Get to know the workspace

Checked 2026-10-13. Unless noted otherwise, this page applies to Beav desktop 2.9.0.

Everything in Beav, including sources, chats, manuscripts, and generated results, belongs to an account space. A space usually corresponds to one social media account, one client, or one long-term project. If you cannot find something, check the current space first.

Create a space and run first-time setup

When you first use Beav and have no space yet, click New account, enter a space name, and create it. Beav goes straight to the main interface and opens a setup chat for that space.

• You already have an account: Enter the account's homepage link in the chat, confirm the account details Beav reads, and continue. You can click Paste on the right of the input box, press Ctrl+V (Command+V on macOS), or paste from the right-click menu. Links inside shared text are extracted automatically. Bilibili b23.tv short share links are also recognized.
• You do not have an account yet: Choose New account and talk it through with AI to settle your goals and content direction (see "Talk the direction through with AI" below).

Setup saves the selected content, downloads images and videos, and transcribes videos, then learns the account profile, writing style, and image style. Downloads and transcription wait up to 30 minutes in total. After that, setup continues with whatever was obtained and does not treat missing content as learned. You can use other features during setup, such as a new chat or browser extension capture.

Version 2.9.0 builds containing the October 12, 2026 update show media preparation in two stages: image and video downloads first, with a percentage, file count, downloaded size, and transfer speed; then transcription and text extraction, with the number of completed videos. Content without videos proceeds directly to account learning. Download progress weights files equally and uses the current file’s measured download fraction. When a media URL provides no file size, the display shows transferred bytes and speed without an estimated percentage. Each file has a 30-minute download timeout and a separate 15-second connection timeout. This update is included in 2.9.0; real platform downloads and desktop appearance remain to be verified.

Version 2.9.0 builds containing the October 13, 2026 fix prefer complete video files and try supplied backup URLs when a source fails. Interrupted temporary files are not reported as saved. Retrying the same source reuses a valid local file, and transcription failures preserve the original video. Transcription reads the actual audio stream and uploads an extracted M4A file, including for common media formats with incorrect extensions. Each file is limited to 1 GiB and a shared 30-minute budget across sources; while alternatives remain, one source can use up to 10 minutes. Missing audio, damaged files, playlist-only sources, and unsupported codecs fail explicitly. This fix is included in 2.9.0. Local fixture validation is separate from real platform testing; Windows hardware, network speed, and end-to-end ASR remain to be verified.

If setup pauses because you ran out of points or your sign-in expired, a Setup paused notice and the reason appear at the bottom. After you sort out the points or sign-in, return to the original setup chat and click Resume setup. Saved sources are reused, so you do not need to create the space again.

Version 2.9.0 builds with the retry fix from the same date try every pending file once, then retry only unsuccessful files, up to five retries after the first attempt. Reopening the app restores saved files and used attempts. After retries, if more than 50% of files have been saved, setup skips remaining failures, transcribes only downloaded videos, and starts style analysis with available sources. Counts still reflect files actually saved. Exactly half or fewer does not trigger this early handoff; the existing 30-minute preparation limit remains in place. A manual retry keeps saved sources and gives only unfinished files a fresh allowance. This behavior was checked against source on October 13, 2026 and is included in 2.9.0; real desktop and platform-network testing remain pending.

In new builds containing the October 23, 2026 change (not yet released), these cases no longer stop setup:

• The account has no posts yet, or the post list still can't be read after a retry: setup saves the account archive (avatar, name, bio, followers), writes those facts into the space profile, leaves positioning, audience, and writing style marked to fill in, and finishes right away. They fill in as you chat; after you post, sync the account archive and choose Learn more about my style.
• Image or video downloads fail: after retries, setup analyzes the saved text and whatever files downloaded, however many that is. This replaces the 50% rule above.

