Release Notes

July 2026

This release focuses exclusively on the PDF Accessibility module, delivering new reporting and search capabilities, analytics filtering by user and group, significant remediation quality improvements, AI cost instrumentation, and a suite of bug fixes affecting document processing, report accuracy, and the viewer experience.

🗂️ PDF Accessibility & Remediation

What's new

Downloadable accessibility detail report — Users can now download a structured PDF accessibility report for any processed file directly from the accessibility results screen. The report includes document metadata (title, file name, language, pages, file size, creation and modification times), an accessibility grade (A–F), aggregate Pass / Warning / Error counts, and a categorized breakdown across Basic Requirements, Metadata and Settings, and Logical Structure checks. The report is branded with the CampusMind logo and downloads as {filename}_accessibility_report.pdf. The download button is only shown for successfully completed files.

Batch search on the All Batches dashboard — A search field is now available above the batch list, allowing users to filter batches by name, uploader, or batch ID without paginating through hundreds of entries. Search is case-insensitive, debounced, and works in combination with existing status filters. Pagination updates to reflect the filtered count, and an empty state is shown when no batches match.

AI token and cost logging per file — Every accessibility AI call — full reconstruction and layout/autotag — now records the model used, input tokens, output tokens, and cost per file. Data is stored per file with tenant, user, batch, and pipeline context, and is queryable for aggregation by user, group, and tenant. This serves as the data foundation for cost reporting and utilization dashboards.

Analytics dashboard — user, group, and date filtering — The accessibility analytics dashboard now supports filtering by individual user or group, so administrators can view accessibility activity and outcomes scoped to a specific person or department rather than across the entire tenant. Date range controls have also been added, with presets for Last Day, Yesterday, and Last 30 Days, plus a custom date range selector for more precise reporting windows.

Document metadata collection before processing — Users can now provide document context before processing begins: document topic (STEM, business, chemistry, medical, etc.), page number presence and location, single- vs. two-page scan format, and whether to ignore headers and footers. This metadata is passed into the AI processing pipeline to improve tagging accuracy and remediation quality.

Bug fixes

Full reconstruction — heading hierarchy preserved — Full Accessibility Reconstruction previously tagged all headings as H1 regardless of the source document's structure. Heading levels (H1–H6) are now preserved from the original document, satisfying WCAG 1.3.1 logical heading order requirements and passing VeraPDF/PAC structure checks.

Full reconstruction — hyperlinks preserved — URLs and email links present in the original PDF were being dropped during full reconstruction, leaving plain text where clickable annotations should be. Reconstructed PDFs now retain all hyperlink annotations with their original destinations, and screen readers announce them correctly as links.

Download report — inaccurate issue counts corrected — The "Problems Resolved" and "Original Issues" counts shown in the downloaded accessibility report were not accurate due to hidden backend elements being included in the totals. Counts now reflect only the relevant visible elements.

Document viewer — file name no longer shows previous file — When navigating from one file to another within the same batch, the document viewer header was still showing the previous file's name until the new file fully loaded. The header now updates immediately to match the file being opened.

Upload page — colour contrast corrected — The upload and batch naming UI contained colour contrast errors flagged by WCAG 2.1 AA compliance tools (WAVE). Contrast has been corrected across text, buttons, labels, and placeholders on the upload page, consistent with the CampusMind design system.

Notes

• The downloadable accessibility report covers single-file download only in this release. Bulk export across multiple files is not included.

• AI token and cost data is available for new processing runs from this release forward. Historical backfill is out of scope.

• Document metadata collection inputs (topic, page numbers, scan format, headers/footers) are passed as processing context; they do not affect how files are stored or displayed in the batch list.

For questions about any of these changes, please reach out to your CampusMind account team.

Key highlights in the document

• Agent KB settings — chunk size, overlap, and top-K now configurable per agent

• Carlson environment upgrades — both Dev and Prod upgraded to latest Azure AI Foundry

• Migration automation — script to recreate agents with KB, tools, and access

• Pipedream admin controls — tenant-level feature flag to disable integrations

BUGS FIXES

KNOWN ISSUES:

1. Modifying LLM settings (such as chunking or KB size) created by another owner not fully reset custom file data. (Targeted for a fix in the upcoming release)

2. Removing web crawlers created by other owners does not properly clear their associated data. (Targeted for a fix in the upcoming release)

May 2026

This release brings meaningful improvements across PDF accessibility, Canvas LMS integration, AI-powered workflows, agent knowledge and connectors, and platform experience — including a new analytics dashboard, department-level security, per-user Canvas authentication, and stronger remediation reliability.

🗂️ PDF Accessibility & Remediation

What's new

Tenant-level analytics dashboard — Administrators can now view accessibility activity and outcomes across the entire tenant, supporting reporting and operational decisions without needing to export raw data.

Per-department security — Accessibility batches and files can be scoped by department or group so users only see and interact with content permitted for their organizational unit.

Manual review filter and rerun — Filter files flagged for manual review and create a new batch to rerun low-scoring or critical items with updated remediation settings.

Document queue visibility — Processing queue depth is now visible in the UI so admins and users can gauge system load and expected wait times.

Name your batch on upload — Users can assign a meaningful batch name at upload time rather than relying on system-generated labels.

Keyboard navigation and shortcuts — Full keyboard navigation, a / command palette, and navigation hotkeys improve WCAG-aligned use of the accessibility module.

Improvements

• Processing columns and color-coded scores — The file list shows how each file was processed and uses color to highlight PDF accessibility scores for quicker triage.

• Accessible queue for authorized users — Users beyond global admins can now see the accessible queue when their permissions allow.

• Analytics dashboard access rules — Corrected layout issues and enforced role-based visibility for analytics data.

