Changelog

Changelog v0.1.6

New Features

Text Edits — Add text to any image in three quick steps. Open an image from your Library, drag your mouse over the area where you want text to appear, then describe what it should say and hit send. ZeroTwo renders the text directly into the selected region — no separate editor or mask drawing required.

• Open from Library — Pick any image you've generated, uploaded, or edited to start a text edit.
• Drag-to-select region — Highlight the exact area where text should land so ZeroTwo knows where to place it.
• Send to render — Type the wording you want and click send. Your edited image is ready in seconds.

Changelog v0.1.5

New Features

New Models - Kimi k2.6 and gpt image 2 are now available.

Improvements

Deep Research - Deep research agent now consistently score above 9.0 on the official "DeepResearchEval"

Bug Fixes

Chat Input - All chat input component UI quirks are now resolved and the transitions in state are smoother

Gemini - Gemini provider code refactored to be more reliable

Changelog v1.4

New Features

Buddies — New buddies feature will help reflect on your conversations and agent's research while you chat.

Agent — Alpha release of "agent zero" (official zerotwo agent) release, capable of calling connectors, doing longer running sessions, and spawning subagents.

New Models — New models Qwen 3.6 and 3.5.

Opus Usage Based — Credits can now be purchased by users for Opus Access.

Image Aspect Ratio Controls — Each model's unique image aspect ratios are now accessible via an aspect ratio selector.

Image Tools — New image processing and editing menu with tools grouped by workflow: creative edits, quality enhancement, precise editing, and utilities.

• Relight — Adjust lighting, shadows, and mood on an existing image without redrawing the whole scene.
• Object Insertion — Add new objects into your image so they blend with perspective and lighting.
• Character Swap — Replace or swap people or characters while keeping pose and composition coherent.
• Upscale — Increase resolution and clarity for sharper prints, crops, and zoomed views.
• Background Removal — Cut out the subject or strip the background for composites and transparent exports.
• Inpainting — Paint over a region and regenerate that area to match the surrounding image.
• Magic Erase — Quickly remove small distractions and blemishes with a single pass.
• Object Removal — Remove a chosen object and fill the area so the edit looks natural.
• Image Expansion — Extend the canvas beyond the original frame (outpainting) to widen or reframe the shot.
• Format Conversion — Convert between image formats (for example PNG, JPEG, WebP) for sharing or workflows.

Improvements

Chat Input — Persistent state across threads and more stable animation.

Sidebar — Much smoother sidebar animation and cleaner sidebar organization updates.

Search Mode — Search mode now uses the same loop architecture as the main agent allowing it to more autonomously decide research depth given its findings.

Codeblocks — New design and enhanced ux during streaming.

Files Page — Files page was redesigned and users can now easily access and chat with any previously uploaded files.

Auto Scroll — Platform now autoscrolls on message send.

Reasoning UI Quirks — Resolved quirky ui for different types of reasoning models into single consistent pattern.

New Markdown Table — Markdown table ui has been modified and simplified.

Connectors — New connectors dialog ui and optimized connection performance and UX.

Project Sharing — Projects shared between users are perfectly synced now and data fetching on inital shared projects load has been improved.

Bug Fixes

Image Gen — Fixed aspect ratio requests not being requested.

Web Search Citations — Fixed web search citations occasionally not rendering citation information.

Projects — Resolved project invite acceptance failure due to frontend security constraints.

Changelog v0.1.3

New Features

FIM Autocomplete — Removed history and regular model suggestion and replaced with official Mercury FIM autocomplete.

Document Search — Enhanced document parsing for RAG using Document AI for PDFs, DOCX, PPTX and more file types.

Improvements

Sidebar UI — Simplified sidebar UI and optimized its performance by refactoring.

Bug Fixes

Changelog v0.1.2

New Features

Files Page Displays All Files — The files page has been updated to now show all files stored in the user's file storage space in our database.

Agent Can Now Make Slides — Our agent is now capable of generating slides in html for easy previewing.

New Studio Page — The Images page has been replaced with a new studio page which now allows you to generate images, videos and audio in a single creative studio.

Realtime Model Update — The realtime model has been updated from gpt-realtime to gpt-realtime-1.5.

