Release Notes - Rhizar

Release notes for Rhizar, including new features, improvements, and bug fixes.

June 26, 2026

Connected Tools

Added connected tools so agents can read from and act in external apps from within a chat
HubSpot and Notion are available to connect now, plus no-sign-in Web Search and Google Maps
Connecting an account once works across every agent in the workspace that has the tool enabled
Added a Tools section in account settings to connect, disconnect, reconnect, and check status per provider
Owners and admins choose which toolkits each agent exposes; members see the list read-only
Added a read-only admin catalog showing every toolkit, what it does, and its availability
Added an inline Connect step in chat that continues the action automatically once a needed account is connected
Tool activity now renders as plain-language cards ("Used Notion")

Action Approval

Actions that change data in a connected tool now pause for approval before running
Approval cards show the exact fields the agent will submit, not just its description
Approve or deny in one click, or choose "Other" to redirect the agent with a free-form instruction
Approval stays disabled until an account is connected, then auto-advances
Read-only actions (search, list, fetch) now run automatically; only changes wait for approval
Approvals, connections, and questions from one reply are now grouped into a single card

Clarifying Questions

The agent can now pause to ask a structured single- or multiple-choice question, with an "Other" free-text option

Web Search

The agent can now search the public web for current information beyond the knowledge base
Added a "Web search · N" chip under replies that opens a drawer listing the pages searched and read

Chats

Thinking now renders as separate blocks in the order things happened, each labeled "Thought for Xs"
Multi-step replies now render in correct chronological order instead of collapsing into one block
Actions that change data are no longer silently retried; only transient network errors retry
Shared and read-only chat views now show thinking blocks, tool cards, and questions
The message box is now held while an action or question is pending so nothing sends out of order
Loading states during agent responses have been optimized

Chat Attachments

Increased the Word document attachment limit to 10 mb (15 mb across all attachments)
Oversized files now show a clear dialog naming the file and the limit, at both pick and send time

Knowledge Base

Added a $50-per-day knowledge base processing cap per workspace to prevent runaway upload costs
Hitting the cap now shows a clear dialog and ingestion resumes automatically at midnight UTC

Admin Dashboard

Added a Tool Usage card on the admin and per-agent analytics pages showing how often each toolkit is used
Added admin visibility into the platform tool catalog and Arcade connection-quota status

Agent Settings

Agent instructions now have a character limit with a live counter that warns as you near the cap

May 14, 2026

Chats

Assistant messages that consulted other connected agents now show a "Consulted N" chip beneath the message that opens a drawer listing each consultation
The consultation drawer surfaces each consulted agent's name, the query the orchestrator sent, the answer it returned, the model used, and any sources the consulted agent grounded its answer in
The chip self-hides on messages where no agents were consulted, so it never adds visual noise to a normal reply
The assistant now consults connected specialist agents more readily and can ask several of them in parallel within a single reply, so cross-discipline questions surface the right perspectives without waiting on multiple turns

Sources & Citations

Every source passage now shows an explicit "via [Agent Name]" attribution badge, including main-agent sources — previously only connected-agent sources had the badge, so users had to infer "no badge means main agent"
Source passages in the Sources drawer now render with full markdown formatting (headings, lists, emphasis, line breaks) instead of being flattened to plain text
The "show surrounding context" view now displays an ~800-character window centered on the cited text — long text no longer requires scrolling to find the relevant span
Truncated context shows leading/trailing "…" indicators outside the markdown blocks for a cleaner read

Documents

Inserting a table is now a single click — the toolbar's table button drops a 3×3 grid with a header row in place rather than asking for a separate "New table" step
Deleting a table now opens an in-app confirmation dialog (with explicit messaging about undo) instead of relying on the browser's native confirm prompt
DOCX imports now preserve Word's text highlighting, mapping all 16 Word highlight colors to the nearest of Rhizar's 5 highlight colors so highlighted passages survive the round-trip
Submitting a document to the knowledge base now shows "Added to knowledge base" when it auto-ingests, or "Submitted to KB queue / Awaiting admin review" when admin approval is still needed, so submitters always know which state their document landed in

