ai‑digest

What it does

• A command‑line tool (and Node library) that walks through a project’s files, filters them with ignore/minify rules, optionally strips whitespace, and writes everything into a single Markdown document (codebase.md).
• The resulting Markdown can be uploaded to Claude Projects or a custom ChatGPT so the model can “see” the whole codebase for analysis, debugging, or review.

How you use it

1. Install/run the CLI with npx ai-digest (or npm i ai-digest and call it from a script).
2. Choose options such as:
   • -i / --input to point at one or more directories.
   • -o to name the output file.
   • --whitespace-removal to shrink the text size.
   • --ignore-file / --minify-file to supply .aidigestignore and .aidigestminify pattern files.
   • --watch to keep the markdown up‑to‑date while you edit.
   • --show-output-files to see a bar‑chart of each file’s size.
3. The tool creates a Markdown file that contains each source file wrapped in a code block, with placeholders for files that are minified (i.e., listed but not fully included).
4. Upload that Markdown to Claude or a custom GPT’s knowledge base, then start a chat.

Key features

• Default ignore patterns (build artefacts, config files) plus custom .aidigestignore support.
• Minify patterns (.aidigestminify) that replace large files with a short placeholder; you can customise the placeholder via a callback when using the library.
• Whitespace removal (disabled for whitespace‑sensitive languages like Python/YAML) to cut token count.
• File‑size statistics with visual bar charts; can be sorted by size.
• Watch mode – automatically rebuilds the markdown on file changes.
• Library API – all CLI functionality is exposed as async functions (generateDigest, generateDigestFiles, getFileStats, etc.) for programmatic use.

Who it helps

• Developers who want to give an LLM a full view of a codebase without manually zipping or copying files.
• Teams building custom ChatGPT agents or Claude Projects that need a concise, token‑efficient representation of their source.

Typical workflow

# From the root of your repo
npx ai-digest            # creates codebase.md
# In Claude UI → New Project → upload codebase.md
# or in ChatGPT → Custom GPT → upload the same file

Limitations

• The tool only aggregates text files; binary/SVG files are noted but not embedded.
• Whitespace removal is skipped for languages where it would change semantics.
• Token estimates are provided, but actual model limits still apply.

License

• MIT, so you can freely use or modify it.