Skip to content
June 18, 2026 · Summer Test Kitchen

OpenClaw for Work

Most AI tools run in someone else's cloud. OpenClaw is an agent you run yourself, so your files and data stay on your own machine. Xiaohan Liu shows how she uses it for real work with sensitive data: analyzing autonomous-driving issues, generating slides from that analysis, and building a personal research digest. Then everyone tries it hands-on through a Discord bot, no setup required.

Presenter: Xiaohan Liu (Autonomy Systems Engineer, DiDi Global) · 2:30–4:30 p.m. · SCDI 3115, Santa Clara University

Watch the recording

Watch on YouTube →

What is OpenClaw

Xiaohan opens with a short, non-technical definition of OpenClaw, then makes the case for when this approach beats a regular chatbot like ChatGPT, Claude, or Codex. A few things set it apart.

  • Private, local data. Your files stay on your own machine, so you can work with sensitive data without uploading it to a third-party service.
  • Multi-model cross-checking. Connect different model APIs and have them check each other, which catches hallucinations and gets a better result than any single model alone.
  • Lives in your chat apps. It runs as a bot inside tools like Discord or Telegram, can stay always-on, and a group can share one deployment, which is how today's exercise works.

Reach for it when the data has to stay on your hardware, you want to mix or run local models, or you want an always-on agent inside a chat app your team already uses. For a quick one-off or a scheduled job, a managed tool like Claude or Codex is usually simpler.

What Xiaohan will demo

Three short projects, drawn from her real work and personal use, shown in order. The first one introduces the idea of a reusable "skill": a workflow you package once, then point at new data to run end to end.

Demo 1
Issue Analysis
Turn raw autonomous-driving issue data into insight: which issues happen where, by city and road structure. A reusable skill extracts the data, writes a script to plot it, and outputs a markdown report, all end to end.
Demo 2
PowerPoint Generation
Turn the issue-analysis results into a basic deck. Less about slide design (covered in an earlier session) and more about how the agent manages its context and memory across a multi-step task.
Demo 3
Personal Information System
Xiaohan's own use case: the agent pulls from AI-lab blogs and papers and summarizes them into a daily report. New to AI Kitchen, and the demo she plans to spend the most time on.

Also covered

  • Choosing a model. A live comparison of Opus, GPT-5, Kimi, and GLM, with the pros and cons of each for tasks like analysis and presentation.
  • How agents remember. How OpenClaw manages memory (today, plain markdown files), possibly with a lightweight memory add-on (mem0) and a look at Hermes as a more advanced, self-improving example.
  • Inside a real company. A look at SmartWork, DiDi's internal OpenClaw-style tool that wraps their chat app, email, and simulation database, so you can see what production agent tooling looks like.

Try it yourself

After the demos, everyone works with the agent directly. Nothing to install, it all happens in the AI Kitchen Discord.

The hands-on exercise

  1. Open the AI Kitchen Discord. Go to the Workshop area, where the openclaw.didi bot is running for the session.
  2. React to open your private channel. In the Workshop area, react with an emoji to Linus's message. That opens your own private OpenClaw channel, an independent session, so your work never gets crossed with anyone else's. We'll walk through this live, so nothing to set up beforehand.
  3. Work with the sample data. Xiaohan provides a sample autonomous-driving dataset (download it below). Discuss it with the bot and have it produce a markdown report on the data distribution, building on the Issue Analysis demo.
⬇ Download the sample dataset CSV · 1,000 rows of synthetic data
No setup required. Running your own OpenClaw instance takes real work (a server, external services, and API costs), so for the session you just use the shared Discord bot. If you want your own instance afterward, stick around and the team can help you get started.

Going further

Part of the AI Kitchen Summer Test Kitchen series at Santa Clara University.
Questions? klukoff@scu.edu · Full schedule.