Knowledge Base

Knowledge base file processing has been rebuilt on a fault-tolerant event-driven pipeline — uploads now recover automatically from transient parser errors and rate limits instead of failing visibly
A background watcher scans every 10 minutes for files that got stuck mid-parse (e.g. a missed parser callback) and re-runs the next step, so files finish processing without anyone having to retry them
Per-agent and global limits on simultaneous parsing keep one large upload batch from blocking smaller uploads in the same workspace
Admins reviewing the KB Submissions section now see Approve and Decline buttons on every pending submission, including their own — the previous "Awaiting admin review" badge on self-submissions has been replaced with the standard action buttons

Chat Attachments

File attachments in chats (PDFs, Word docs, etc) now flow through the same robust extraction pipeline as knowledge base files, with the same automatic recovery from transient parser errors
A background watcher catches attachment extractions that fall behind and re-runs them, so attachments that previously appeared "stuck" in older builds now finish on their own

Admin Dashboard

The Top Expensive Turns drawer now includes an "Open in Langfuse" link that jumps straight to the full trace for that turn — span timing, nested tool calls, and prompt/response payloads are all one click away
The trace link in the drawer automatically points at the correct environment (production, staging, or development) so admins reviewing a turn never have to re-find it manually

AI Behavior

Every step of an assistant reply — knowledge base intent classification, query rewriting, the main response, each connected-agent consultation, and the post-turn chat summary — now reports under a single trace, making it possible to see the full decision path for any reply in one view

Feedback

The in-app feedback modal now works correctly when opened from admin pages (outside the workspace context), populating its workspace and agent dropdowns dynamically instead of appearing empty

May 6, 2026

Agent Settings

Moved show sources toggle to knowledge base tab
Updated document type list to simpler phrasing in knowledge base upload
Knowledge base file count is now the same in list and capacity sections
Removed agent connections AI model selector entirely
Made knowledge base file list scrollable
Opus 4.7 set as default for all chats and agent connections
Made knowledge base submission review history list scrollable
Added ability to delete multiple knowledge base files at once
Members can now see and retry their own pending or failed knowledge base files
Added "Awaiting review" badge on a member's own pending knowledge base submissions
Removed file size hint text from the knowledge base upload dropzone
Approved knowledge base submissions now embed inline images directly in the saved file (instead of falling back to alt text)
Replaced agent-connection count badges with a chain-link "has connections" indicator
Knowledge base file list now loads up to 500 files so nothing is silently hidden from delete or retry actions
Added a search bar to the knowledge base file list so files can be filtered by name
Added a sort dropdown to the knowledge base file list (last updated newest/oldest, name A→Z / Z→A, size largest/smallest)
Increased the per-agent knowledge base document cap from 300 to 500

Documents

Fixed bug where uploading a document from Word caused an error
Added ability to add images to a document
Added underline to the document editor
Increased document file size limit to 100 mb
Added a Table menu in the editor toolbar with shortcuts for adding and removing rows and columns
Images can now be dragged, dropped, or pasted directly into a document
Inline images are now embedded as real images when exporting a document to Word
Word export now uses US Letter page sizing with 1-inch margins
Word import now preserves underline formatting from the source file
Word import now prompts for a title before creating the document (pre-populated with the filename)
Duplicating a document now suggests a unique title (e.g. "Copy of X (1)") to avoid collisions
Duplicating a document now copies its inline images so they remain available in the copy
Updated knowledge base submission status labels on documents to clearer wording with hover tooltips ("Awaiting Review", "KB Approved", "KB Declined")
Document list rows are now real links — supports right-click "Open in new tab" and keyboard tab-to-open
Sources button in the editor toolbar is now a compact icon button (Database icon, matching the Knowledge Base tab) with the count surfaced in a tooltip
Version History sidebar now refreshes automatically after autosave so new draft versions appear without reloading the page
Comments, Version History, and Sources sidebars now use a unified right-rail drawer for consistent close/back behavior on mobile

Account Settings

Removed agent model selector
Sidebar collapsed/expanded state now persists per user and syncs across devices
Sidebar width is now saved per user and synced across devices
Account deletion card now shows a clear error state with a "Try again" button when the precheck fails (previously the card silently appeared blank)
Account deletion button is now always visible — disabled with a hint message ("Transfer ownership of the workspace(s) above before deleting your account") when blocked
General feedback button on the user profile menu now has a proper accessibility label

