Your Repository's Mentat
"It is by will alone I set my mind in motion."
Welcome back,
You'll need this token in Step 5
You'll need this token in Step 5
"Every sietch has its own ways. Choose wisely."
The Mentat requires certain instruments
First, get a Claude Code subscription ($20/mo). Once you have it, open your terminal and run these two commands — the first installs the Claude Code CLI, and the second logs you in with your Anthropic account:
npm install -g @anthropic-ai/claude
claude auth loginnpm install -g @anthropic-ai/claude
claude auth loginRequires Node.js (v18+). If npm isn't recognized, install Node.js first.
Your AI colleague uses GitHub to manage code. Install the GitHub CLI and log in with your GitHub account:
brew install gh
gh auth loginwinget install GitHub.cli
gh auth loginDon't have a GitHub account? Create one here — it's free.
Create a new folder for your project and initialize it as a git repository. Replace my-project with whatever you'd like to name it:
mkdir my-project && cd my-project
git initmkdir my-project; cd my-project
git initDownload Typeless for speech-to-text. Instead of typing, you can talk to your AI colleague by voice — it transcribes as you speak.
Access the spice reserves
Request access if you don't have it. You'll need a workspace login to proceed.
We'll add this screenshot together
In Databricks, go to Settings → User Settings → Developer → Access Tokens. Generate a new token and save it somewhere safe.
We'll add this screenshot together
Go to SQL Warehouses, click your warehouse, and copy the ID from the connection details.
We'll add this screenshot together
Open Claude Code in your project and run:
claude "Check my Databricks connection and list available schemas"Initialize your repository's intelligence
Your token:
Make sure you're inside your project folder (from Step 3), then run this command. It will ask for your token — paste the one from Step 1 above:
npx thufir initConfirm Claude Code is installed and working. You should see a version number:
claude --versionStart Claude Code and talk to your new AI colleague. That's it — you're ready for the workshop:
claude
The pre-work is complete. Your tools are sharp.
The spice will flow when the council convenes.
Workshop begins in
--:--:--
"The sleeper must awaken."
Give your colleague deep knowledge of your company's data systems
This is the big one. This prompt teaches your Mentat everything about how data works at Toloka — the four layers, the source tables, quality scoring, naming conventions, and how to build a project-specific data mart. Copy it and paste it into Claude Code.
Loading prompt...After your Mentat processes this, it will set up your Databricks connection and save the knowledge for future sessions.
The first step to understanding is naming things precisely
Now talk about YOUR project. Use Typeless to speak naturally for 2-3 minutes about what you manage and what data you need. Then invoke the brainstorming skill.
Two ways to do this:
/brainstorm Define a data glossary for my area of the projectPlease use the brainstorming skill to help me define my project's data glossary. I'll tell you what I manage.The brainstorming skill will ask you questions one at a time to help you define precise terms. Use your own language — it adapts to you.
Turn your glossary into queryable data
Your glossary becomes SQL views in your Databricks namespace. Ask your Mentat to plan and build them:
Build my project-specific data mart from the glossary we just defined. Use brainstorming to plan the views first, then create them in my namespace.Your Mentat already knows the source tables, naming conventions, and performance patterns from the setup prompt. It will create the views, the SQL files, and the documentation.
When it's done, push your work: git add -A && git commit -m "feat: initial data mart" && git push
Ask questions. Get answers. Debug what breaks.
Ask your Mentat anything about your data. Keep it simple:
How many people are in each pipeline step right now?More examples — adapt to YOUR project:
Use systematic debugging. Tell your Mentat:
The result doesn't look right. Use systematic debugging to investigate.Continue exploring at home — your Mentat remembers everything. Push your work before you close: git push