• Upload flow resilience — Files remain in the list when navigating back; upload dates, metadata, and file count descriptions display correctly.

• "New files" indication — The upload page again indicates which files are newly added.

• Improved OCR for auto-tagging — More accurate tags on scanned PDFs through improvements to the auto-tag pipeline.

• Canvas course selection for remediation — Users can change the Canvas course when making a course accessible; batch names are now consistent between batch view and the remediation report.

Bug fixes

• Batch sharing — Group descriptions now appear in sharing dialogs; deleting a group updates the UI immediately; batches no longer remain visible after a group is removed.

• Department security UI — Group names display correctly in security dialogs; previously selected groups are no longer cleared when editing access.

• Remediation reports — "Scanned on" shows the correct timestamp; score colors reflect critical issues; "Changes Made" text is no longer truncated.

• Batch status accuracy — Status text and icons align between the accessibility hub and batch detail pages; stuck in-progress batches and failed processing are handled more reliably; original PDF download 404s are resolved.

• Preserve layout and full remediation — Fixes for figure alternate text and title failures in full remediation paths (Montgomery and Adobe checker scenarios).

• Bookmark in preserve layout — Resolved failures when bookmark remediation ran under the preserve layout strategy.

• Batch sharing reliability — Sharing dialogs, group delete behavior, and batch visibility after group removal all corrected.

• Manual review filter — Filters behave consistently across batch states.

• Filter grouping in batch — Filters inside a batch now group files correctly.

• Remediation UI cleanup — Removed confusing steps (duration, inspect PDF tags, proceed to export) where not applicable; cleaned up hand-pointer styling on the remediation complete page.

📚 Canvas & LMS Integration

What's new

Per-user OAuth for Canvas — Each user now connects their own Canvas account via OAuth, replacing the previous shared admin token. This improves security and auditability across all Canvas-linked workflows.

Processing history on course files — The Canvas course files list shows prior processing runs, status, and history so teams can avoid re-processing the same file.

Predictable Canvas batch naming — Canvas batches are now prefixed with Canvas- and include batch numbers, making tracking and support easier.

Bug fixes

• Canvas session and configuration — Fixed logout when opening Canvas pages and errors loading courses after reconfiguring Canvas integration.

• LTI menu — Canvas LTI now shows the updated left navigation menu instead of a legacy layout.

⚙️ AI Workflow Builder

What's new

Scheduler for workflow automation — Workflows can be configured with scheduler-driven user input steps for recurring or timed runs.

File input node — Upload a file within a workflow and pass a file ID to downstream nodes for document processing, OCR, and related operations.

Pipedream inside workflows — Native agent Pipedream actions can now be invoked directly from workflow graphs, not only from standalone agent chat.

Bug fixes

• Pipedream file upload — Resolved failures when uploading files as part of a Pipedream action in a workflow.

• Account reconnect — Reconnecting an integration no longer lists duplicate email addresses.

• Trigger permissions — Creating triggers now correctly respects role permissions.

🤖 Agents & Chat

What's new

Configurable knowledge base settings — Chunk size, chunk overlap, and Top K are now configurable per agent, allowing retrieval to be tuned for different content types and use cases without code changes.

Tenant-level Pipedream controls — Tenant administrators can disable Pipedream actions and triggers organization-wide.

Environment-level connector toggle — Administrators can enable or disable the Pipedream connector per environment (e.g., different settings for non-production vs. production).

Clickable web crawler sources — URLs and documents from the web crawler now appear as clickable references in agent conversations.

Improvements

• Knowledge base settings UX — "Reset to default" applies standard values (800 / 50% / 10); the settings icon matches design; an obsolete field has been removed; updates persist and apply to chat responses correctly.

• Citation rendering across models — File and web sources display correctly with Claude Opus/Sonnet, Grok-3, and other LLM options; web page titles render in chat.

• Cleaner crawled content — Boilerplate content (navigation, footers) is stripped from crawled pages; configurable chunk selection reduces unnecessary token usage.

• Global LMS settings — Canvas and other LMS configuration now lives under a global Settings tab accessible across all modules.

• Large document support — Single documents over 30 MB are now handled without silent failure.

• Runtime file visibility — Files uploaded for an agent session display correctly and are not hidden when tools are enabled.

Bug fixes

• Chunk overlap validation — The UI previously allowed 0% chunk overlap that the API rejected; frontend and backend are now aligned.

• Memory on Top K-only change — Updating only Top K no longer unintentionally clears other knowledge configuration.

• Connector enforcement — When connectors are disabled, users can no longer access Actions and Triggers.

• Pipedream reliability — Core Pipedream actions and triggers restored after a regression.

• Connection status for Actions — Success messages now only appear when the external account is actually connected.

• User Access Finish button — Saving user access from the Agents wizard now completes correctly.

🏗️ Platform & Experience

What's new

Default landing screen — Users now land on the Home screen after login instead of Accessibility, matching the primary navigation experience for most roles.

Improvements

• WorkOS branding — Sign-in now displays the CampusMind logo; invitation emails use the CampusMind product name.

• Home icon in navigation — The sidebar now uses a Home icon rather than a Chat icon for the main entry point.

Bug fixes

• Create Group validation — Minimum character rules are now enforced consistently when creating groups.

Notes

• Knowledge base chunk size and overlap changes clear and rebuild indexed knowledge for the affected agent. Allow time for re-indexing to complete after saving.

• Pipedream tenant and environment controls are tenant-wide. Confirm the desired settings with your administrator before go-live if your organization uses or plans to restrict Pipedream.

• Canvas per-user OAuth replaces the previous shared admin token. Existing Canvas configurations should be reviewed and reconnected by individual users where applicable.

For questions about any of these changes, please reach out to your CampusMind account team.