Talk the direction through with AI

In new builds containing the October 26, 2026 change (not yet released), setup no longer has the BPTI questionnaire, the organization positioning form, or the content direction picker. Instead you talk with AI in the setup chat:

1. AI starts with one open question: what do you want this account to achieve, and are there creators or posts you like or want to learn from? Paste homepage or post links, or just describe them.
2. AI reads the links and saves 3–5 representative posts per creator to References. It does not subscribe to anyone for you.
3. For anything still unclear, AI shows clarification cards to tap through, with up to 3 questions per card and at most two cards in the whole setup. Skipped questions take AI's recommended option.
4. To start right away, click That's enough, start now above the input box. AI stops asking and builds the space from what it knows; anything filled in from a recommendation is listed under "To confirm" in the profile, and you can change it in chat any time.

AI works out from the conversation whether this is a personal or an organization account. You can change it later in Settings → Personalization → Account type.

For an existing account, after you enter its homepage Beav first reads only the profile and post list, then asks which stage the account is in:

• Direction is set: learn from my posts: continues the learning flow above, reading top posts to learn your writing and image style.
• Still exploring: talk it through first: saves the account archive without reading posts in full, and moves to the conversation above.

The first option is recommended when the account has at least 20 posts and posted in the last 30 days; otherwise the second. If you don't pick within 10 seconds, setup continues with the recommendation. An account with no posts, or whose post list can't be read, saves the profile and goes straight to the conversation (replacing the "finishes right away" behavior above). If capture fails, Talk it through with AI instead opens the same conversation. Spaces already set up with the questionnaire or form are not affected. Real accounts and desktop interaction remain to be verified.

• Video transcription fails: analysis continues, and the AI says which spoken content it could not check.

Running out of points or an expired sign-in still pauses setup until you fix it. This change has been checked against source and unit tests only; real empty Douyin, X, and other accounts and desktop interaction remain to be verified.

You can rename the space later. The space directory uses a fixed ID, so renaming does not move existing files.

In the rebuilt 2.9.0 release dated October 13, 2026, deleting the current space opens another existing space when available, preserving its knowledge, assets and chats. Setup appears only when no spaces remain.

What setup learns

In builds with the October 20, 2026 update (not yet released):

• Broader sample: Beav reads the first two pages of the homepage (about 40 posts) and studies up to 12 representative posts in depth: the best performing, the most recent, and at least two each of videos, image posts, and articles. Titles, stats, and dates of the rest are also used for the overall picture. Image posts that arrive with only a title get one extra detail read, which may use a few more social credits.
• Faster: up to three videos are transcribed at the same time.
• Style per content type: if your account has clearly different kinds of content, such as reviews, tutorials, and opinion pieces, Beav records the structure, openings, pacing, and annotated examples of your own posts for each type. It then drafts a sample title and opening for each type and asks you to point out any line that does not sound like you.
• Who you are: Beav notes the identity, expertise, recurring views, stories, and offerings you have stated yourself, and lists them under "What I know about you". Tell the AI in chat if anything is wrong. These memories stay in the current space.

Ask Beav to learn more later

• In chat, say something like "pull more of my posts and learn my style again".
• Or on the account profile page, click Sync recent posts, then Learn more about my style in the pop-up.

Beav starts with archived posts it has not learned yet. If it needs to sync more, it shows the estimated credits and asks first. It then updates only what changed and tells you what is different.

Switch spaces quickly

Spaces are switched from the account menu at the bottom left. The account button shows your current space under your name; open the menu and the space list sits right below your profile. Click another space to switch. You no longer go through the full-page space manager, and the window no longer reloads. While the menu is open, press 1–9 to pick a space by position. Without the menu, press ⌥⌘1–9 on macOS or Ctrl+Alt+1–9 on Windows.

If you switch while on Home, Knowledge, Hotspots, Schedule and similar pages, you stay on the same page and see the new space's content. Switching from a chat or the manuscript editor takes you to Home. To create a space, use "New account space" below the space list. Renaming, deleting and opening an existing space are under "Manage account spaces…".

Applies to Beav desktop 2.9.0; not yet verified on a real desktop.

Tasks keep running when you switch spaces

Switching spaces only changes what you see. AI tasks, image and video generation, voiceover, transcription, and edit exports that have already started keep using the original space's sources and save their results back to that space. Switch back to the original space and open the original chat to see progress and results.

