strainer-cli

Offline document summarization and tagging for text and Markdown.

strainer-cli turns .txt and .md files into extractive summaries, keyword tags, and compression metrics — entirely offline, with zero dependencies in fast mode.

---

Quick start

pip install .            # or just run directly — no install required
strainer my_notes.md

Output:

── Summary ─────────────────────────────────────
We observed three colonies near the edge of the
meadow during the late afternoon survey.

── Tags ────────────────────────────────────────
colonies · meadow · honeybee · survey · observed

── Stats ───────────────────────────────────────
21 words → 11 words (52.4%)

No network, no API keys, no accounts.

---

Installation

Requires Python 3.10+.

Fast mode (default)

pip install .

Or run without installing:

python -m strainer my_notes.md

Smart mode (optional)

Install with embedding support for higher-quality sentence ranking:

pip install ".[smart]"

This pulls in sentence-transformers, torch, and numpy. If these aren't available, strainer falls back to fast mode automatically.

---

CLI usage

strainer <file>                    Summarize a file (human-readable output)
strainer -                         Read from stdin
strainer <file> --json             Machine-readable JSON output
strainer <file> --output-md        Obsidian-friendly Markdown
strainer <file> --output-md out.md Write Markdown to a file

Options

Flag	Effect
--json	Output as JSON instead of human-readable text
--output-md [path]	Obsidian-friendly Markdown (to stdout or a file)
-a, --include-anchors	Include source-position evidence for sentences and tags
-h, --help	Show help
--version	Show version

Environment variable

Set STRAINER_INCLUDE_ANCHORS to any non-empty value to always include evidence anchors.

---

Output formats

Human-readable (default)

── Summary ─────────────────────────────────────
Top-ranked sentences from the source.

── Tags ────────────────────────────────────────
keyword · tags · inferred · from · text

── Stats ───────────────────────────────────────
150 words → 32 words (21.3%)

JSON (--json)

{
  "summary": "Top-ranked sentences from the source.",
  "tags": ["keyword", "tags", "inferred", "from", "text"],
  "metrics": {
    "original_words": 150,
    "summary_words": 32,
    "compression": "21.3%"
  }
}

Markdown (--output-md)

Obsidian-compatible frontmatter with summary body — useful for piping into vaults and note systems.

---

How it works

strainer-cli supports two ranking modes. The output is the same for both.

Mode	Dependencies	Ranking	Speed
Fast	None	Keyword-density ranking	Faster, lightweight
Smart	sentence-transformers, torch, numpy	Embedding-based cosine similarity	Slower, higher quality

Before analysis, strainer strips Markdown noise: frontmatter, links, images, formatting artifacts, and reference sections. This keeps summaries focused on content, not syntax.

---

Examples

# Basic summary
strainer my_notes.md

# JSON for scripting
strainer my_notes.md --json

# Pipe from another command
cat article.txt | strainer -

# Write Obsidian-friendly Markdown
strainer my_notes.md --output-md summary.md

# Include evidence anchors (sentence source positions)
strainer my_notes.md --json --include-anchors

Windows drag & drop

Keep Drag_Text_Here.bat in the same folder as the project. Drag any .txt or .md file onto it.

---

When to use strainer-cli vs an LLM API

Choose strainer-cli when data must stay on your machine, you need deterministic outputs, or you're feeding summaries into scripts, search, or local agents.

Choose an LLM API when you need abstractive rewriting, style adaptation, or reasoning over content.

---

Development

pip install ".[dev]"
pytest

---

License

MIT License. See LICENSE.