Does Claude remember previous conversations?

By default, Claude does not retain memory between separate conversation sessions. Each new chat starts fresh. However, Claude Projects solve this for business users: when you create a Project, you can add persistent context (company information, your role, preferences, document libraries) that Claude loads at the start of every Project conversation. This gives Claude the equivalent of institutional memory — it always knows your company, your style guide, your terminology, and your preferences without you re-explaining them.

What's the difference between context window and memory?

The context window is the amount of text Claude can work with in a single conversation — Claude's models support 200,000 tokens, roughly 150,000 words. Within one conversation, Claude has perfect recall of everything in the context. Memory in the persistent sense refers to information that survives across separate conversations. This is provided via Claude Projects (persistent system context) or, for API users, via system prompts and retrieved content (RAG — retrieval augmented generation) that is injected into each request.

How much context can I put in a Claude Project?

Claude Projects support up to 200,000 tokens of project knowledge (documents, instructions, style guides). In practice, this is sufficient for 50-100 pages of company documents, your complete style guide, product documentation, process manuals, and detailed role instructions. For knowledge bases larger than this, API-based RAG implementations retrieve the most relevant documents dynamically and inject them into each request, effectively giving Claude access to unlimited organisational knowledge.

How do I set up Claude to remember my company's style and terminology?

Create a Claude Project and add a 'Company Knowledge' document that covers: your company name, what you do, your target audience, your tone of voice, terminology preferences (words you use, words to avoid), product names and descriptions, key facts and stats, and your typical output formats. Claude will apply this context to every task in the Project. For Claude API deployments, put this information in the system prompt. Our training programme covers building these knowledge documents for every department.

Claude Memory System Guide: How Claude Remembers Context for Business

How Claude's Memory Actually Works

Claude's memory architecture has three distinct layers, and understanding them is essential for designing effective enterprise deployments. The first layer is the context window — everything Claude can see in the current conversation. Claude's models support 200,000 tokens (roughly 150,000 words), which is enormous. Within a single conversation, Claude remembers everything perfectly — every message, every document you've shared, every instruction you've given.

The second layer is Projects (the persistent knowledge layer). When you create a Claude Project, you can add documents, instructions, and context that automatically appears at the start of every conversation within that Project. This is Claude's equivalent of long-term memory — it persists across sessions. A properly configured Project means Claude always knows your company, your role, your preferences, and your common tasks without you repeating yourself.

The third layer, available via the Claude API, is retrieved context (RAG — retrieval-augmented generation). Your systems dynamically identify the most relevant documents from a knowledge base and inject them into each request. This allows Claude to effectively access unlimited organisational knowledge while staying within the context window limit.

Most enterprises underuse layers two and three. Teams that configure them properly report 40-60% productivity gains compared to teams using Claude without persistent context — because every interaction starts from an informed baseline rather than a blank slate.

Want help designing the right Claude context architecture for your organisation? Our readiness assessment maps your knowledge needs and recommends the right memory approach for your team size and use cases. Free, 90 minutes.

Request Free Assessment →

Using Claude Projects for Persistent Business Context

Claude Projects are the primary tool for giving Claude persistent memory in Claude.ai. Every Project can have up to 200,000 tokens of project knowledge — enough for your complete style guide, company overview, product documentation, process manuals, and role-specific instructions. When you start a conversation in a Project, Claude loads all of this context before you type your first message.

What to Put in Your Project Knowledge

The most effective Project knowledge documents cover five areas: company context (who you are, what you do, your key metrics and facts), role context (your specific responsibilities and what "good" looks like for your work), style and voice (how you want Claude to write — tone, formality, vocabulary, structure preferences), terminology (your internal jargon, product names, things to avoid), and standard output formats (the templates and structures Claude should use by default for your most common tasks).

Company Context Document Template for Projects

## Company Overview Company: [Name] Industry: [Industry] What we do: [2-3 sentence description] Our customers: [Who they are, typical size/profile] Key facts to use consistently: [Revenue, employee count, key stats] Competitors: [2-3 main competitors] ## My Role Title: [Your title] Team: [Your team/department] Key responsibilities: [3-5 bullet points] Common tasks Claude helps with: [List your top 5-8 use cases] ## Communication Style Tone: [Formal/Professional/Conversational] Audience: [Who typically reads my outputs] Format preferences: [Short bullets / flowing prose / executive summaries / etc] Language: [British English / American English] Words to avoid: [Any banned terms or jargon] Always include: [Anything to always add — e.g., disclaimers, sign-offs] ## Standard Templates [Paste any standard formats or structures you want Claude to use] ## Key Documents [Paste or link any reference materials — style guide, brand guidelines, product descriptions]

Department-Specific Projects

Best practice in our deployments is to create separate Projects for each major use case or team. A legal team might have separate Projects for contract review, research memos, and client communications — each pre-loaded with the relevant precedents and style guides. A marketing team might have one Project per brand they manage. This keeps each Project focused and ensures Claude always applies the right context for the task at hand.

Free White Paper

The Claude Training Curriculum: Role-by-Role Guide

How to train every department to use Claude effectively — including how to set up Projects, build company knowledge documents, and establish team workflows.

Download Free →

System Prompts: Memory for API Deployments

For teams using the Claude API — building internal tools, chatbots, or automated workflows — the system prompt is the primary memory mechanism. The system prompt appears before every conversation and can contain all the same elements as a Project knowledge document: company context, role instructions, output formats, and constraints.

A well-designed system prompt does three things: it tells Claude what it is (a contract review assistant, a customer support agent, a financial analyst), it gives Claude the knowledge it needs (company-specific context, relevant policies, standard terminology), and it defines how Claude should behave (output format, level of detail, what to escalate). System prompts of 1,000-3,000 tokens are typical for enterprise deployments; some complex use cases benefit from 5,000+ token prompts that pre-load substantial domain knowledge.

RAG: Giving Claude Access to Your Entire Knowledge Base

The context window and Projects solve persistent memory for commonly-used information, but some use cases require access to larger, more dynamic knowledge bases — your full HR policy library, your complete contract database, your product catalogue with thousands of SKUs, your entire email archive. This is where retrieval-augmented generation (RAG) comes in.

A RAG architecture works like this: when a user asks Claude a question, your system first searches a vector database of your documents to find the most relevant passages, then injects those passages into the Claude API request along with the user's question. Claude answers based on the retrieved content. The result: Claude appears to "know" your entire knowledge base, even though it only ever reads the relevant portions in any given request.

Our implementation team designs RAG architectures for HR knowledge bases, legal research tools, sales enablement systems, and customer support agents. A typical deployment combines a vector database (Pinecone, Weaviate, or pgvector) with the Claude API and an MCP server that handles retrieval. See our MCP Servers Enterprise Guide for details on connecting Claude to knowledge bases via MCP.

Claude Memory System Guide: How to Give Claude Lasting Business Context