When working with AI in Eden, you'll encounter "context limits" — the maximum amount of information a model can process in a single conversation. Understanding these limits helps you choose the right model for your task and get better results.

What Is a Context Window?

A context window is the AI model's working memory — the total amount of text, transcripts, documents, and conversation history it can "see" at once. This is measured in tokens, which are roughly ¾ of a word (so 100 tokens ≈ 75 words).

When you connect items to an AI chat on Canvas, everything — your question, the connected content (transcripts, documents, notes), and the AI's response — must fit within the context window.

If you exceed the limit:

• The AI may lose track of earlier content

• Responses may become incomplete or less accurate

• You might need to break your task into smaller pieces

Context Limits by Model

Eden gives you access to models from multiple providers, each with different context capacities. Here's how they compare:

Claude Models (Anthropic)

What 200K tokens looks like:

• ~150,000 words

• ~500 pages of text

• A full novel or several research papers

• Multiple hour-long video transcripts

ChatGPT Models (OpenAI)

What 128K tokens looks like:

• ~96,000 words

• ~300 pages of text

• Several long documents or transcripts

Gemini Models (Google)

What 1M tokens looks like:

• ~750,000 words

• ~1,500 pages of text

• ~50,000 lines of code

• Transcripts from 200+ podcast episodes

• Entire books or document collections

Grok Models (xAI)

What 256K tokens looks like:

• ~192,000 words

• Full legal contracts, technical manuals, or research papers

What 2M tokens looks like:

• ~1.5 million words

• Entire document libraries, massive codebases

Practical Examples by Use Case

Analyzing a Single Video (10-30 minutes)

Transcript size: ~2,000–6,000 tokens

Recommended models: Any model works well

• Claude 4.5 Haiku for quick summaries

• ChatGPT 4o for fast analysis

• Gemini 2.5 Flash for speed

Working with Multiple Videos (2-5 hours total)

Transcript size: ~25,000–60,000 tokens

Recommended models:

• Claude 4.5 Sonnet for detailed analysis

• ChatGPT 5 for comprehensive review

• Gemini 3 Pro for comparing across many sources

Analyzing a Full Course or Podcast Series

Transcript size: 100,000–500,000+ tokens

Recommended models:

• Gemini 3 Pro or Gemini 2.5 Pro (1M context)

• Grok 4 Fast (2M context)

• These models can hold entire course libraries at once

Research Synthesis (Multiple PDFs + Videos + Notes)

Combined size: Varies widely, often 50,000–200,000+ tokens

Recommended models:

• For moderate collections: Claude 4.5 Sonnet or ChatGPT 5

• For large research projects: Gemini 3 Pro or Grok 4

• For massive document sets: Grok 4 Fast

Codebase Analysis

Size: Varies by project

Recommended models:

• Small projects (< 50K tokens): Claude 4.5 Sonnet, ChatGPT 5

• Medium projects (50K–200K tokens): Grok 4, any Claude model

• Large projects (200K+ tokens): Gemini 3 Pro, Grok 4 Fast

Model Tiers in Eden

Eden organizes models into tiers based on capability and cost:

Base Models (1x requests)

Fast and efficient for everyday tasks:

• Gemini 2.5 Flash

• Gemini 2.5 Pro

• Claude 4.5 Haiku

Pro Models (1.5x requests)

More capable for complex work:

• Gemini 3

• ChatGPT 4o

• ChatGPT 5

• ChatGPT 5.1

• Claude Sonnet 4.5

Expert Models (2x requests)

Maximum capability for important tasks:

• Claude Opus 4.5

How Eden Extends Your Context

While every AI model has a fixed context window, Eden uses two techniques to help you work with more content than would otherwise fit: RAG (Retrieval-Augmented Generation) and Smart Chat Summarization.

RAG: Searching Your Workspace

RAG allows AI to pull relevant information from your entire Eden workspace — even content that isn't directly connected to your chat.

How it works:

1. When you ask a question, Eden searches your workspace (or a specific folder you've referenced) for relevant content

2. Only the most relevant snippets are pulled into the AI's context

3. The AI uses these snippets to answer your question accurately

When RAG is used:

• When the AI searches your Eden workspace for information

• When you reference a folder in your chat using @mentions

• When you ask questions that require finding specific information across many files

Why this matters: Instead of loading entire documents into the context window, RAG selectively retrieves just the parts you need. This means you can effectively "search" across gigabytes of content — far more than any model's context window could hold directly.

Example: You have 50 hours of podcast transcripts in a folder. Instead of connecting all of them to a chat (which would exceed most context limits), you can reference the folder and ask "What did the guests say about pricing strategies?" Eden's RAG system finds the relevant moments across all 50 hours and brings just those excerpts into your chat.

Smart Chat Summarization

Long conversations can exceed context limits too. Eden handles this automatically with smart summarization.

How it works:

1. As your conversation grows and approaches the context limit, Eden detects this

2. Earlier parts of the conversation are intelligently summarized

3. The summary preserves key information, decisions, and context

4. Your conversation continues seamlessly with the condensed history

What you'll notice:

• You may occasionally see Claude "organizing its thoughts" during long conversations

• This is the automatic context management at work

• Your full chat history is preserved — you can scroll back to see it

• The AI can still reference earlier content through the summary

Why this matters: Without smart summarization, long conversations would simply stop when you hit the context limit. With it, you can have effectively unlimited conversation length — the AI maintains continuity by working with compressed versions of earlier exchanges.

Important note: Smart summarization requires code execution to be enabled in your settings. In rare cases with very large initial messages, you may still encounter context limits.

Combining Both Techniques

These features work together powerfully:

• RAG lets you access vast amounts of stored content without loading it all at once

• Smart summarization lets conversations continue indefinitely without losing context

This means your practical working capacity in Eden far exceeds the raw context window numbers listed above. A 200K context window becomes a gateway to your entire workspace, not a hard ceiling.

When to Use Direct Connection vs. RAG

Tips for Working Within Context Limits

Choose the Right Model

Match your model to your content size:

• Short conversations → Any model

• Multiple documents → Pro tier models

• Massive research projects → Gemini or Grok with 1M+ context

Be Strategic About What You Connect

On Canvas, connect only the items relevant to your current question. You can always add more later, but including everything at once may not improve results and uses more requests.

Use Multiple Chats for Different Purposes

Instead of one massive AI chat with everything connected, create focused chats:

• One for summarizing

• One for brainstorming

• One for drafting

Break Large Tasks into Steps

For very large projects:

1. Summarize each source separately

2. Combine summaries in a new chat

3. Ask synthesis questions

Watch for Signs You're Hitting Limits

• Responses that ignore earlier context

• Incomplete or truncated answers

• The AI repeating itself or losing track

Quick Reference Table

Common Questions

Which model should I use if I'm not sure? Start with a Pro tier model like Claude 4.5 Sonnet or ChatGPT 5. They handle most tasks well and have generous context limits.

What if my content is too large for any model? For content exceeding even 1M tokens, break it into logical sections, summarize each section, then work with the summaries. Or use Grok 4 Fast with its 2M token window.

Do larger context windows mean better quality? Not necessarily. Larger windows let you include more content, but the quality of reasoning depends on the model's capabilities. A smaller, more capable model might give better analysis than a larger window with less capable reasoning.

How do I know how many tokens my content uses? Eden handles this automatically when you connect items to AI chats. As a rough guide: 1 minute of video transcript ≈ 150 tokens; 1 page of PDF ≈ 500-800 tokens.

Can I see my token usage? Eden tracks your AI requests, which factor in token usage. Check Settings > Billing to see your usage.