfaces.
The early release prioritizes provenance, review, and state integrity. Some media parsers and platform adapters return explicit failures when a source is unsupported instead of creating empty evidence.