New Video Gen Models — WAN 2.6 Flash, Grok Imagine Video and Kling VIDEO 3 Standard have been added to video models list.

Podcast and Music Generation — Podcasts and music can now be generated on the audio page.

New Voice UI — Voice mode now has a cleaner, audio reactive design.

Agent Planning — Agent now has a tool for planning and running longer tasks.

New Image Models — grok-imagine-image and grok-imagine-image-pro have been added

Improvements

4o Responses Update — 4o now returns much more thorough responses.

Memory Now Updates in Realtime — New trigger has been added for memory system which updates the model with new memories every time a new thread is created.

Voice Model Updates — Voice model now uses the new gpt realtime 1.5 model from OpenAI.

Model MCP Server Awareness — Servers and tools are now listed in system prompt for reliable server tool calling and awareness.

File Uploads — Models that support certain file types now directly receive those files instead of using our RAG system.

Bug Fixes

Branch in Chat — Branch in chat tool message types fix.

Business Organization Persistence — Resolved business organization loss of persistence on logout.

Seperation of Assets in Business — Files uploaded and assets generated by private user accounts no longer appear in business account studio or files.

Deepseek Model Hallucination Fix — Resolved Deepseek Reasoner having issues calling tools and hallucinating during longer conversations.

Chat Input Connectors Dropdown — Part of the dropdown wasn't selectable due to being covered by a canvas.

Files API Falling Asleep — Fixed an issue with files api for handling attachments falling asleep after any inactivity during user session.

Projects Sharing Invite Link — Fixed an issue with project invite link sometimes showing undefined for link code.

Smooth Scroll — Scroll is smoother now.

Changelog v0.1.1

New Features

Human-in-the-Loop (Ask Human) — The agent can now pause mid-run and ask you a multiple-choice question. Answer directly in the sidebar, and the agent continues with your input. Supports custom text responses and a 2-minute timeout with auto-select.

Skills System — Install skills from the database and load them on demand. Skills provide instructions and optional Python scripts the agent runs in its sandbox. Example: the report-builder skill gathers requirements via Ask Human, then generates PDF, Markdown, or HTML reports.

Slash Commands — Type /p for prompts and /s for skills. Prompts insert template content; skills insert /skill-slug when agent mode is on. Both support search and keyboard navigation.

GitHub Skill Import — One-click import from any public GitHub repo. Recursively fetches SKILL.md plus scripts, references, agents, and other resource files. Re-imports safely update existing skills without duplicates.

Agents Page & Telegram — New /agents route with a dedicated landing page. Link your Telegram account via QR code or deep link, then chat with the Telegram agent in the full web interface. Threads stay in sync between web and Telegram.

Sharing & Thread Management — Share, pin, move to project, and delete threads from the top of the chat page. Share button generates a public link; the More menu gives quick access to thread actions for all conversations.

Improvements

Skill Viewer — Code files in the skill viewer now use ZeroTwo’s Shiki themes (matching chat code blocks) instead of generic CodeMirror. Deferred highlighting prevents UI freezes on large files; file tree rows use consistent 27.5px height.

Skill Resources — Database layer supports upsert-by-slug and full resource replacement. Expanded resource types (script, reference, asset, agent, template, etc.) and null-byte sanitization fix PostgreSQL import errors from GitHub content.

Interrupt UI — Compact question panel in the Agent Activity sidebar with pre-selected recommended option, custom text input, and proper session handling for live and persisted runs.

New Image Row for Web Search — Models can now provide a neat row of images from search results for users

Bug Fixes

Verify Email — Resolved frequent 400s on email verification. Added body validation, structured logging, and frontend guardrails (debounce, cache, double-submit prevention). Error messages now map reason codes for banned, gibberish, and unavailable cases.

Send Magic Link — Fixed 500s and Supabase error handling. Safe logging (no raw emails), correct HTTP status codes (400/429/502/500), redirect URL allowlist validation, and retry policy for transient errors only.

Gemini 3 — Fixed three critical issues: tool call format mismatch crash (handle both OpenAI and Gemini formats), reasoning tokens in follow-up streams (Gemini 3 only supports reasoning in initial request), and missing reasoning deltas (extra_content flag for delta processing).

