KroWork Changelog

v0.9.10 — Multimodal Agent, Memory, and Workshop Search

Release Date: 2026-06-24

Summary: This release extends the main Agent's multimodal capabilities from text to image understanding and generation, introduces an early memory system, and adds Workshop search. Users can now interact with the Agent using images, let the Agent remember key preferences, and locate target apps in the Workshop with a keyword.

New

1. Multimodal Understanding for the Main Agent: The main Agent now has image understanding capabilities, supporting image-to-text scenarios. Users can submit images directly for the Agent to analyze, interpret, or generate text-based content — no need to describe the image in words first.

2. Multimodal Responses for the Main Agent: The main Agent also supports multimodal output, including image-to-image and text-to-image generation. Users can input images and receive image outputs within the same conversation — visual creation, from request to result, no longer requires leaving the chat.

3. Early Memory: When a user explicitly emphasizes that a piece of information needs to be remembered, or confirms a correction after an Agent mistake, the system automatically generates a memory entry. In subsequent conversations, the Agent naturally applies these preferences and corrections, reducing the cost of repeated explanations.

4. In-thought Tips: Tips are now displayed randomly during the Agent’s reasoning process. Users can use the waiting time to discover more product capabilities and usage guidance, making the experience feel less idle.

5. Kro Workshop Search: The Workshop now supports full-text search. Whether browsing available apps or locating a specific case, a keyword gets you there directly — no more manual scrolling.

6. Background App Import: App imports now run in the background, with a notification upon completion. Users can continue working during the import — no need to stare at a progress bar.

7. App Pioneer Program: The App Pioneer Program is now live, offering review incentives and install-based rewards for high-quality apps. Once an app passes review, it receives a 5000-credit reward; after it is listed, each installation by another user earns an additional 10 credits. This gives creators not only exposure, but also ongoing and visible credit-based returns.

8. Kimi K2.6 Added to the International Version: International users now have a stronger model option to choose from, along with more complete model coverage.

Improvements

Plan Mode Refinement: The Plan Mode workflow has been fine-tuned. The path from reviewing a plan to confirming execution is now shorter, with clearer confirmation checkpoints.

Billing

Upgrade Discounts and Plan Recommendations: Upgrade discounts and personalized plan recommendations are now available. Users can more quickly find the most cost-effective plan that matches their usage.

Notes

New App Case Study: An AI novel-writing case study has been added in this release. It serves as both a capability showcase and a reference template for long-form creative workflows.

v0.8.6 — Plan Mode, Version Management, and Workspace Preview

Release Date: 2026-06-15

Summary: This release focuses on pre-execution control, full lifecycle management of app state, and workspace visibility. Users can review a plan before the Agent acts, save and restore versions at any time, roll back to any historical conversation node, and preview workspace files without leaving KroWork.

New

1. Plan Mode: After enabling Plan Mode from the input bar, the Agent generates a detailed execution plan before writing code or performing tasks. Users review, adjust, and confirm — only then does the Agent proceed. Every step stays within the user's line of sight, eliminating the uncertainty of "what will the Agent do next."

2. Version Management: Save the current app state as a named version at any time. Saved versions can be viewed, loaded, renamed, or deleted. Users can also roll back an app to the state at any specific Agent message — every experiment has a way back, no fear of breaking things beyond repair.

3. Workspace File Preview: Workspace files can now be previewed directly inside KroWork. Generated code, documents, and reports no longer require jumping to an external tool for inspection — interruptions drop dramatically.

4. Error Diagnosis and Reporting: KroWork now automatically diagnoses runtime errors, and users can submit error reports with one click. The team receives more precise diagnostic information, fixes arrive faster, and users don't have to repeatedly attempt reproduction.

Improvements

1. Agent Interaction: The Agent can now ask multiple questions at once, each with selectable options. What used to take several rounds of back-and-forth clarification now wraps up in a single round.

2. Agent Self-awareness: The Agent's understanding of its own capabilities and boundaries has been updated. It proceeds without hesitation when appropriate, requests confirmation without overstepping, and asks for input without waiting for the user to notice a gap.

3. Right-side Task Panel: The right-side panel uses a collapsible section layout, grouped by Control, Sharing, Archive, Task List, and Files. Related information stays together — no jumping between multiple panels.

4. Input Bar: Image upload, file upload, and Instructions are now consolidated under a unified "More" menu, while Plan Mode and permission mode are promoted as primary controls. High-frequency actions move to the front; low-frequency settings don't occupy prime space.

Billing

Annual Plans: Annual subscription plans are now available for both international and domestic versions. Long-term users enjoy more favorable yearly pricing, avoiding the cumulative premium of monthly subscriptions.

v0.7.X — Multimodal Intelligent Apps and a Better Conversation Experience

Release Date: 2026-06-08

Summary: This update centers on multimodal capabilities for Intelligent Apps and the conversation experience. Apps can now understand images, edit images, and handle emails, while the conversation system gains fork, queuing, and refresh capabilities.

New

1. Image Input and Understanding: Intelligent Apps now accept images as input and can understand their content. Users can embed images directly into task descriptions, no need to verbally describe what's on screen first.

