Unstructured.io vs MarkItDown for PDF ingestion

Unstructured.io and MarkItDown both help you pull text out of PDFs, but they optimize for different next steps. Unstructured is a library and platform story: partition documents into typed elements you can chunk, filter, and load into retrieval pipelines. MarkItDown (including Toolsy’s browser path) aims at readable Markdown you can edit, commit, or paste into a notes app. Searching “unstructured pdf to markdown” often mixes those intents. This article separates them. Use Unstructured when ingestion is a productized pipeline. Use MarkItDown when you need a clean .md draft without standing up partition infrastructure. For MarkItDown basics, read what is MarkItDown. For local CLI tradeoffs, see Pandoc PDF to Markdown: when to skip the CLI.

What “PDF ingestion” means in practice

Ingestion is not one verb. Some teams mean “get string content into a vector store tonight.” Others mean “I want headings I can edit in Obsidian.” Others mean “CI should reprocess a folder every night with the same schema.”

Unstructured leans into structured elements: titles, narrative text, tables, images as objects you can route. MarkItDown leans into a Markdown document that looks like something a human wrote. Both can feed LLMs. The glue code differs.

If you only need to see whether a PDF even has a text layer, a two-minute browser convert on PDF to Markdown answers that before you wire Python.

What Unstructured.io optimizes for

Unstructured targets developers building ETL for documents. You install or call an API, choose a partition strategy, and receive elements with metadata. That shape helps when chunking rules depend on element type, when you drop headers and footers by category, or when multiple file types must land in one schema.

Ops cost is real: versions, workers, GPU options for some strategies, and monitoring. You pay that cost when the pipeline runs daily and quality regressions matter.

Partition elements versus a single Markdown file

Element lists are great for code. They are awkward for a human who wanted a handout in Git. If your reviewer is a person with a text editor, Markdown wins. If your reviewer is a chunker function, elements win.

Where Unstructured is the wrong default

Skip Unstructured for a one-off PDF on a locked laptop with no Python. Skip it when the team has no one to own dependency updates. Skip it when you only needed to copy a section into Notion once.

What MarkItDown and Toolsy optimize for

MarkItDown is Microsoft’s open approach to turning office-ish files into Markdown. Toolsy exposes that path in the browser via PDF to Markdown and the multi-format convert to Markdown hub. Free accounts get up to three files per day, up to 5 MB each. The upload is used for conversion and not kept after the response.

Output is a .md you can open anywhere. Tables and complex layouts still need a human pass. Scanned image PDFs need OCR thinking; text PDFs convert cleaner.

Browser convert versus installing MarkItDown locally

Local MarkItDown fits scripted folders and offline rules. Browser convert fits spikes and shared machines that block installs. The Pandoc sibling post covers the same one-off versus pipeline fork for a different CLI. Same instinct: install when you repeat; click when you do not.

Sibling comparisons worth reading

Marker vs MarkItDown vs online and MarkItDown CLI vs browser converter go deeper on tooling variants. This post stays on Unstructured versus MarkItDown job fit.

Side-by-side: choose by next consumer

Next consumer	Prefer
Custom chunker / element filters	Unstructured
Human editor / Git notes	MarkItDown / Toolsy
Nightly folder reprocess in CI	Unstructured or local MarkItDown/Pandoc
One PDF today, no install	Toolsy browser
Strict offline confidential batch	Local stack you control

No row crowns a universal winner. Write the next consumer in the ticket before you pick a library.

How to spike PDF quality before you commit to a pipeline

Take three representative PDFs: a clean digital report, a slide export, and a scanned form. Run MarkItDown-style convert. Inspect headings and tables. If the scan is empty garbage, budget OCR. If the digital report looks good, decide whether you need elements or Markdown files as the stored artifact.

Only after the spike should you invest in Unstructured partition configs. Premature pipeline work on bad source PDFs wastes a sprint.

Tables, headers, and footers

Converters guess. Footer page numbers glue into body text. Multi-column layouts shuffle. Unstructured element types can help you drop noise in code. Markdown drafts need manual cleanup. Plan for either cost.

Size, quota, and production scale

Toolsy free caps are intentional. A 200 MB corpus is not a browser job. Move to local MarkItDown, Pandoc, Unstructured, or an API with retention you accept. See prepare documents for RAG for the longer Markdown pipeline.

Privacy and ops honesty for both paths

Browser uploads leave your machine. Local Unstructured or MarkItDown can stay offline if you configure them that way. Hosted Unstructured APIs are another trust decision. Match the path to data class.

Neither tool magically fixes image-only PDFs. Neither replaces a human review for legal or medical text. Product honesty beats benchmark screenshots.

A practical decision rule

If you are writing application code around document elements this quarter, evaluate Unstructured. If you are producing Markdown for people and models with minimal glue, evaluate MarkItDown. If you are unsure, spike in the browser, then graduate the winners to local automation.

Keep Pandoc in the shortlist when you already live in CLI land; the skip-CLI guide explains when not to. Keep the Markdown for AI hub for format habits after convert.

Frequently asked questions

Is Unstructured.io the same as MarkItDown?

No. Unstructured focuses on partitioning documents into structured elements for pipelines. MarkItDown focuses on Markdown output from common file types. Both can support LLM prep. They are not drop-in replacements for each other.

Can Toolsy replace Unstructured for production RAG?

For production RAG with custom chunking and monitoring, many teams still want a library or API they run under their own ops. Toolsy’s browser convert is strong for spikes and modest quotas. Use it to validate PDF quality, then wire Unstructured or local MarkItDown for scale.

Does MarkItDown use Unstructured under the hood on Toolsy?

Toolsy’s convert tools are MarkItDown-backed paths as described on the product side. Do not assume Unstructured runs behind the browser upload. If you need Unstructured specifically, install or call it yourself.

Which is better for scanned PDFs?

Neither magic wand fixes bad scans alone. You need OCR-aware steps and human checks. Test your real scans early. Blurry phone photos fail on every stack.

When should I prefer Pandoc over both?

Prefer Pandoc when your team already scripts Pandoc, needs its writers and filters, or must stay fully local with a known CLI. Read Pandoc PDF to Markdown: when to skip the CLI. Prefer MarkItDown when Markdown-from-Office is the main goal with less Pandoc complexity.

How do I learn MarkItDown basics first?

Start with what is MarkItDown (Microsoft guide), then try PDF to Markdown on a text PDF under 5 MB. Inspect the .md before you automate anything.

What file size works on Toolsy free PDF to Markdown?

Free accounts typically allow up to 5 MB per file and a small daily quota (about three files). Pro raises limits. Huge corpora belong in local or API pipelines.

Should I store Unstructured elements or Markdown in git?

Store what humans must review in Git as Markdown when possible. Store element JSON in object storage or a database when only code consumes it. Mixing both is fine if you document the source of truth.

Do I need both tools in one architecture?

Sometimes. Example: MarkItDown for human-editable canon files, Unstructured for an experimental partition path on the same PDFs. Start with one path. Add the second when a concrete gap appears.

Is this a ranking of which library is “best”?

No. It is a job map. Best means fit for the next consumer, privacy constraints, and who will maintain the pipeline at 2 a.m. Pick that, then pick the tool.

For RAG-oriented Markdown habits after you choose, continue with prepare documents for RAG.