Rust Processor 502 — Added timeout (90s), retries with exponential backoff, circuit breaker (opens after 3 consecutive 502/503/504), concurrency limit (5), and pending_retry status with cron retry. File processing no longer fails silently on upstream timeouts.

Browser-Use Timeout — WebSocket server now starts automatically with the API; no separate npm run ws needed. Improved "No extension connected" error message and port conflict handling for the browser-use agent.

Telegram Agent Routing — Fixed send-path precedence so explicit agent and agent_type context override UI toggle state. Prevents Agents-page messages from accidentally routing away from Telegram agent mode.

Unnecessary Data Fetching — Persisted initialized flag to localStorage so the app no longer refetches all user data on every page load. Faster navigation and fewer database queries.

Message Stream Completion Jump - Fixed ui bug causing the assistant message to jump a little bit at the end of the stream.

Detailed Changelog (Frontend & Backend)

Complete changelog of all bug fixes, feature updates, improvements, and implementations across ZeroTwo (frontend) and ZeroTwoApi (backend).

February 2026

Backend

February 9, 2026 - Anthropic System Message Fix

• Fixed dual system messages not reaching Anthropic API
• Additional assistant message (dynamicContext) was not being properly passed to Anthropic models
• Merged staticInstructions and dynamicContext into single system message
• Files: lib/ai-providers/anthropic-provider.js

February 9, 2026 - Image Attachment Retrieval Fix

• Fixed images uploaded by users not being retrieved from vector store
• Retrieval system was not running for image attachments
• Added MIME type detection and proper retrieval triggering
• AI can now answer questions about uploaded images

February 9, 2026 - Image Processing Wait Fix

• Fixed images not accessible immediately after upload
• Added wait mechanism for image processing completion before retrieval

February 9, 2026 - MCP Storage Consolidation Fix

• Fixed duplicate configuration entries for MCP token storage
• Consolidated into single configuration entry with proper structure

February 9, 2026 - Realtime Subscription Optimization

• Fixed multiple duplicate threads appearing in frontend
• SSE events and Realtime subscription both adding threads; optimized event handling to prevent duplicates

February 9, 2026 - Search Mode File Reading State Definition

• Fixed file reading not executing when files attached
• Added proper state definition and execution flow

February 9, 2026 - XAI Responses API Fix

• Fixed invalid arguments error for all XAI models using Responses API
• Fixed tool schema conversion for XAI Responses API format

February 9, 2026 - MCP Tool Name Prefix Fix

• Fixed Manus agent unable to execute MCP tools
• Added proper tool name prefix conversion from OpenAI format to MCP format

February 9, 2026 - Gemini Tool Schema Compatibility Filter

• Fixed Gemini 3 Flash models rejecting complex JSON schemas
• Added schema filtering for Gemini compatibility

February 9, 2026 - Extension Tool Filtering

• Only bio_memory and python tools available when from_extension=true
• Improved security and performance for extension requests

Frontend

February 6, 2026 - Chat Input Pivot Animation

• Replaced chat input animation with pivot-style transition
• Textarea swivels up around the plus button (speed dial)
• Badges and buttons fade in smoothly once textarea reaches multiline position
• Uses Framer Motion for smooth animations

January 2026

Backend

Citations Implementation for Thread Messages

• Added citations field to thread_messages table
• Automatic citation storage from RAG operations

Conversation History System

• Complete conversation history system implementation
• Separates memory layer from conversation history layer
• Recent message context + searchable summaries
• Claude-like architecture for better context management

Memory Summary System

• Edge function generate-memory-summary using GPT-5.2
• Database infrastructure, cron job triggering, logging
• Automatic memory summarization and deduplication

Past Chats Tools Implementation

• Claude-style past conversation retrieval tools
• AI can search and retrieve past conversations

Business Workspace Invitation System

• Workspace owners can invite members via email
• Similar experience to ChatGPT Business

Search Mode Conversation History Support

• Conversation history support in search mode agent
• Search mode can leverage full conversation history

Rust Processor Vision Fallback System

• Intelligent fallback for image/video processing
• Automatic fallback to GPT-4.1-nano when Gemini fails