Admin Dashboard

Added consultation costs to costs tab
Renamed Members tab to Users
Added "Last active" column to the Users tab
Added Workspaces column on the Users tab showing each user's workspace memberships as badges
Added search by name or email on the Users tab
Each admin tab is now a real URL (/admin/users, /admin/workspaces, /admin/agents, /admin/analytics, /admin/costs, /admin/feedback, /admin/legal, /admin/docs) so tabs are bookmarkable, shareable, and the browser back button works
Admin sidebar tab links now support right-click "Open in new tab" and middle-click
Admin sidebar can now be dragged to resize on desktop, with width persisted per user and synced across devices (matches the chat sidebar)
On mobile, the admin sidebar drawer auto-collapses after navigating to a tab
Workspace and agent logos now render in admin tables (signed image URLs)

Costs

Top Expensive Turns rows are now clickable and open a side drawer with the full per-turn breakdown (replaces the standalone drill-down page)
Top Expensive Turns table now paginates 10 rows per page
Top Expensive Turns table hides the Agent column on mobile and the Model + Events columns on tablet, with a fade overlay signaling the horizontal-scroll affordance on small screens
Top Agents by Cost card now paginates 10 rows per page

Analytics

Added a per-user filter on Platform Stats, Platform Costs, Workspace Analytics, and Agent Analytics
The user filter is searchable by name or email and persists in the URL for deep-linking
Active Users KPI is hidden when filtering to a single user, with the layout reflowing to a 2-column grid
Top Agents card title becomes "This user's top agents" when a user filter is active
Top Agents card on Platform Stats and Workspace Analytics now paginates 10 per page
Errors on analytics pages now show in a styled alert box with a Retry button instead of plain text

Feedback

In-drawer status dropdowns no longer accidentally close the drawer when a value is selected (Radix portal click is now treated as inside the drawer)
Status changes made inside a feedback drawer now update the parent table row immediately (optimistic, no refresh needed)
Mobile feedback cards are now keyboard-accessible — focusable buttons that open the drawer on Tab + Enter
Platform feedback drawers (chat preview + general feedback) migrated to the unified right-rail drawer
General feedback modal no longer pre-fills workspace and agent based on the current URL — defaults to blank so the choice is always explicit

Chats

Sidebar is now resizable by dragging its right edge
Sidebar resize handle is now a 12px hit zone centered on the divider for an easier grab
Saving a chat message as a document now validates against the 100 mb size limit before attempting to save
Clearer error message when saving a chat message as a document fails

AI Behavior

Automatic fallback to Sonnet 4.6 when Opus 4.7 is overloaded or rate-limited (transparent mid-conversation)
Questions about what's in the knowledge base ("what files do you have?", "list your KB") are now answered directly from the system prompt without firing retrieval
Centralized effort tier policy: orchestrator chat runs at xhigh, consultation calls and background utilities at low, chat summary at medium — tunable in one place
Consultation prompt and orchestrator system prompt rewritten following Anthropic's prompt-engineering best practices for Opus 4.7 (positive examples, no length scaffolding, no caps-lock urgency)
Chat summary max output bumped from 300 → 500 tokens so longer chats summarize without truncation

April 23, 2026

Chat

Streaming responses can now exceed the previous 32k output cap, up to 64k
Truncated responses show a "Continue" chip to resume without losing context
"Compacting context…" loading status appears when long conversations compact
Workspace switcher refreshes automatically when you return to the tab

Documents

New Document dialog with "Start blank" or "Import from Word" options
Duplicate any document with a custom name (prompts before creating)
3-dot menu on sidebar documents (Duplicate + Delete)
"Sources (N)" chip in the document editor toolbar opens the sources drawer
Citation count badge on document cards
AI-suggested title confirmation before saving a chat response as a new document
Duplicating a document no longer copies comments or sharing permissions (matches Google Docs / Notion)
Added ability to highlight text

Agents

Unified "Agent Settings" label for members and admins
Per-agent "Show sources" toggle — suppresses citations, strips them on save-as-document, hides source UI for non-admins
Required instructions field when creating a new agent
Default agent instructions updated — removed conflicting "Cite sources" sentence now managed by the Show sources toggle
Updated chat message loading icon design

Workspace

Members now see "Workspace Settings" in the workspace switcher (page renders read-only for non-admins)
Centralized role enforcement (requireOrgRole) across all mutation endpoints
Top Agents card on workspace analytics matches the design used on admin
Renamed the default agent to “[Workspace Name] Agent”

