ctx

Lightning-fast context engine for LLMs. Pack your entire codebase into LLM-ready context in milliseconds.

$ ctx pack ./my-project --budget 200k --model claude-sonnet

  ctx pack complete

  Tokens  [████████████████████████████████████████] 99.2%
          198.4k / 200.0k
  Files   265 packed (386 skipped, 651 scanned)
  Model   Claude Sonnet
  Cost    $0.5953 per request
  Time    0.1s
  Output  stdout

Why ctx?

Every developer working with LLMs needs to feed code context into prompts. Existing tools are slow (Node.js/Python overhead), lack budget awareness, and don't prioritize files intelligently.

ctx is a single Rust binary that:

• Packs 651 files / 1.4M tokens in 0.1s (parallel tokenization via rayon)
• Smart budget packing — automatically selects the most important files to fit your token budget
• Multi-model cost estimation — know exactly what each request will cost before sending it
• Zero dependencies — single binary, no runtime required

Install

cargo install --path .

Or build from source:

git clone https://github.com/sebasnallar/ctx.git
cd ctx
cargo build --release
# Binary at ./target/release/ctx

Usage

Pack a directory

# Pack current directory (outputs markdown to stdout)
ctx pack .

# Pack with a token budget
ctx pack ./src --budget 100k

# Pack for a specific model with cost estimation
ctx pack . --model gpt4o --budget 200k

# Output as XML (Repomix-compatible)
ctx pack . --format xml

# Write to file
ctx pack . --output context.md

# Copy to clipboard
ctx pack . --clipboard

# Pipe-friendly (no progress output)
ctx pack . --quiet | pbcopy

Count tokens

See which files are eating your token budget:

$ ctx count ./src --model claude-sonnet

>> Token count for Claude Sonnet (42 files in 0.1s)

  File                          Tokens       %
  src/main.rs                     2669   32.0%
  src/cli.rs                      1355   16.2%
  src/output.rs                   1250   15.0%
  src/packer.rs                    981   11.7%
  ...

  Total: 8.4k total tokens | Est. cost: $0.0251

Pack git changes only

# Pack files changed since main branch
ctx diff

# Pack files changed since a specific branch
ctx diff --base develop

# Pack only staged files
ctx diff --staged

Smart Packing

When you set a --budget, ctx doesn't just grab files randomly. It uses a priority system:

Priority	File Type	Examples
0 (highest)	Project config	Cargo.toml, package.json, pyproject.toml
1	README	README.md
2	Entry points	main.rs, index.ts, app.tsx
3	Types/schemas	types.ts, *.d.ts, *.proto, *.graphql
4-7	Source code	By nesting depth (shallower = higher priority)
8	Config files	*.yaml, *.toml, *.json
9	Tests	Files containing test or spec
10	Docs	*.md, *.txt, *.rst

Output Formats

• Markdown (default) — Code blocks with language detection, file index
• XML — Repomix-compatible format with CDATA sections
• Plain — Simple text with file separators

Configuration

Create a .ctx.toml in your project root:

budget = 100000

include = ["src/**/*.rs", "Cargo.toml"]
exclude = ["**/generated/**", "**/vendor/**"]

Or a global config at ~/.config/ctx/config.toml.

Supported Models

Model	Tokenizer	Cost/1M input tokens
claude-opus	cl100k_base	$15.00
claude-sonnet (default)	cl100k_base	$3.00
claude-haiku	cl100k_base	$1.00
gpt4o	o200k_base	$2.50
gpt4-1	o200k_base	$2.00
gpt4-1-mini	o200k_base	$0.40
gemini-pro	cl100k_base*	$1.25
gemini-flash	cl100k_base*	$0.15
generic	cl100k_base	$3.00

*Gemini uses its own tokenizer; cl100k_base is used as an approximation.

Ignoring Files

ctx respects .gitignore by default. For additional exclusions, create a .ctxignore file:

# .ctxignore
node_modules/
*.generated.ts
dist/
coverage/

License

MIT