System Prompt Caching Strategy

• System prompts never built during runtime
• Pre-built and cached system prompts
• Eliminated prompt building overhead

Performance Optimization Implementation

• Complete performance optimization system
• Database functions, timing tracking, cache prewarming
• Reduced time-to-first-response significantly

Frontend Cache Prewarm Integration

• Cache prewarming for frontend; eliminates cold starts

Timing Tracker Integration

• Request timing tracker for bottleneck identification
• Tracks all operations in request flow

Missing Database Functions

• Batch fetch functions: get_image_tool_calls_batch, get_thread_messages_batch
• Eliminated N+1 query patterns

DCR OAuth System Rewrite

• Complete DCR OAuth system using official @modelcontextprotocol/sdk OAuth functions

DCR Token Refresh Implementation

• Critical OAuth token refresh failure fixed
• Proper token refresh implementation; seamless token refresh

DCR OAuth 401 Error Fix

• Fixed 401 Unauthorized on OAuth callback
• Updated middleware to allow OAuth endpoints

Atlassian OAuth Integration

• Complete Atlassian OAuth integration with custom MCP-specific OAuth flow

Airtable OAuth Error Fix

• Fixed Airtable connection failing immediately; proper token validation and error handling

Notion MCP Token Refresh Fix

• Fixed token refresh errors for Notion; added proper token refresh endpoint

OAuth Callback Token Refresh Issue

• Fixed oauth-callback library errors; proper token endpoint configuration

Manus Agent Flow Logging

• Comprehensive flow logging system; complete execution tracking

Manus Agent Simplification

• Removed complex XML-based reasoning; uses native reasoning capabilities

Manus Agent Task Completion Fix

• Fixed agent providing final responses before completing tasks

Manus Agent Reasoning Format Fix

• Fixed model using markdown instead of XML tags; proper XML tag enforcement

Search Mode Iterations Fix

• Fixed too many research iterations (3 instead of max 2)

Planning Tool Evaluation Update

• Plan progress display shows checked-off items when tasks complete

Thesys Provider Tool Call Fix (Critical)

• Tool name not accumulating, causing validation failure
• Return accumulated tool name and arguments; tool calls now validate and execute correctly
• Follow-up stream implementation: proper tool call accumulation and follow-up stream handling

Gemini 3 Thinking Display Fixes

• Thinking events not showing; updated to use response_metadata.thinking for Gemini 3

Gemini 3 Thinking Format Fix

• Proper XML tag enforcement instead of markdown

Gemini 3 Thinking Event Title Update

• Custom titles from <reasoning_title> tags

Kimi K2 Thinking Follow-up Stream Fix

• Fixed duplicate reasoning content in follow-up; proper tool result handling

Kimi Provider Timeout Issue Resolution

• Fixed consistent timeouts in search mode; proper timeout handling and Venice AI routing

GPT-5.2 Video Generation Prompt Fix

• Added video_gen to system prompt; video generation now works

Database Migration Applied

• Custom memory system migration (January 27, 2026) via Supabase MCP

Mem0 to Supabase Migration

• Migrated from paid mem0 to custom Supabase implementation
• AI-powered extraction, semantic search, automatic deduplication
• Eliminated $200–500/month subscription cost

First User Message Fix

• first_user_message only saved for new threads; proper first message tracking

Token Encryption Implementation

• Encryption for security tokens; signed tokens and CSRF tokens encrypted

Security Test Results

• January 15, 2026; all tests passed in production

File Logging Environment Controls

• Debug file logging only in development; no debug files in production

Unused Exports Cleanup System

• Multi-agent system for finding unused exports; parallel child processes

Tool Call Title Improvements

• More contextual tool call titles; shows specific files and data being used

Removed Structured Thought Events

• Now uses native thinking events only; simpler, more reliable

January 30, 2026 - OAuth Credential Mismatch (Notion)

• Fixed Notion token refresh failing with 401
• Atomic credential fallback implementation

January 30, 2026 - Kimi Temperature / Terminal Errors

• Fixed strict temperature parameter requirements; proper integer handling (1 vs 1.0)