Knowledge Base

Fix knowledge base upload failures for .md, .json, .csv, .yaml, and other text-based files that were incorrectly routed to LlamaParse
Stuck uploads now auto-recover in 3 minutes instead of 20 when a webhook is lost, without interrupting legitimately-running parses
Improved parsing tier from cost efficient to agentic for PDFs
Added cancel button for in progress uploads
Keep file name the same when downloading a file in the knowledge base
KB submissions show the latest resubmission time in the subtitle
Cohere reranking improves knowledge base search relevance
KB list now scrolls within its container in agent settings
KB submissions show a skeleton while loading
Document explorer shows a skeleton instead of "loading docs" when empty

Account

Preferred AI model picker in Account → Preferences (Opus 4.7, Opus 4.6, Sonnet 4.6, Haiku)
"About me" text injected into every chat as personal context
Email marketing opt-in at signup with GDPR-compliant consent metadata
Resend contact sync on signup and account deletion

Feedback

Feedback modal with optional Workspace, Agent, and Topic dropdowns

Admin

Top Expensive Turns card on the costs dashboard — 10 costliest user turns in the selected period
Per-turn drill-down page showing every event (consultation, embedding, completion, etc.) in a single turn
Friendly display names for Consultation (Haiku/Sonnet/Opus) and Continue generation operations
Every per-turn cost event now correlated with a turn_id for accurate per-turn cost rollups
Chat summary embeddings, title generation, knowledge base embeddings and reranks, and connected-agent consultations all tagged to the originating turn

Scale

1 million token primary context window (up from 200k)
Up to 20 connected agents per agent (up from 6)

Brand

Browser tab title now reads "Rhizar — Agents for Teams"
Docs site logo links to app.rhizar.com
Renamed “Analyst” to “Agent” throughout the app
Updated URL to app.rhizar.com

Tests

Added codebase wide test coverage (408 tests) with Vitest and PGlite
Test suite grew to 511 tests
Coverage includes admin cost queries, attachment orphan cleanup, document duplication, summary embedding, shared UI components and more

Code quality

Shared TopAgentsCard component replaces three inline copies across admin and workspace analytics
Shared FileDropZone primitive used by knowledge-base upload and DOCX import
Shared HITL prompt component used for duplicate, title confirmation, and continue flows
Consolidated document API error handling (clean 403/404/500 responses)
Using logger instead of console.log throughout application
Completed code review on entire codebase, major refactor, modularity
Fixed all lint and type warnings

Bugs

Streaming responses can now exceed the previous 32k output cap, up to 64k
When deleting a user, it now sets usage events email to NULL
Fixed bug where large PDFs in knowledge base were getting stuck in parsing
Fixed shared document continuous reloading bug
Chat rename now updates the sidebar and navbar instantly without a page reload
Pending file attachments no longer reappear in the input after reloading
Deleting a document you don't own now shows a clear permission toast instead of a 500 error
Deleting the document you're currently viewing now navigates you out to a new chat automatically
Duplicated documents no longer show orphaned yellow highlights from the original's comments
Admin costs dashboard loads reliably
"N sources" footer on agent messages is hidden when an agent has source display turned off
Orchestrator "Starting stream" log entry now shows the user's actual runtime model instead of the env default

March 26, 2026

Bugs

Fixed attachment and knowledge base uploads to correctly process Word documents

Chat

Added smooth typewriter animation for streamed responses
Fixed auto-scroll to resume after scrolling away
Fixed streaming to continue when navigating away and back
Increased response streaming timeout
AI reasoning streams in live as an expandable block, showing how long the model thought
Chat title now visible in the top navigation bar
Word docs and text files attach near-instantly via local extraction instead of LlamaParse
Send a message and navigate anywhere in the app - streaming continues in the background and your chat state is preserved when you return

Knowledge Base

Added support for image uploads with AI-generated text summaries
Added a refresh button for website sources to immediately re-fetch content

Analyst

Increased maximum analyst connections from 3 to 6
Redesigned and refactored analytics page

Admin Dashboard

Redesigned and refactored cost tracking page with full multi-service breakdown
Redesigned and refactored analytics page
Cost tracking now uses the usage event ledger across all services (Anthropic, Cohere, LlamaParse)

Workspace

Redesigned and refactored analytics page

API Limits

Raised rate limits for all APIs (LlamaParse, Supabase, Resend, Cohere, Claude) and added email notifications at 80% usage

