Support

Context Pack Support

For the capture-to-export walkthrough, start with Context Pack Help. Use Support for errors, limits, or compatibility questions.

For help, email support@tryaiblock.com. Describe the file type and visible error without sending sensitive source documents.

Version scope: This page describes Context Pack 1.3.0.

Supported inputs

Selectable-text PDF, DOCX, Markdown, plain text, the readable current page, the current main-frame selection, and supported direct HTTPS PDF URLs opened in the active tab.

Unsupported inputs

Scanned or image-only PDF, encrypted PDF or DOCX, and legacy .doc are not supported. OCR is not included. Embedded DOCX images are omitted, and complex layouts or merged tables may lose fidelity.

1.2.0: Markdown / Preview

In Context Pack 1.2.0, after Build context pack, use Markdown to edit or Preview to read. Arrow keys switch the tabs. Copy Markdown and Download .md always use the same editable Markdown, not rendered HTML.

HTML appears literally and images appear as alt text. If Preview times out or is too complex, switch back to Markdown; valid output can still be copied/downloaded. Changing sources clears stale output and Preview: build again to refresh both. See the real UI walkthrough.

1.2.0: dashboard and table warnings

Supported visible action labels, disabled/pressed/expanded states and selected non-editable choices are retained as bounded text. Input values, hidden choices and editable regions stay excluded. The control-simplification warning means labels or choice summaries are limited to 200 code points; review them before adding.

Complex tables may use explicit row/cell text and span annotations instead of a Markdown grid. Cell numbers describe source order, not reconstructed columns. No visual-layout guarantee, OCR, crawling, automatic PII redaction or protected-viewer bypass is added. Read the 1.2.0 release notes.

1.3.0: tables, Quick Copy and links

These controls are included in Context Pack 1.3.0. See real screens and steps.

Capture table: choose one visible HTML table, inspect Markdown/Preview, then Copy Markdown or Add one Table source. Edited drafts survive table switching, including blank drafts; Add/Copy still require valid non-empty content. Cancel discards provisional choices without changing your pack.
No readable visible HTML tables: PDF/iframe tables, editable/form tables and virtualized grids are unsupported. Load required rows yourself. More than 20 eligible tables or 500,000 capture code units is rejected, not partially added. Merged grids may repeat content with a warning; complex/headerless layouts use explicit row/cell text.
Quick Copy: page/selection/table/link review supports Markdown/Preview and Copy Markdown without Add/Build. Direct-PDF review is unchanged. At 10 retained sources, review/Copy still work but another Add is refused.
Literal link labels: Preview keeps escaped labels that look like Markdown readable, without extra nested destinations. Code backslashes and ordinary formatting remain intact. Copy URLs ignores destinations mentioned only in escaped labels; raw Markdown exports are unchanged.
Table fidelity: inline code with a backslash immediately before a pipe uses warned row/cell text to keep its exact contents. Images, alt text and image-only links are omitted from table capture, not downloaded. Check Markdown and removal warnings; safe text links remain.
Missing context menu: on a supported page, reload the extension to retry registration. Inspect its service-worker console through chrome://extensions for a fixed registration warning that contains no page data. Popup capture is a separate route.
Extract links: highlight a visible region or clear the selection for page links. Apply keyword, origin and duplicate filters; replacing edited Markdown needs confirmation. Copy URLs reads the current draft; Copy Markdown exports it exactly. Add creates one Links source, not a source per destination.
No supported visible links: clear the selection or try another page; input/textarea selections are excluded, not whole-page fallbacks. More than 1,000 inspected anchors or 500,000 capture units is rejected, not truncated. No linked-page fetching, crawling or Batch URL import is added.
URL parsing/clipboard failure: Copy URLs stays disabled while parsing, with no usable URLs, or after a bounded-worker failure. Edit/retry or use Copy Markdown. If clipboard access fails, switch to Markdown and copy the selected text manually. Windows may adapt clipboard line endings; your Markdown draft is not rewritten.