January 21, 2026 - DCR Token Refresh Fix

• Critical OAuth token refresh failure; missing tokenUrl parameter
• Proper token refresh implementation for DCR providers

January 11, 2026 - Thesys Tool Call Fixes

• Tool name not accumulating; return accumulated tool name and arguments
• Custom SSE event format; proper accumulation and follow-up streams

Frontend

January 31, 2026 - Retry Functionality Implementation

• Comprehensive retry functionality for AI assistant responses
• Multiple retry options: Try Again, Add Details, More Concise, Search the Web, Switch Model
• Enhanced ActionButtons component with retry menu handlers

January 31, 2026 - Chat Input Container Query Migration

• Migrated chat input width from viewport-based breakpoints to container-query-driven CSS variables
• Matches modern ChatGPT layout pattern; more fluid and responsive

January 30, 2026 - MCP Configured Connectors Frontend Implementation

• Frontend for MCP servers requiring API keys or configuration
• Connectors: Zapier, Make, GitHub, GitLab, Linear, Jira, Notion, Slack, Discord, Gmail, Outlook, Google Drive, Dropbox, OneDrive, AWS S3, Google Cloud Storage, Azure Blob Storage
• Uses PreConfiguredMcpDialog for setup

January 29, 2026 - DCR OAuth Frontend Integration

• Implemented Dynamic Client Registration (DCR) OAuth flow
• Backward compatibility with legacy OAuth
• Providers: Notion, Linear, Asana, Airtable, ClickUp, Stripe, Monday, Sentry, Box, Neon, Intercom, Square, PayPal, HubSpot

January 28, 2026 - Instant Feedback Dialog

• Created instant feedback dialog for feedback, feature requests, bug reports, help requests
• Sends emails to noah@zerotwo.ai and reed@zerotwo.ai

January 7, 2026 - Compare Pages Implementation

• New "Compare" section in footer
• Comparison pages: vs ChatGPT, vs Claude, vs Grok, vs Perplexity, vs Gemini

January 6, 2026 - Search Agent Sources Integration

• Integrated search agent sources into ActionButtons and SourcesAvatars
• Displays source citations from search mode agent responses
• Extracts sources from sources_review step within message's steps array

January 30, 2026 - Fix Unnecessary Data Fetching

• Fixed app fetching all user data on every page load despite realtime subscriptions
• Eliminated unnecessary RPC calls; now uses realtime-first data management

January 30, 2026 - Realtime Data Management System

• Realtime-first data management; data fetched once during login, kept in sync via Supabase realtime

January 29, 2026 - Initialize App Data Optimization

• Refactored initialize_app_data RPC from sequential to parallel
• Split into 7 parallel database functions; ~5–10x speed improvement (519ms → ~50–100ms)

January 27, 2026 - Markdown Worker Integration

• Integrated markdown processor Web Worker into Markdown component
• Improves performance for large messages and streaming content

January 25, 2026 - OAuth Redirect Invitation Parameters Fix

• Fixed workspace invitation parameters lost during OAuth redirect
• Preserves invitation parameters through Google/Microsoft/Apple OAuth

January 24, 2025 - File Processing Status Timing Fix

• Fixed images and videos showing "processed" before upload to vector store
• Corrected backend upload.js and frontend useFileUpload.js status handling

January 23, 2026 - File Upload Spinner Fix

• Fixed image uploads showing "uploading" spinner indefinitely after successful upload
• Separated RAG processing status from upload completion status

January 21, 2026 - Landing Page Lazy Loading Fix

• Fixed landing page loading ALL app components on initial load
• Implemented lazy loading for authenticated-only features

January 21, 2026 - Complete Lazy Loading Optimization

• Comprehensive lazy loading for landing pages and authenticated app sections
• App loads only code needed for current route

January 18, 2026 - Anti-Bot & CAPTCHA Protection

• Anti-bot protection for login: fingerprinting, behavioral analysis, automated tool detection
• Google reCAPTCHA v2 when suspicious; robots.txt protection

January 18, 2026 - Backend CAPTCHA Implementation

• Server-side CAPTCHA verification; POST /api/auth/send-magic-link
• Handles reCAPTCHA v2 and v3

