TRAE Work Deep Hands-On: From One-Sentence HTML Pages to Building Your Own Stock Screener

TRAE Work is ByteDance's AI-native workbench, syncing across desktop (Windows / macOS) plus web and mobile. Its core difference isn't "a stronger model"—it's closing the loop on "talk while it works": generate a shareable web page from one sentence, turn a GitHub repo into a local database, run scheduled jobs that update data automatically.

This article starts with a 3-minute HTML hook, then runs two real tasks (a self-built stock screening system and batch paper clustering) through TRAE Work's core capabilities, and closes with a methodology portable to any AI tool.

Official knowledge base (400K-word AI work tutorials): bytedance.larkoffice.com/wiki/RxVXweukSi7JtIkhOKBcuSlbnfC

Before You Start

• Install: desktop supports Windows and macOS (both Apple Silicon and Intel builds available)
• Three modes: Code (coding, debugging, Git workflows), Design (designing, revising, finalizing), Work (documents, data, presentations—the focus of this article)
• GitHub account: section two uses TRAE Work's direct GitHub connection
• Optional: your own server (for deploying the workflow as a permanent URL)

The Hook: Turning a Long Document into an Interactive Web Page in One Sentence

TRAE Work's newly launched HTML capability is the best entry point for understanding the whole workflow. It doesn't generate page code—it directly generates something you can deliver.

Full Flow (About 10 Minutes)

Enter a one-sentence request—for example, turn 4 WAIC 2026 conference summaries (tens of thousands of words scattered across four links) into a conference overview page:

Turn these four WAIC 2026 conference summaries into a reader-facing conference overview page, four themes in four blocks, each block leading with the core takeaways, details expandable on click, and it should read well on mobile.

TRAE Work decomposes the task automatically, executes step by step, and shows progress on screen. In our test, generating a dynamic page from 4 sources took about 9 minutes 48 seconds.

Two Ways to Revise

• Select + comment: circle whatever's wrong, tell it how to fix it, and everything else stays untouched
• Direct editing: like editing in Word, with alignment, font size, bold, and other formatting

Get a Link + Roll Back

After editing, hit share to generate an outward-linkable URL. The recipient installs nothing and can open it on a phone, and the link always shows the latest version. Hidden feature: a rollback button that instantly returns to the state before any edit.

The whole chain: one sentence → generated page → select-edit / manual edit → get a link → recipient opens it on a phone.

💡 The test: will this artifact be opened by more than one person, and opened more than once? If both answers are yes, it's worth building in TRAE Work.

Deep Build 1: A Self-Built Local Stock Screening System

⚠️ Disclaimer: all stock content in this section only demonstrates tool capability and does not constitute any investment advice.

The goal of this task is to build a complete stock-screening workflow from scratch: pull data → land it in a local database → feed in strategies → output results → deploy online → auto-update.

Step 1: Solve the Data Source

Use the open-source instock stock data repo on GitHub. TRAE Work's differentiated capabilities show clearly here:

• Direct GitHub connection: upload finished code without wrangling SSH keys
• Have TRAE Work build the system on the local repo; it reads the entire repo first, then lists a task plan: analyze the code → set up the database → test the features

When it finished, the output: basic info complete for 4,589 stocks, market data all in place.

💡 Key lesson: verify every step AI completes. Have it build a visual database dashboard listing which tables exist, how many rows each has, and how current the data is. Manually spot-check two stocks to confirm the numbers are correct.

Step 2: Feed It the Strategy (Natural Language, Not One Line of Code)

The strategy is all trader jargon—"pullback on shrinking volume," "long lower shadow," "breakout on heavy volume." Send it verbatim:

From A-share stocks that are non-ST, non-delisted, listed over a year, with average daily turnover above 100 million yuan, filter those in an uptrend (close > 20-day MA > 60-day MA) that have pulled back 3–8 days recently with volume shrinking stepwise to below 70% of the 5-day average, showing a doji or small candlestick with a long lower shadow near the 20-day MA or a prior breakout platform... rank by trend strength, volume shrinkage, and support validity, and output the top 10.

The key closing line: add one sentence to the end of the prompt—

When data is insufficient, do not guess or fabricate recommendations.

TRAE Work first queries the database and fills in the missing historical candle data, then translates the jargon into computable rules line by line, finally producing the qualifying results (2 stocks in our test), each with indicators, entry conditions, stop-loss reference, and a one-line rationale.

Step 3: Multi-Channel Delivery

Have TRAE Work output the same results to three channels:

• PPT: in research-report form, ready in minutes
• Feishu multi-dimensional table: daily K-lines and key indicators stored in, ready for your own filtering and sorting
• HTML web version: opens on a phone, nothing to install

Step 4: Deploy as a Permanent URL

Start /plan plan mode—have TRAE Work produce a deployment plan first, then, with the Server Skill, deploy the whole system to your own server, get a real live URL, and add password protection.

Step 5: Scheduled Jobs That Clock In Automatically

Following the official knowledge base's automation tutorial, configure a scheduled job. Spell out three things in the dialog:

1. When to run: before market open every trading day
2. What to do: sync the latest market data
3. Notify me when done

The next morning your phone buzzes: data sync complete.

The full chain: pull data → translate strategy → run the screen → generate the display → deploy online → auto-update.

Deep Build 2: Batch Paper Clustering into a Knowledge Base

Task goal: organize 60 papers on Loop and Agent topics into one Feishu sheet—categorized, clustered, with the research lineage visualized.

Have TRAE Work produce in one shot:

• A table organizing all 60 papers one by one
• Clustering results grouped by theme
• One visualization that makes the whole research field's lineage legible

Advanced Play: Trend Briefings Built on the Table

A few days ago Grok 4.5 launched, touting graph-theoretic reasoning. We simply had TRAE Work pull the graph-reasoning-related papers out of the existing table and explain, against Grok 4.5's public information, what the trend is actually new on and which ideas the papers had already covered. A few minutes later, a briefing arrived.

💡 Key point: the table's standalone value is limited—the value is that the knowledge is now in a database. Any later follow-up question, comparison, or trend tracking can be produced from that table in seconds.

The Core Insight: Intelligence Is Rented, Workflows Are Yours

After running these two tasks, the biggest takeaway isn't TRAE Work tips—it's a process portable to any AI tool:

• How to decompose a request (natural-language strategy + a "no guessing" constraint)
• How to write prompts (role + data + acceptance criteria)
• How to verify (visual checks at every step)
• How to persist (design specs, data tables, Skills)

A model works today; tomorrow it might raise prices, change rules, or ban accounts. But the methods built up while working with the model run just the same on another tool. That's also why the TRAE Work team folded this process into the official knowledge base as a general template—tools change, methodology doesn't.

Use Cases

TRAE Work's "closed-loop workflow" creates the most value in these scenarios:

• Independent researchers / investment research: self-built databases + scheduled updates + multi-channel delivery
• Content creators: meeting notes, industry roundups, weekly and monthly reports turned into HTML pages in one shot
• Academia / research: batch paper clustering, knowledge graphs, trend briefings
• Operations / product: data reviews, event summaries, product requirement documents, job portfolios

Entry points:

• TRAE Work knowledge base
• Full workflow templates

Copy a knowledge-base link and send it to your own TRAE to use.