EVO CAPABILITIES / Documents, images and layout
Work with the material, not just a summary of it.
Bring selected documents, tables and images into a reviewable workflow that keeps extracted information connected to the original material.
Mac development · See current scopeThe real problem
A useful document answer must often do more than summarise. You may need to compare two specifications, trace a number to a spreadsheet cell or understand which paragraph supports a recommendation. Different formats introduce different risks: an image contains visible text, a spreadsheet may hold a stale formula result, and a PDF's reading order may differ from its visual arrangement.
What chat history leaves you to do
Pasting extracted text into a chat removes much of its location and structure. A paragraph can lose its page, a value its sheet, and a slide its context. At the other extreme, saying an AI has read a file can imply more than happened. Text extraction, viewing an image and understanding a layout are distinct capabilities and need different evidence.
How EVO approaches it
Read the chosen material in its own structure
EVO's development workflow supports selected text, structured data and common document formats. Depending on the format, extracted information retains useful locations such as lines, paragraphs, sheets and cells, slides or PDF pages. This helps turn a broad question into something you can check against a specific part of the original file.
Keep the original as the reference
A source file remains distinct from its extracted view and any proposed output. Supported local text-copy workflows preserve the original and record the edited copy as a new version. Reading a file does not mean its contents were edited, approved or sent elsewhere. You can use the original to resolve a doubtful extraction instead of trusting a flattened summary.
Show the difference between text and visual evidence
For supported PDFs, the Mac workflow can extract page text and create limited rendered previews. Supported images can provide recognised text. Those results help locate information; a preview is not automatically an AI visual review, and recognised characters are not a complete understanding of a chart, photo or page design.
Choose the right check for the question
A spreadsheet question may require checking source cells and whether stored formula results are current. A layout question needs the rendered page rather than only its words. EVO keeps these limits visible so that the next action can be a targeted inspection, a selected comparison or a request for the missing source instead of an invented conclusion.
A concrete example
ILLUSTRATIVE WORKFLOW · NOT A CUSTOMER RESULT
Illustrative workflow: compare a specification with a table
You provide a product specification and a spreadsheet extract, then ask which requirements have matching evidence in the table.
Inspect the selected files and organise the question around named requirements.
Use paragraph and cell locations to make each comparison checkable, marking missing or ambiguous matches.
Return a comparison brief that separates the extracted values, EVO's interpretation and questions that still require the original material.
What you take away
The intended output is a traceable review rather than a general summary: where a requirement appears, which value supports it and where the evidence is incomplete. The comparison can then guide a targeted follow-up instead of another full read-through.
Your choices
- Choose the files and the question; importing a reference file does not automatically include it in every conversation.
- Inspect the original and the extracted view, and decide whether cloud processing may receive selected material.
- Keep proposed edits separate from application and verify the resulting file or page before accepting it.
Current scope
Implemented foundation
- A real Swift run read a selected 123-character reference and correctly returned its release name, review slots and people per slot.
- This is a small-file example. It does not qualify long PDFs, scanned documents, image interpretation or a complete research report.
Next milestones
- Public access, broad visual-document understanding and reliable interpretation of every chart or layout are not released capabilities.
- Office pagination and styling are not rendered by text extraction. Scanned PDFs are not automatically read through image recognition, and spreadsheet formulas are not recalculated by the parser. A preview or extracted text must not be described as a verified visual conclusion.
These are development capabilities, not a public release. See platform availability before requesting a trial.
Platforms and availabilityRelated questions
Does adding a file upload it?
No. Adding a reference file to the Mac workspace does not automatically upload it or make it a shared memory.
Storage, selection for a task and cloud processing are separate choices. A conversation receives supported content you have selected within that task’s scope.
You can keep a project brief locally, then choose an appropriate extract for a specific comparison.
Review selected material before sending and choose the processing mode. Manage reference files separately from memories and conversation records.
Supported formats and extraction limits vary. Removing a local file does not recall a copy already sent elsewhere or erase separate backups.
Can EVO inspect the website people actually see?
The Mac development build can load a rendered page and preserve desktop and mobile-size screenshots alongside page structure and recorded findings.
This captures the result of the page loading, rather than treating its original HTML as proof of its final appearance.
An inspection can show a narrow-screen overflow or a small control with the page, viewport and captured evidence attached.
Choose the page and inspection scope. Cloud review of selected screenshots requires a separate data-sharing choice.
The published example demonstrates browser capture and rule findings. It does not qualify model visual judgment, every interaction, logged-in pages or full accessibility compliance.
Can EVO understand PDFs, tables and images?
EVO can work with supported extracted document text and structured tables. Image content and page layout need a separate visual-capable path.
Selected text can be searched or compared, and supported tables inspected and aggregated. Supported PDFs provide page text and limited rendered previews; supported images can yield recognized text.
A pricing CSV can be compared by column, while a scanned PDF needs readable extraction or visual review before its contents can support a conclusion.
Select the material and inspect extraction limits before relying on an answer or sharing it with cloud processing.
Text extraction is not proof of layout understanding. Complex PDFs, scans and images do not have universal qualified support, and model visual review remains separately limited.
How is reading page text different from inspecting layout?
Text reveals what a page says. A rendered screenshot and element positions reveal how that content appears at a particular screen size.
EVO’s browser inspection keeps those evidence types separate, so a text-only reading cannot be presented as a completed visual review.
A headline can read well in extracted text yet overlap a button on a narrow screen; that requires rendered-page evidence.
Choose desktop or mobile inspection and open the recorded captures beside the findings.
A screenshot covers its recorded state and time. It does not prove every responsive width, interaction, visual interpretation or accessibility behavior is correct.
Try the workflow
Bring a task you want to move forward.
Explore the Mac release, or tell us which workflow you would like to evaluate. An application does not guarantee an invitation.