January 17, 2026 - Stripe Card Validation for Trial Subscriptions

• Validates cards before 14-day free trial; handles 3D Secure and SCA

January 15, 2026 - Enterprise SEO Optimization Complete

• 9.7/10 SEO score; SSR for crawlers; @graph structured data; GEO meta tags; FAQ schema

January 15, 2026 - SEO Optimization Completed

• Optimized /solutions, /features, /pricing; SERP descriptions; schema.org

January 15, 2026 - FAQ Implementation

• FAQ sections (3–5 questions) on 10 routes with schema.org

January 15, 2026 - Compare Index Page

• New /compare landing page; comparison cards for 5 major AI platforms

January 15, 2026 - SEO Image Breadcrumb Fixes

• Fixed Google Search Console breadcrumb errors; image alt text and structured data

January 15, 2026 - Sitemap Canonical URL Cleanup

• Fixed duplicate/alias model URLs; accurate lastmod; Google News tags on recent articles only

January 12, 2026 - Stripe Security Fix

• Server-side verification of Stripe session IDs; prevents URL manipulation and fake payments

January 12, 2026 - Edge Function Backend Security Fix

• Scheduled task edge functions: origin validation, signed/JWT tokens

January 12, 2026 - Frontend JWT Authentication Update

• All frontend API calls include Supabase JWT; fetchWithAuth utility; fetchGet, fetchPost, etc.

January 12, 2026 - Kitty Images Optimization

• Kitty images 50MB → 11MB (78% reduction); preloading and caching; service worker

January 25, 2026 - Admin Pages Styling Update

• Admin pages match PersonalizationSettings styling; unified layout and spacing

January 25, 2026 - Admin Component Restructuring

• Restructured admin from Business/ to pages/main/admin/; updated routes

January 26, 2025 - Business Subscription Separation

• Personal and business subscriptions in different storage; business in profiles.settings.business_subscription

January 27, 2026 - CSS and Code Cleanup

• Stylelint; removed unused CSS and code; improved bundle size

January 27, 2026 - Unused Exports Summary

• 401 unused exports identified; opportunity for ~10–15% bundle reduction

December 2025 / 2024

Backend

December 29, 2024 - Reasoning Deltas Fix

• Fixed reasoning models (OpenAI o1/o3/o4-mini, xAI Grok) not emitting thinking deltas
• Proper thinking delta emission

Frontend

December 31, 2025 - Canvas Version Switch Not Updating Fix

• Canvas version dropdown not updating content; added proper remounting/re-rendering

December 30, 2024 - Surgical Updates Final Fix

• Fixed surgical updates for CodeEditor; canvas_surgical_delta handler; removed invalid canvasService.pushToHistory() call

December 30, 2024 - Surgical Updates CodeEditor Fix

• Surgical updates not working; metadata passing and visual feedback for added/removed lines

December 30, 2024 - Old Content Preservation Fix

• Old content disappearing when surgical update starts; preserves content until new content ready

December 30, 2024 - MCP Tool Block Responsive Improvements

• MCP Tool Block fully responsive; text wrapping and JSON preview overflow

December 29, 2024 - Research Message Loader Fix

• Research messages showing loading indefinitely; fixed PendingMessageLoader logic

December 28, 2024 - Onboarding Completion Bug Fix

• onboarding_completed not set in Supabase; fixed infinite redirect; RPC and constraint handling

December 28, 2024 - Onboarding Flow Cleanup

• Standardized onboarding UI; consistent layouts, button styles, positioning

December 28, 2024 - OAuth Login Onboarding Flash Fix

• Brief redirect to onboarding on every OAuth login; ProtectedRoute checks after profile loads

December 28, 2024 - Locale 404 Fix

• Production 404 for locale files; i18n base codes and fallback handling

December 28, 2024 - Sidebar Animation Simplification

• Sidebar animations simplified to color transitions; removed glitchy animations

December 28, 2024 - Deep Research Progress Real-Time Updates Fix

• DeepResearchProgress not showing real-time updates; subscribes to thread store

December 28, 2024 - SimpleToolBox Locale Fallback

• SimpleToolBox showing translation keys; locale fallback handling

December 27, 2025 - Deep Research Manus Format Fix

• Deep Research Progress not opening Split View; updated for manus-research agent format

