Skip to Content

Projects, Knowledge Base, and System Prompts

Section 3 — Lesson 2

Projects, Knowledge Base, and System Prompts

Stop repeating yourself in every conversation. Learn how to build persistent Claude workspaces with your own documents, rules, and personality — so Claude already knows everything it needs before you type a single word.

What Are Claude Projects?

Claude Projects is a feature on claude.ai (available on Pro, Team, and Enterprise plans) that lets you create dedicated workspaces for specific tasks or domains. Think of a Project as a persistent environment where Claude already has access to your reference materials, understands your rules and preferences, and behaves according to your custom instructions — all before you even start typing. Instead of re-uploading documents and re-explaining context in every new conversation, you set it up once in a Project and every chat within that Project inherits all of it automatically.

This is a game-changer for anyone who uses Claude repeatedly for similar tasks. Whether you are a marketer who always needs on-brand content, a developer who needs code reviewed against specific style guidelines, or a customer support manager building an internal knowledge assistant — Projects eliminate the repetitive setup that wastes time and tokens in every conversation. It is the difference between hiring a new intern every day versus working with a seasoned colleague who already knows your systems, documents, and preferences.

Creating a Project — Step by Step

Setting up a Claude Project takes about two minutes. Here is the process from start to finish:

1

Open the Projects Panel

On claude.ai, click the Projects icon in the left sidebar. You will see a list of your existing projects (if any) and a button to create a new one. Click “Create Project” to start.

2

Name and Describe Your Project

Give your project a clear, descriptive name like “Customer Support Assistant” or “Blog Content Writer.” Add an optional description so you (and team members on Team/Enterprise plans) know what this project is for at a glance.

3

Upload Knowledge Base Documents

This is where the magic happens. Click “Add content” and upload your reference documents — PDFs, text files, code files, spreadsheets, or any other document Claude should know about. You can upload up to approximately 200,000 tokens of content (roughly 150,000 words or 500 pages). Claude will read and reference these documents in every conversation within the project.

4

Write Custom Instructions

The “Custom instructions” field is your system prompt. Everything you write here is prepended to every conversation in the project. This is where you define Claude’s role, rules, tone, and behavior. We will cover how to write great custom instructions in detail below.

Custom Instructions as System Prompts

Custom instructions are the most powerful part of a Claude Project. They act as a system prompt — instructions that Claude reads before every conversation in the project, shaping its behavior, tone, knowledge boundaries, and output format. Unlike regular conversation messages that get lost when you start a new chat, custom instructions persist across every single conversation within the project.

Think of custom instructions as programming Claude’s personality and behavior for a specific role. You are not just telling Claude what to do — you are telling it who to be. This is fundamentally different from giving instructions in a regular chat, because the instructions are always active and always take priority.

💡 Key Insight
Custom instructions are read by Claude before your first message in every chat. This means Claude already knows its role, your rules, and your preferences before you ask anything. It is the closest thing to having a trained, briefed employee who never forgets their training.

How to Write Great Custom Instructions

The structure of your custom instructions determines how well Claude performs in the project. Here is a proven template that covers all the essential elements:

## CUSTOM INSTRUCTIONS TEMPLATE
ROLE: You are [specific role]. You work for [company].

RULES:
- Always respond in [language/tone]
- Use the uploaded documents as your primary source
- If unsure, say so — never make up information
- Keep responses under [N] words unless asked otherwise

TONE: [Professional / Casual / Friendly / Technical]

AVOID:
- Do not mention competitor products
- Do not give legal or medical advice
- Do not use emojis in formal communications

FORMAT: Default to [bullet points / paragraphs / structured].

Real-World Project Use Cases

Here are three practical examples of Claude Projects that professionals use every day. Each demonstrates how the combination of uploaded documents and custom instructions creates a specialized AI assistant that would be impossible to replicate in a regular chat.

