tpress 0.2.0

Press repositories, code files, and HTML into compact, prompt-ready context bundles.

TextPress (tpress)

Press repositories and code into compact, prompt-ready context bundles for LLMs.

Why tpress?

When working with LLMs, you often need to share code context. Copy-pasting files is tedious, and full repos exceed token limits. tpress solves this by:

• 📦 Bundling entire repos into a single markdown file
• 🎯 Prioritizing code over data files (Python/JS first, JSON last)
• 🔐 Redacting secrets automatically before sharing
• ✂️ Compressing intelligently to fit token budgets

Install

pip install tpress

Quick Start

# Bundle current directory → bundle.md
tpress .

# Bundle with custom output path
tpress . -o context.md

# Compact a single file
tpress README.md -o readme.txt

# Custom token limit
tpress . -t 50000

Usage

Default Command (Auto-detect)

tpress auto-detects whether you're passing a file or directory:

tpress .                    # Directory → repo mode
tpress src/main.py          # File → file mode
tpress /path/to/project     # Directory → repo mode

Repository Bundling

tpress repo . -o bundle.md

Output:

✓ bundle.md · 47k tokens · 19 files

Adaptive Bundling

tpress uses relevance-based adaptive bundling by default:

• File Ranking (0 = most relevant):

  • Rank 0: Core code (.py, .js, .ts, .go, .rs)
  • Rank 1-4: Compiled, shell, SQL, config
  • Rank 5-6: Documentation, HTML/CSS
  • Rank 7-8: Data files (.json, .csv)

• Smart Ordering: Most relevant files appear first in the bundle

• Token Budget: Default 100k limit with intelligent reduction:

  1. Summarize large data files (JSON/CSV → schema only)
  2. Compress lower-priority files (strip docstrings/comments)
  3. Exclude lowest-relevance files if needed

# Verbose output
tpress . -v

# Very verbose (full analysis)
tpress . -vv

# Disable adaptive bundling
tpress repo . --no-adaptive

Filtering Modes

tpress repo .           # Default: code, config, docs
tpress repo . --strict  # Code only
tpress repo . --all     # Everything including data files

Split Large Repos

tpress repo . --split -t 50000
# Creates: bundle_1.md, bundle_2.md, ...

Single File Compaction

tpress main.py                    # Output to stdout
tpress main.py -o compact.py      # Output to file
tpress main.py -p llm-safe        # With secret redaction

HTML Extraction

tpress html page.html -o text.md

Profiles

Profile	Description
lossless	Conservative normalization (whitespace only)
readable	Unwrap paragraphs, light cleanup
llm	Aggressive (dedupe, structure-aware)
llm-safe	Same as llm + automatic secret redaction

Secret Redaction

The llm-safe profile automatically detects and redacts:

• API keys (OpenAI, AWS, Google Cloud, Azure)
• Tokens (GitHub, Slack, JWT)
• Private keys and connection strings
• High-entropy strings

# Force redaction on any profile
tpress . --redact

# Disable redaction
tpress . --no-redact

CLI Reference

tpress [PATH] [OPTIONS]

Arguments:
  PATH          File or directory to process

Options:
  -o, --out PATH          Output file path
  -c, --copy              Copy output to clipboard (no file created)
  -p, --profile TEXT      Compression profile [default: llm]
  -t, --token-limit INT   Soft token limit [default: 100000]
  -v, --verbose           Verbosity: -v steps, -vv detailed
  --redact/--no-redact    Enable/disable secret redaction
  -V, --version           Show version
  --help                  Show help

Subcommands:
  repo     Bundle a repository (full options)
  file     Compact a single file
  html     Extract text from HTML

Clipboard Support

Copy output directly to clipboard instead of writing a file:

# Copy bundle to clipboard
tpress . -c

# Works with all modes
tpress file.py -c
tpress doc.html -c --html

Token Counting

Token counts are displayed after each operation:

✓ bundle.md · 47k tokens · 19 files

Tokens are counted using tiktoken (GPT tokenizer).

Roadmap

• Repository bundling with adaptive token limits
• Relevance-based file ordering
• Secret redaction
• HTML pipeline: extract | clean | outline
• Language-aware strategies (AST-based)
• Custom redaction patterns via config

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.

License

MIT