December 27, 2025 - Deep Research Citation Support

• Citation support in deep research; sources from research events to Markdown

December 27, 2025 - Research Agent Step Content Fix

• Deep research steps not storing content; steps now store rich data from SSE events

December 27, 2025 - Service Worker Cache Variable Duplication Fix

• Duplicate const cache SyntaxError in service worker fixed

December 27, 2024 - Research Report Storage Fix

• Final research report storage conflict; summary separated; report in agent_run.report

December 27, 2024 - Deep Research Split View Refactor

• Reduced complexity; eliminated duplicate data fetching

December 27, 2024 - Agent Run Field Added to Thread Messages

• All get_thread_messages* functions include agent_run field

December 27, 2024 - Supabase Preconfigured MCP Dialog

• Configured MCP connectors use PreConfiguredMcpDialog; Supabase and Hugging Face

December 27, 2024 - Web Workers Performance Optimization

• Web Workers for SSE stream and markdown processing; prevents UI blocking

December 26, 2024 - Plan Created Event Handling

• plan_created event not extracting/storing plan data; updated for lightweight event format

December 26, 2024 - New MCP Servers Added

• Intercom, Neon, Square, PayPal, HubSpot + simple connector; 5 OAuth MCP with DCR

December 26, 2024 - Configured MCP Connector Type Implementation

• configured_mcp connector type; property-based form generation

December 26, 2024 - Agent Run Data Merge Fix

• backend-provider-service no longer replaces agent_run; merges to prevent data loss

December 26, 2024 - Tool Call Metadata Preservation Fix

• Tool call metadata preserved during SSE; from delta and tool execution started events

December 26, 2024 - MCP Tool Metadata Simplification

• Simplified MCP tool call rendering using metadata; reduced pattern matching

December 27, 2025 - UnicornStudio Script Preload Optimization

• Preconnect and preload for landing page animated background

December 27, 2025 - Chat Header Liquid Glass Effect Fix

• Inconsistent blur in chat header; backdrop blur and opacity corrected

December 29, 2025 - Google Site Verification & robots.txt

• Google Site Verification; robots.txt; Search Console; AI crawler support

December 28, 2025 - SEO Crawling Verification Report

• Verified /assets/* and robots.txt; sitemap 1,107 URLs

December 27, 2025 - Mintlify SEO Enhancement

• SEO system per Mintlify; contentStructure, internalLinking, geoOptimization modules

December 30, 2024 - Implementation Status

• Surgical updates for CodeEditor; critical errors fixed; testing completed

Provider Integrations (Backend)

| Provider | Status | Notes | |---------------|---------------------|--------------------------------------------| | Atlassian | Complete | Custom MCP-specific OAuth | | HubSpot | Guide created | Remote MCP at https://mcp.hubspot.com/ | | Notion | Full DCR OAuth | Token refresh, error handling | | Linear | DCR OAuth | Dynamic client registration | | Box | DCR OAuth | — | | Google Drive / Gmail | Non-DCR OAuth | Pre-registered credentials | | Microsoft | Non-DCR OAuth | — | | Airtable | OAuth error fixed | Proper token handling | | Salesforce | JWT issue identified| Documented for future resolution |

Documentation

• Performance Optimization Guide
• DCR OAuth Setup Guide
• Frontend OAuth Implementation Guide
• MCP OAuth Flows Reference
• Atlassian OAuth Quick Start & Scopes
• HubSpot MCP Integration Guide
• Venice AI / Kimi K2 Thinking
• Timing Tracker Integration
• Research Plan Editing (Frontend)
• Search Mode File Reading Implementation
• Manus Flow Logging System
• System Prompt Caching Strategy
• SEO Quick Reference & Enterprise SEO Guide
• Edge Function Logging Debug Guide
• Environment Variables Configuration

Summary Statistics

By Category

• Bug Fixes: 90+ (frontend + backend)
• New Features: 35+
• Performance Optimizations: 20+
• Security Updates: 10+
• UI/UX Improvements: 10+
• SEO Enhancements: 20+
• Infrastructure & Code Quality: 10+

By Layer

• Frontend: 50+ entries (UI, onboarding, realtime, SEO, lazy loading, admin, Stripe, etc.)
• Backend: 50+ entries (OAuth, providers, agents, memory, MCP, database, security, etc.)