Capture selected table: highlight any ordinary text in a cell of one main-page HTML table, then right-click the highlight to review that whole outer table without choosing it again. No special word is required; changing a dropdown option is not page-text selection. Discovery keeps its chooser only for multiple tables. Invalid selections are rejected with guidance; hidden/editable content and iframe tables are unsupported. The menu is limited to HTTP/HTTPS document or frame URLs. It does not appear in non-matching frames such as about:srcdoc, and no workspace guidance is triggered there. Received editable/iframe selected-table events show guidance without reading content or replacing pending work. Open an embedded frame's page in its own tab to retry. Use Capture selected text for only the highlighted content, or Capture table for discovery.

No table shortcut, point-without-selection picker, CSV/XLSX input/export, OCR or protected-viewer bypass. Capture stays main-frame, explicit and local with unchanged permissions. Visible content and retained link queries may be private; review before sharing. Read all changes.

Current limits

Markdown/plain text: valid UTF-8, 1.5 MB per file.
PDF/DOCX: best under 10 MB; 20 MB hard limit per file.
PDF: 30 pages per file and 60 PDF pages per pack.
Mixed batch: 50 MB.
DOCX expanded package: 80 MB and 2,000 ZIP entries.
Pack: 10 retained sources and 500,000 UTF-16 code units.
Page capture: before cloning, up to 40,000 main-frame elements, 100,000 total DOM nodes, and 5,000,000 raw UTF-16 code units across character data, attributes, and live form values; normalized Markdown is limited to 500,000 code units.
Selection capture: before cloning, up to 10,000 selected, boundary, or restored-inline ancestor nodes and 500,000 raw UTF-16 code units across selected character data, cloned attributes, and live form values; normalized Markdown is also limited to 500,000 code units.

Token estimates

Per-source estimates cover normalized source content. The total covers the complete generated or edited Markdown. Both use packaged OpenAI o200k_base tokenization and may differ for other models.

Page capture

Chrome grants temporary access to the tab where the extension was invoked. Choose Capture page or Capture selection in the popup; Context Pack opens or focuses the current workspace and stages editable Markdown plus the captured URL for review before anything is added. The first-run guide appears inside that workspace and can be reopened through Help. Wait until the current capture, review, or conversion is finished before invoking it from another tab; the next capture then returns to the same open pack. Open current pack opens a new empty workspace when none is loaded or focuses the current workspace, without capturing. New pack clears the current in-memory session, with confirmation when it contains work. Closing or reloading always discards the editable session; copying or downloading keeps only the exported Markdown. Navigating away or closing the source tab may end access. A page or selection above the complexity or size limits is rejected without adding partial content.

Capture blocked, access expired, or source tab closed?

“Chrome does not allow capture on this page.” Chrome-internal pages such as chrome://version/ cannot be captured. The Chrome Web Store also restricts capture. Use a regular HTTP or HTTPS page; adding more permissions does not make a protected page capturable.
“Access to the source tab expired.” Temporary access can end after navigation to another origin. Return to the desired source tab, open the extension again, and choose Capture page or Capture selection.
“The source tab is no longer available.” Reopen the source page, invoke Context Pack from that tab, and choose a capture action.

Keep the workspace open; do not reload it or choose New pack to fix capture. A rejected capture does not add a partial source: earlier sources and built Markdown stay in that open workspace. A fresh capture from a regular page returns there for review before Add to pack.

1.2.0: Chrome-internal page rejection now shows the correct restricted-page guidance even when Chrome withholds the tab URL, instead of reporting expired access. This fix is included in version 1.2.0.

Page capture options

Right-click and choose Context Pack → Capture this page or Capture selected text. To capture the current page by keyboard, press Alt+Shift+M on Windows/Linux or Option+Shift+M on macOS. Every path focuses the same workspace, stages content for review, and never adds it automatically.
Direct-PDF capture works only for the active top-level HTTPS URL without an embedded username or password and whose path ends in .pdf. The response must be application/pdf, must not redirect, must contain a recognizable PDF signature, and must remain within 20 MB. The one request can include the current site's credentials.
HTTP direct PDF URLs are not fetched. Download the PDF and add it as a local file. Use the same fallback for embedded, custom, blob:, DRM-backed, redirected, non-PDF, or inaccessible viewers.
The Capture this page menu item shows the currently assigned shortcut when one is bound. Inspect or change the assignment at chrome://extensions/shortcuts; an unbound command leaves the menu title plain.

Browser requirement

Chrome 140 or newer is required by the packaged PDF runtime.