Left navigation

Entry	Purpose
New chat / Home	Pinned at the top of the sidebar. Home contains today's plan, review cards, the content schedule calendar, the daily report, the topic pool, and recent output. See Daily report and today's plan and Content schedule
Chats / Drafts	Switch between chat history and manuscripts in the same list
Knowledge	Saved notes, web pages, documents, subscriptions, and linked local folders. The top of the category bar opens Assets
Account profile	Each account profile in the current space. The first item is the space-level Manuscripts page
Hotspots	After authorizing NewsBing, view overall and platform trends. See Explore trends
Schedule	In newer versions it moved from the sidebar to Content schedule on Home (Month / Week / Year views). See Content schedule; older versions keep it in the sidebar
Tools & Skills	Shows the mini-tool catalog by default, plus the skill marketplace and plugins. See Built-in mini tools and Skills, plugins, and MCP

Click your avatar at the bottom left to open the account menu, where you can open Settings, membership, Use a redemption code, and Help and feedback. Drag the sidebar's right edge to resize it, or collapse it.

Account profiles

When you expand Account profile on the left, the first item is the current space's Manuscripts entry, followed by each profile with its avatar, listed by account. A space can hold several account profiles: click + to the right of Account profile and paste the account's homepage link. You can paste the whole share text: Beav keeps only the link and shows the detected platform's icon. Xiaohongshu, Douyin, and Bilibili share short links are first resolved to the account homepage, then the profile is created (resolving reads the account info once and may use credits). Adding a profile does not replace the space's original primary account or create a new space.

For profile sync, daily refresh, and data history, see Enter links, subscriptions, and account profiles.

Interface language

Switch between 中文 and English in Settings → General → Language. On first use, Beav picks the system language; if there is no match, it uses Chinese.

• Switching languages does not translate your existing articles, chat content, file names, or the names of mini tools you created.
• Beav desktop 2.8.14 brings all fixed text across the desktop app into the English interface, including Knowledge, manuscripts and publishing, the video editor, settings, membership top-up, and built-in mini tools. AI replies, progress notes, and summaries also default to the current interface language. If you explicitly ask for another language in a chat, your request takes precedence. A switch takes effect from the next reply.
• Original content returned by platforms, raw server error messages, and data you saved may still appear in Chinese.

Choose a model

You can choose a chat model in the new chat input box. Beav's official models show the output price as a multiplier of the official list price to the right of their names (for example, ×0.60). Non-members see both the standard multiplier and the discounted member multiplier; members see only their discounted multiplier. Actual point consumption follows the price list and usage records. The bottom of each reply shows the points it used.

The picker lists only the models of the provider that Chat and agents currently uses. If you route chat to a third-party provider in the model settings, the picker shows that provider’s models instead of the official models and their multipliers.

To connect your own model service, see Account, points, and model settings.

Next steps

• Chat and the right workspace
• Memory and creative style
• Manuscripts, asset packages, and publishing

Home session board

Checked 2026-10-13 against desktop 2.9.0 in Agent mode. Desktop interactions, narrow layouts, and model status declarations have not been manually verified.

Home groups chats into waiting for you, ready for review, and working. Completed chats can be expanded; archived chats are hidden. Click a card to continue the full conversation in a side panel or open the chat page. Viewing clears unread status without marking the chat done. Mark it done manually or move it using the context menu. To drop a session, choose Abandon from the ⋯ menu at the top right of the session or from the context menu; it is archived, leaves the board, and stays in archived chat history. When the AI needs your answer, a question card appears above the input: click an option (or press its number key), or write your own reply in the card and send it. Skip lets the AI go with its recommendation, and × only hides the card. It asks up to three questions, answered one by one and sent together (applies to the App version that includes this change). working status comes from the runtime. Failed refreshes retain existing data; focus the window again to retry.

The manuscript context menu offers draft, final, and published stages. AI can mark draft or final while preserving user labels. Published comes from publication evidence or a manual label and does not publish to a platform. When all linked manuscripts show published, the board groups the chat as done. The board replaces the former home approval cards and continue-creating list; topics, reports, and recent outputs remain.