Key Achievements

1. Performance: Lazy loading, image optimization, cache prewarming, parallel RPCs, Web Workers, system prompt caching.
2. Security: CAPTCHA, Stripe verification, token encryption, JWT auth, file logging controls.
3. SEO: 9.7/10 on optimized pages, GEO, structured data, sitemap cleanup.
4. User Experience: Onboarding fixes, realtime data, retry UI, DCR OAuth frontend, compare pages.
5. Integrations: DCR OAuth, MCP connectors, provider fixes (Thesys, Gemini, Kimi, XAI Responses API).
6. Code Quality: Unused exports cleanup, CSS/style cleanup, tool call titles, structured thought removal.

Last Updated: February 9, 2026

AI Platform Revolution

Character-Based Assistant Generation

• Complete overhaul of assistant creation system
• Assistants use famous personalities (e.g., Gordon Ramsay for cooking)
• Full AI-powered creation replacing template-based generation
• Enhanced prompts embodying character traits professionally
• Added character_inspiration metadata field

Enterprise & Projects

• Company onboarding flow
• Admin portal for enterprises
• Knowledge graphing for projects
• Google Drive file integration

Advanced AI Capabilities

• Agentic reasoning system
• Include reasoning in response toggle feature
• System hierarchy levels (PLATFORM_SAFETY, META, USER_PROFILE)
• Hosted models for chat completions
• Custom RAG Vision using open vision models

Content Creation & Multimedia Suite

Video & Audio Generation

• Generate videos using Veo or Sora
• Create audio books and transcriptions
• Advanced multimedia processing capabilities

Slides & Presentations

• Slide creation agent workflow (similar to deep research)
• Custom themes and layouts
• Automated presentation generation

Writing & Styles

• Analyze writing styles and create prompt extensions
• Apply custom writing styles to AI responses
• Enhanced content personalization

Image Generation

• Multiple model options (Flux, FAL, etc.)
• Advanced image generation capabilities
• Integrated visual content creation

MCP Integration & Automation

MCP (Model Context Protocol)

• Gmail Read integration
• Notion Read capabilities
• Calendar Read functionality
• Automated MCP task scheduling

Tasks & Automation

• Scheduled tasks for deep research and web search
• MCP-powered automated workflows
• Background task management

Deep Research Enhancement

• Export to Google Docs
• Advanced connectors (Dropbox, Notion, etc.)
• Enhanced research workflows

Code Interpreter

• Python sandboxed code runner
• Secure code execution environment
• Development tool integration

Mobile & Cross-Platform

Mobile Applications

• Launch iOS App for mobile access and chat management
• Launch Android App for mobile access and chat management
• Cross-platform synchronization
• Mobile-optimized user interface

Canvas Collaboration

• Autosave functionality
• Real-time collaborative editing
• Enhanced workspace management

Realtime Features

• Meeting transcription and summarization
• Vision agent capabilities
• Live collaboration tools

Intelligence & Search

Web Search

• Multiple search engine options (Tavily, Brave Search, Exa)
• Stream sources to skeleton
• Advanced search result processing

Templates & Customization

• Model and instructions loadouts
• Custom template system
• Personalized AI configurations

Retrieval & Vision

• Custom PostgreSQL vector for retrieval
• Advanced RAG (Retrieval-Augmented Generation)
• Vision processing capabilities

Production Release

Core Platform Stability

• Initial stable release
• Core application framework stability improvements
• Production-ready API endpoints
• Complete documentation and deployment guides

Quality Assurance

• Automated testing for database tables
• Automated testing for API endpoints
• Development server optimization

Personalization Foundation

• Memory embeddings system
• User preference management
• Basic personalization features

Beta Release

User Experience

• Performance optimizations
• Security audit and improvements
• User interface polish
• Accessibility improvements

Free Version Launch

• Free tier implementation
• Usage limitations and management
• Upgrade path to premium features

Alpha Release

Project Foundation

• Initial development release
• Basic project structure
• Core dependencies setup
• Development tooling configuration