# Concepts

A few key ideas to help you get the most out of Treasure Studio.

## Projects

A **project** saves your working context so you don't repeat yourself across conversations. Think of it as a preset that loads your preferences, files, and skills every time you start a new chat.

Each project includes:

| Field | What It Does |
|  --- | --- |
| **Name & Description** | Identify the project in the sidebar and project switcher |
| **Working folder** | The directory on disk the AI can read and write files to |
| **Instructions** | Custom rules the AI follows — coding conventions, preferred databases, output formats |
| **Skills** | Skills that are always active in this project's chats |
| **TD context** | Default database, table, or segment the AI uses when none is specified |


**Example:** A "CDP Ops" project might set the working folder to `~/td/cdp-configs/`, activate segment and journey skills, and include instructions like "Always validate YAML before pushing."

Use `Cmd+P` to switch projects instantly. Your chat history is preserved — switching only changes the configuration for new conversations.

## Working Folder

The **working folder** is the AI's file system sandbox. When the AI generates files — query results, segment definitions, charts, reports — they go here. The AI can also read files from this folder, which is useful for reviewing existing configs or iterating on exports.

Set a working folder per project, or configure a global default in settings.

## AI Credits

Each AI interaction consumes credits. The amount depends on which model you use and how much the AI processes (longer conversations and tool-heavy sessions cost more).

| Model | Relative Cost | When to Use |
|  --- | --- | --- |
| **Opus** | Highest | Complex multi-step analysis, building segment rules, writing workflows |
| **Sonnet** | Medium | General-purpose work, moderate reasoning tasks |
| **Haiku** | Lowest | Quick lookups, simple questions, conserving budget |


## Context Window

The **context window** is how much conversation the AI can "remember" at once. Studio shows the current usage as a percentage indicator.

![Context window indicator](/assets/studio-context-window.1a31e7df465c30399a216b262c7f878f2643cc076165630287bc1b0fda9b897a.5c07e18c.png)

As you chat, the percentage grows. When it gets close to the limit, Studio's **auto-compact** feature kicks in — it summarizes older messages to free up space without losing important context. Your conversation continues normally.

You can also manage context manually:

- `/context` — See detailed context usage breakdown
- `/compact` — Summarize the conversation to free up context space
- `/clear` — Clear the context entirely and start fresh


For best results on long-running work, use [projects](#projects) to carry context across multiple chats rather than pushing a single conversation to its limit.

## Model Selection

Studio supports multiple Claude models. You can switch models mid-conversation — useful for starting with Haiku for quick exploration, then switching to Opus for the complex part.

![Model selector](/assets/studio-model-selector.7b9d54a8d0f9520c38aea783b242271889ad715fe94509ea542386a5281ee04c.5c07e18c.png)

| Model | Strengths |
|  --- | --- |
| **Opus** | Deepest reasoning. Best for multi-step data analysis, complex segment logic, and code generation. |
| **Sonnet** | Balanced speed and quality. Good default for everyday tasks. |
| **Haiku** | Fastest and cheapest. Great for quick lookups and simple operations. |


Use the model selector dropdown in the chat to switch at any time.

## Next Steps

- [Introduction](/treasure-code/studio) — Explore the full feature set
- [Skills](/treasure-code/studio/skills) — Install agent skills for specialized tasks
- [Connections](/treasure-code/studio/connections) — Connect external services