ctxstuff

Pack codebases into LLM-ready context. Token counting, optimization, cost estimation. Built different.

Why ctxstuff?

When working with LLMs like GPT-5, Claude, or Llama, you need to pack your codebase into a format they can understand. ctxstuff does this perfectly:

• 📦 Smart packing - Respects .gitignore, skips binaries, prioritizes important files
• 🔢 Token counting - Estimate or accurate counts (PRO with tiktoken)
• 💰 Cost estimation - Know API costs before sending (PRO)
• ✂️ Context splitting - Split large codebases into chunks (PRO)
• 🔧 Optimization - Fit context to any token limit (PRO)
• 👁️ Watch mode - Auto-repack on file changes (PRO)

Installation

npm install -g ctxstuff

Quick Start

# Pack current directory
ctxstuff pack .

# Pack and save to file
ctxstuff pack ./my-project -o context.md

# Copy to clipboard
ctxstuff pack . -c

# Count tokens
ctxstuff count ./src

# Compare models
ctxstuff compare ./src

Commands

FREE Commands

| Command | Description | |---------|-------------| | pack [dir] | Pack directory into LLM context | | count [target] | Count tokens in file/directory | | compare [target] | Compare token counts across models | | license | Manage PRO license |

PRO Commands ⚡

| Command | Description | |---------|-------------| | optimize [dir] | Optimize context to fit token limit | | split [dir] | Split context into manageable chunks | | cost [target] | Estimate API costs | | watch [dir] | Watch and repack on changes | | profile [action] | Manage custom model profiles |

Usage Examples

Pack a Project

# Basic pack
ctxstuff pack ./my-project

# Only JavaScript/TypeScript files
ctxstuff pack ./src -e js,ts,jsx,tsx

# Ignore test files
ctxstuff pack . -i test,spec,mock

# Output as XML
ctxstuff pack . -f xml -o context.xml

# Show detailed stats
ctxstuff pack . -s

Count Tokens

# Count tokens in a file
ctxstuff count ./src/index.js

# Count directory with breakdown
ctxstuff count ./src -b

# List all supported models
ctxstuff count --models

Compare Models

# Compare against all models
ctxstuff compare ./src

# Compare specific models
ctxstuff compare ./src -m gpt-5-turbo,claude-5,claude-4.5-haiku

PRO: Optimize Context

# Fit to 50K tokens
ctxstuff optimize ./src --tokens 50000

# Target specific model's context
ctxstuff optimize . --model gpt-5

# Keep comments
ctxstuff optimize . --keep-comments

PRO: Split Large Codebases

# Auto-split by tokens
ctxstuff split ./large-project

# Split by directory
ctxstuff split . --strategy by_directory

# Save chunks to files
ctxstuff split . -o ./chunks

# Get split suggestions
ctxstuff split . --suggest

PRO: Cost Estimation

# Estimate cost for a model
ctxstuff cost ./src --model gpt-5-turbo

# Compare costs across models
ctxstuff cost ./src --compare

# Estimate with expected output
ctxstuff cost ./src --output 2000

PRO: Watch Mode

# Watch and auto-repack
ctxstuff watch ./src -o context.md

# Copy to clipboard on change
ctxstuff watch ./src -c

Output Formats

• markdown (default) - GitHub-flavored markdown with syntax highlighting
• xml - XML with CDATA sections for content
• plain - Simple text format
• json - Structured JSON output

Supported Models

| Model | Context | Input $/1M | Output $/1M | |-------|---------|------------|-------------| | gpt-5 | 256K | $20.00 | $60.00 | | gpt-5-turbo | 128K | $5.00 | $15.00 | | gpt-4.5-turbo | 64K | $1.00 | $3.00 | | o3 | 200K | $10.00 | $40.00 | | claude-5 | 500K | $20.00 | $80.00 | | claude-4.5-opus | 300K | $15.00 | $75.00 | | claude-4.5-sonnet | 300K | $3.00 | $15.00 | | claude-4.5-haiku | 300K | $0.25 | $1.25 | | gemini-2.0-pro | 2M | $2.50 | $10.00 | | gemini-2.0-flash | 1M | $0.10 | $0.40 | | llama-4 | 128K | $0.50 | $1.50 |

FREE vs PRO

| Feature | FREE | PRO | |---------|------|-----| | pack, count, compare | ✓ | ✓ | | Operations per day | 10 | ∞ | | Files per pack | 20 | ∞ | | Max size | 500KB | ∞ | | Token counting | estimate | accurate (tiktoken) | | optimize command | ✗ | ✓ | | split command | ✗ | ✓ | | cost command | ✗ | ✓ | | watch command | ✗ | ✓ | | profile command | ✗ | ✓ | | .ctxignore support | ✗ | ✓ |

Get PRO

$14.99 one-time payment. No subscription.

# Purchase at
https://pnkd.dev/ctxstuff#pro

# Activate
ctxstuff activate CTX-XXXX-XXXX-XXXX-XXXX

Programmatic API

const { pack, count, format, cost } = require('ctxstuff');

// Pack a directory
const result = await pack('./my-project', {
  extensions: ['js', 'ts'],
  ignore: ['test'],
});

// Count tokens
const tokens = count(result.files, 'gpt-5-turbo');
console.log(`Total tokens: ${tokens.totalTokens}`);

// Format output
const markdown = format(result, 'markdown');

// Calculate cost
const pricing = cost(tokens.totalTokens, 1000, 'gpt-5-turbo');
console.log(`Estimated cost: $${pricing.totalCost.toFixed(4)}`);

Configuration

.ctxignore (PRO)

Create a .ctxignore file to exclude paths:

# Comments start with #
*.test.js
*.spec.ts
__mocks__
fixtures/

Custom Model Profiles (PRO)

# Add custom model
ctxstuff profile add --name my-model --context 32000 --input 5.00 --output 15.00

# List profiles
ctxstuff profile list

# Remove profile
ctxstuff profile remove --name my-model

Tips

1. Start with pack - See what your codebase looks like to an LLM
2. Use compare - Find the cheapest model that fits your context
3. Check token counts - Don't exceed model limits
4. Use stats flag - Identify large files that might need trimming

Links

• Homepage: https://pnkd.dev/ctxstuff
• Issues: https://github.com/pnkd-dev/ctxstuff/issues
• PRO: https://pnkd.dev/ctxstuff#pro

More PRO Tools from pnkd.dev

• llmcache PRO - Cache LLM responses, cut costs by 90% ($18.99)
• aiproxy PRO - One API for GPT, Claude, Llama & more ($18.99)

pnkd.dev - built different.

License

MIT © pnkd.dev