Customer Support Agent
Knowledge: Product docs, FAQ, policies

Upload your product documentation, FAQ database, return policies, and troubleshooting guides. Set instructions: “You are a customer support specialist for [Company]. Answer questions using only the uploaded documents. If the answer is not in the docs, say you will escalate to a human agent. Tone: friendly, patient, professional.” Every chat becomes an instant support session with zero setup time.

Content Writer
Knowledge: Brand guide, sample posts, style rules

Upload your brand guidelines, tone-of-voice document, and 10–20 sample blog posts or social media posts that represent your ideal style. Set instructions: “You are a content writer for [Brand]. Follow the brand voice in the uploaded guidelines exactly. Match the style of the sample posts. All content should be SEO-optimized and include relevant keywords.” Claude will produce on-brand content from the first message.

Code Reviewer
Knowledge: Style guide, architecture docs, patterns

Upload your team’s coding style guide, architecture decision records, and examples of approved code patterns. Set instructions: “You are a senior code reviewer. Review all code against the uploaded style guide. Check for: naming conventions, error handling, test coverage, and security issues. Output format: numbered list of issues with severity (critical/warning/info).” Paste any code snippet and get an instant, standards-compliant review.

The Knowledge Base — What to Upload and How

The knowledge base is where you give Claude domain-specific information that it would not otherwise have. This can include proprietary company data, internal processes, product specifications, research papers, historical records, or any other reference material. Here are practical guidelines for building an effective knowledge base:

Supported Formats
PDF, TXT, CSV, MD, code files (.py, .js, .ts, etc.), and more. Claude reads the text content of these files. For best results, ensure your documents are text-based (not scanned images) and well-structured with clear headings and sections.
Size Limit: ~200K Tokens
You can upload approximately 200,000 tokens of content (roughly 150,000 words). This is enough for most use cases — an entire product documentation site, a comprehensive style guide, or dozens of reference articles. If you exceed the limit, prioritize the most essential documents.
Organization Matters
Name your files descriptively (“product-catalog-2025.pdf” instead of “doc1.pdf”). Claude uses file names as context clues. Group related documents logically. If you have a very large document, consider splitting it into topic-based sections for better retrieval.

Memory — Claude Remembers Your Preferences

In addition to Projects, Claude offers a Memory feature (when enabled in your settings) that allows it to remember key facts and preferences across conversations. Unlike Projects, which are workspace-specific, Memory applies globally to all your chats with Claude. When you tell Claude “I prefer bullet points over paragraphs” or “my company name is Acme Corp,” it can store that preference and apply it in future conversations without you repeating it.

Projects (Workspace-Specific)
  • Custom instructions per project
  • Uploaded documents as knowledge base
  • All chats in the project inherit settings
  • Ideal for role-specific AI assistants
  • Available on Pro, Team, Enterprise
Memory (Global Preferences)
  • Remembers facts and preferences
  • Applies across all conversations
  • User controls what is stored
  • Can be viewed and edited in settings
  • Complements Projects, does not replace them
✅ Pro Tip: Combine Projects and Memory
Use Memory for global preferences (“always respond in Czech,” “I prefer concise answers”) and Projects for domain-specific setups (uploaded documents plus role-based instructions). Together, they create a highly personalized AI experience where Claude knows both who you are and what you are working on.
📚 Lesson Summary
  • Claude Projects create persistent workspaces with uploaded documents and custom instructions
  • Upload up to ~200K tokens of reference material (PDFs, text files, code, spreadsheets)
  • Custom instructions act as a system prompt — they shape every conversation in the project
  • Structure instructions with: Role, Rules, Tone, Avoid, and Format sections
  • Use cases: Customer Support Agent, Content Writer, Code Reviewer, and many more
  • Memory complements Projects by storing global preferences across all conversations
  • Combine Projects (domain-specific) with Memory (global) for the best AI experience

Praktický průvodce Claude Projekty — jak vytvořit trvalé pracovní prostory s vlastní znalostní bází a instrukcí na úrovni systému, které formují každou konverzaci.

Rating
0 0

There are no comments for now.

to be the first to leave a comment.