Usage Tracking

Added granular cost tracking for cache tokens, Cohere embeddings/reranking, and LlamaParse page credits

Rhizar Docs

Restructured docs site with individual page routes and persistent sidebar
Updated docs.rhizar.com to serve directly instead of redirecting

Release Notes

Added Q1 release notes at docs.rhizar.com/release-notes

Cleanup

Removed unused code: command palette, PDF generation, legacy cost calculator, redundant type definitions
Removed custom_domain column from analysts

Stability

Navigating to a deleted chat, invalid analyst, or unknown workspace now redirects gracefully instead of showing an error page

March 2, 2026

Documents

Added document locking — freeze a document in a read-only state and unlock anytime to resume editing
Added document duplication to create a fresh working copy of any document
Added per-user sharing permissions — share documents with specific workspace members and set view or edit access per user
Added "Shared with me" view to see all documents others have shared with you
Added inline comments — highlight text to leave comments with threaded replies, timestamps, and @mentions
Added comment-only access for users viewing read-only or locked documents
Added resolve and reopen actions on comment threads
Added full-page Document Explorer with search, sorting, and custom folder organization
Added document version history — view all saved changes and roll back to any previous version
Documents now retain .docx formatting when attached to a chat or added to a knowledge base (previously converted to .md)

Knowledge Base

Members can now submit documents to an Analyst's knowledge base for admin review
Admins can approve or decline submissions, and submitters can track status (Pending, Approved, Declined)

Chat

Added 50,000-character limit to user messages
Increased stale attachment timeout from 10 minutes to 60 minutes with a notification toast when an attachment expires

Workspace Settings

Added Analytics tab to Workspace Settings
Added a tab showing all Analysts in the workspace
Added ability for workspace owners to transfer ownership to another member

Admin Dashboard

Added Costs tab with KPI cards, a stacked area chart, and breakdowns by service, operation, organization, and top Analysts across all paid services (Anthropic, Cohere, LlamaParse)
Costs tab is fully responsive across mobile, tablet, and desktop

Sidebar

Chats now take up more space in the sidebar with scrollable history (search bar removed)
Added "Documents" link to open the Document Explorer with recent documents in the sidebar

Legal

Added a cookie policy banner for new users to accept or deny cookie use
Added legal information display in Account Settings
Added link to legal documents in the account menu

Bug Fixes

Fixed "Chat with document" opening a blank new chat instead of the correct document context
Fixed sources deleted in the document editor still appearing in the Sources drawer
Fixed undo in the document editor not restoring deleted citations or showing them correctly in the Sources drawer
Fixed sign-in and sign-out buttons briefly flashing on screen when logging out

February 24, 2026

Analysts

Semantic memory enabled — Analysts now search relevant messages and summaries across all past conversations based on the current query, not just recent chats
Expanded live chat context from 12 to 16 messages, plus a rolling summary of earlier conversation history
Improved auto-summaries to include more detail and content from attachments
Upgraded knowledge base reranking to Cohere v4.0 Pro for significantly improved retrieval accuracy on complex queries
Increased file chunk size for richer context per retrieval, improving response quality on complex documents
Extended prompt cache duration from 5 minutes to 1 hour, reducing processing time for returning users
Enabled adaptive thinking — the model now dynamically applies deeper reasoning for complex analytical queries

Sources & Citations

Improved source grounding — responses now prioritize exact figures, data points, and key language from knowledge base documents instead of paraphrasing
Clicking an inline citation in chat now highlights the exact cited passage in the Sources drawer

Documents

Citations in saved documents are now preserved and clickable, linking back to the Sources drawer
Documents now export as editable Microsoft Word (.docx) files

Bug Fixes

Fixed Analyst logos not displaying correctly

February 22, 2026

Analysts

Increased Analyst instructions limit from 4,000 to 10,000 characters
Renamed "Description" field to "Tagline" to reduce confusion with the Analyst Instructions field
Members can now view a read-only Analyst Info page (includes About, Instructions, Knowledge Base, and Connections tabs)
Knowledge base files can now be clicked to download, and website sources open in a new tab
Added Created, Updated, and Created By fields to the Analyst About tab
Improved attachment upload progress to show real-time percentages
Connected Analysts now display as clickable links in the Connections tab
Uploading a file to a knowledge base now shows a toast notification confirming processing continues in the background

