Choose 2–6 sources, each from a single article to a whole shelf. Add PDFs, videos, websites or pasted text alongside your Docsie content. Narrow the question to the subjects that matter.
Docsie works through the selected content and identifies its subjects. It reconciles different names for the same concept while keeping genuinely different subjects separate.
For each comparable subject, the engine looks for the relevant information in each source. The report distinguishes supported comparisons from information it did not find.
Detailed findings roll up into technical areas and a broader overview. Each level explains what each source says, the key difference and its impact.
Inside the product
These product captures show the engine applied to public engineering documentation, followed by a requirements check and a contract-review factory. Open each screenshot to inspect it.
Public wireless-device documentation for ESP32-C5, ESP8285 and TI CC3235MODS is brought into a shared architecture. Each source stays beside the key difference and its impact.
Open the screenshot to inspect the full detail.
Boot and operating modes, peripherals and packaging become comparable technical areas. Explore the subject that matters without losing the larger context.
Open the screenshot to inspect the full detail.
The ESP8285 and TI wireless report links back to both source collections. Its findings remain connected to the material being reviewed, so the investigation can continue.
Open the screenshot to inspect the full detail.
An engineering requirements check breaks a rubric into subcategories, requirements and individual expectations. Scores, pass marks, critical gaps and source evidence remain connected.
Open the screenshot to inspect the full detail.
A review factory repeats baseline checks for each incoming contract. The matrix shows matching, differing and missing reference sections across the batch.
Open the screenshot to inspect the full detail.
The same factory runs a separate rubric check for each document and collects category scores in one matrix. Open a document to investigate the expectations behind its score.
Open the screenshot to inspect the full detail.
Keep investigating
Go from a difference to the material behind it, then keep working in the same chat.
Inspect the quoted source passages and original articles. Figure descriptions add image context and carry an AI-derived label; requirements checks can examine a figure directly when needed.
Ask whether a source covers a topic using the cached extraction index of the selected content. Follow-up questions can reach beyond the findings already displayed in the report.
Revalidate an individual finding against full article text, extend a saved comparison, save it as a Docsie article, or export the report to XLSX or CSV.
When the question becomes “does every incoming document meet our standard?”, turn the baseline and rubric into a review factory.
Import your approved baseline. Add a rubric if you also need requirements checks, then review the expectations and weights.
Inspect one result before processing the intake. Approval freezes the baseline and rubric; changing either requires a new approved sample.
Each document gets its own baseline and rubric checks, a verdict with reasons, and a chat holding its reports. Use the queue to prioritize review.
Open a document’s chat, investigate the cited findings and ask follow-up questions. Export the matrices to carry the review into your existing process.
Common questions
Q: How is this different from a visual document diff?
A: A visual diff shows changed wording or layout. Docsie compares concepts across differently structured sources, researches their counterparts and builds a hierarchy of findings with evidence. Use a dedicated redline alongside it when you need exact tracked insertions and deletions.
Q: Can one source be a whole knowledge base?
A: Yes. A source can be a Docsie shelf, book or article from a workspace you can read. Compare sources accepts 2–6 inputs; source count is different from the number of articles inside those collections. PDFs are imported, videos become documentation and websites are extracted before comparison.
Q: Can I choose the comparison framework?
A: Let source comparison discover the subjects, narrow its scope to a question, choose an approved reference for a baseline check, or supply a requirements rubric. The rubric can be drafted with AI, edited, or brought in from a spreadsheet, document or pasted list.
Q: How do requirements scores work?
A: The model assesses each expectation and must cite evidence for an expectation marked met. Code calculates the share met, compares it with the pass mark and rolls weighted scores up through the rubric. A requirements run grades one body of content; a factory repeats it for each document.
Q: What does a verified quotation mean?
A: It means the quotation resolves to source text. The model’s interpretation still needs review. Figure descriptions are labelled AI-derived, and extraction problems should be checked against the original source.
Q: How do factories handle review?
A: Approve a sample to freeze the baseline and rubric, then apply the same setup to incoming documents. The factory gathers verdict reasons, baseline results and rubric scores, with an individual chat and reports per document. False flags can occur, so people confirm the findings.
Q: Can I export or reuse a comparison?
A: Reports export to XLSX or CSV and can be saved as Docsie articles. Baseline and rubric factory matrices have CSV exports. Saved comparisons also support follow-up questions, edits and individual finding revalidation.
See the workflow on your documents
Book a DemoDiscover how your sources relate. Check them against a reference or requirements. Make the approved review repeatable when more documents arrive.
Start with a comparison on your own material.