Documents

Added copy-link icon to the Shared Documents navigation bar

Feedback

Users can now attach screenshots when submitting feedback via "Leave Feedback"
Added an editable status field to feedback items in the Admin Dashboard with "updated by" and timestamp

Analytics

Improved Analyst Analytics and Admin Dashboard Analytics pages with more accurate metrics and better visualizations

Admin Dashboard

Added Documentation tab to create, edit, reorder, and publish docs and release notes
Public documentation pages now accessible at docs.rhizar.com without authentication

Legal

Updated Legal page to use sidebar navigation matching the docs page styling

Bug Fixes

Fixed navigating away from a loading chat and returning showing a blank screen instead of loading states and messages
Fixed empty paragraph spacing collapsing on document save
Fixed knowledge base tab not being responsive on mobile devices
Fixed "Add document to knowledge base" not navigating to the knowledge base correctly
Fixed various reported bugs received via user feedback

February 20, 2026

Sources & Citations

Inline citation numbers now appear in chat responses, linking directly to the knowledge base content used
Clicking a citation number opens the Sources drawer to that specific source
Source details now show the exact cited text with expandable surrounding context

Legal

Added Legal tab in Admin Settings to create, edit, reorder, and publish legal pages
Public legal pages accessible at /legal without authentication
Sign-up page now links to published legal pages (replaces placeholder terms dialog)
Timestamp and IP address recorded when a user accepts terms
Welcome email automatically sent on registration with terms acceptance details

Account Management

Users can delete their own account, which cascades to all associated data (chats, documents, attachments)
Admins can delete any user from the platform admin panel with a pre-deletion review
Organization owners can transfer ownership to another member
Account deletion requires resolving ownership of sole-owner organizations first (transfer or confirm deletion)

Registration

Company field is now required during sign-up
Required fields are now clearly labeled with red "(required)" indicators

Bug Fixes

Fixed tooltips being clipped inside overflow-hidden containers
Fixed sign-out flow not handling deleted user sessions correctly

February 19, 2026

Analysts

Updated chat messages to use Claude Sonnet 4.6
Added descriptive loading-state labels in chat messages (e.g., "Analyzing your question," "Searching knowledge base," "Consulting connected analysts")
All connected Analysts are now always consulted in parallel, significantly reducing response latency
Analysts are now aware of all files in their knowledge bases and in connected Analysts' knowledge bases — users can ask to list all files with descriptions or ask about a specific file
Attachment extraction now retries automatically on failure (up to 4 attempts with exponential backoff)
Optimized knowledge base retrieval strategy for faster chat response times
Attachment uploads now have reserved processing slots so they are not blocked by background knowledge base processing

Invites & Emails

Improved email deliverability to reduce invites landing in spam
Email replies now go to info@rhizar.com
Expired invites are now visible to admins with an option to resend
Pending invites now show expiration dates

Documents

Improved reliability of document creation and editing

Bug Fixes

Fixed markdown list indentation in chat messages

January 21, 2026

Chat

Added support for new spreadsheet file types (.xlsm, .xlsb, .xlw, .numbers) as chat attachments and in knowledge bases
Added drag-and-drop file attachment in chat

Knowledge Base

Increased maximum file upload size to 150 MB
Improved file upload reliability — files now enter a queued state and are processed sequentially (files stuck from before this update should be deleted and re-uploaded)
Added ability to add URLs and websites to knowledge bases
Added retry button for failed knowledge base file uploads
Improved error messages for knowledge base upload failures

Connected Analysts

Connect up to 3 Analysts in Analyst Settings
The primary Analyst can consult its connected Analysts during chat
Chat message sources from connected Analysts are clearly labeled

Documentation

Launched documentation site at docs.rhizar.com with a detailed overview of Rhizar and how to use it
Documentation is accessible from the user Account menu

Bug Fixes

Fixed Members tab in Workspace Settings failing to load

January 10, 2026

Fully rebranded from "Helix" to "Rhizar" — the platform URL has moved from analyst.helixbuild.com to analyst.rhizar.com
Landing page login buttons at rhizar.com now link to analyst.rhizar.com
Improved attachment reading for Excel, PDF, Word, and other file types in chat messages
Improved knowledge base source usage in chat message responses
Improved knowledge base search accuracy with upgraded embeddings and reranking
Increased knowledge base capacity to 500 files per Analyst
Faster repeat responses with prompt caching