2. Image-to-Image Generation and Editing: Intelligent Apps support image-to-image generation and editing. From understanding an image to generating or modifying a new one, apps now have a complete visual processing pipeline.

3. Email Reading: Intelligent Apps can now receive and read emails. Email-driven automation scenarios can now be executed directly within the app, without relying on external tools.

Improvements

1. Copy Conversation Content: Users can now directly copy text from conversations.

2. Add Conversation Text to Context: Select text from a conversation to add it to the context window. Previously clarified information can be reused directly — no need to re-enter it.

3. Message Queuing: Send multiple messages consecutively; the Agent processes them in order without the user having to wait for one to finish before sending the next. The bottleneck of waiting during thinking is gone.

4. One-click Refresh: When the underlying model stalls, a single click restarts the response. No need to resend or rewrite the message.

5. Sharing Experience: Users can now edit app descriptions during the sharing flow.

6. App Detail Page and Sidebar: The app detail page design has been optimized, and shortcut buttons have been added to the sidebar.

v0.6.5 — Kro Workshop, Intelligent Apps, and Credits System Upgrade

Release Date: 2026-05-29

Summary: Kro Workshop officially launches. Intelligent Apps gain image generation, streaming output, and email sending, while the credits system upgrades in parallel. Users can now browse and remix apps in the Workshop and earn credits through daily check-ins and invites.

New

1. Kro Workshop: Kro Workshop is officially live — users can browse, import, and remix apps, with simultaneous Web support. The sidebar entry "Explore" has been renamed to "Kro Workshop." Discovering usable apps is no longer a secondary feature.

2. Image Generation for Intelligent Apps: Intelligent Apps now support image generation. The boundary of creation expands from text into the visual realm.

3. Streaming Text Output: Intelligent Apps now progressively output generated content in a streaming fashion. Users see results advancing in real time, rather than waiting for a large block of text to appear all at once.

4. Email Sending from Generated Apps: Generated apps can now send emails directly. Notification and workflow automation scenarios no longer depend on external messaging tools.

Improvements

1. Deep Research: Deep Research capabilities have been strengthened, with significantly improved depth in multi-step investigation and comprehensive analysis.

2. Agent Prompting: Agent prompts have been fine-tuned, with greater clarity especially around self-awareness and capability boundaries.

3. Visual Design Quality: Forty-two new design skills systematically improve the visual quality of generated apps. First impressions of an app are no longer constrained by default styles.

4. Website Navigation: Navigation paths for pricing, user center, and help documentation have been streamlined.

5. Mobile Website Experience: The mobile website has been rebuilt. Browsing and operating on small screens is no longer cramped.

6. Login Region Selection: The login page now offers manual region selection for domestic and international versions. Users won't be automatically routed to the wrong version.

Billing

1. Daily Check-in Credits: Users earn credits through daily check-ins.

2. Invite Rewards: Inviting new users earns credit rewards.

3. Invoice Support: Invoice requests are now supported.

v0.5.X — Link Sharing, App Marketplace, and Modular Capabilities

Release Date: 2026-05-22 / 2026-05-25

Summary: This phase centers on app sharing, the App Marketplace, and modularization of Kro capabilities. Sharing upgrades from zip files to direct links, apps now have their own marketplace entry, and core capabilities are gradually split into independent modules with usage-based credit consumption.

New

1. App Sharing 2.0: Share apps via links; recipients open the link to install. Sharing an app is as simple as sending a URL.

2. App Marketplace / Explore (Early Version): An early version of the App Marketplace and Explore page is now live. Users no longer need to discover apps from scratch — they can browse and evaluate existing cases directly.

3. Folder-level Context Referencing: Conversations can now reference entire folders, no longer limited to individual files. When dealing with large volumes of files, there's no need to select them one by one.

4. Conversation Fork: Users can create a new conversation copy from any historical node. Exploring alternatives leaves the original conversation untouched.

5. Feature Introduction Pop-up: A feature introduction pop-up helps users become aware of new capabilities.

Improvements

1. Modularized Kro Capabilities: Core capabilities have been split into independent modules, including:

• Web search
• Fetch
• Office operations
• File reading and writing

2. Modularized Intelligent App Capabilities: Intelligent App capabilities are similarly modularized, consuming credits from the user's Kro account based on actual usage. Pay only for what you use — no need to pay for capabilities you don't need.

3. Search Quality: Search has been upgraded to paid search providers, with significantly improved accuracy and content quality. Search results no longer require secondary vetting.

Billing

Subscription Renewal SMS Reminder: An SMS reminder is sent 5 days before subscription expiration. Users no longer face unexpected interruptions or passive renewals.

v0.4.8 — App Sharing 1.0 and Interaction Redesign

Release Date: 2026-05-15

Summary: App Sharing makes its first landing with zip import and export. At the same time, the in-app interaction experience has been fully redesigned, language support expands to Japanese, Spanish, and Traditional Chinese, and feedback channels plus billing infrastructure are rounded out.

New

1. App Sharing 1.0: Apps can now be exported as zip bundles and imported from zip files. App sharing now has a complete import-export loop for the first time.

2. More Interface Languages: New interface languages include:

• Japanese
• Spanish
• Traditional Chinese

3. User Feedback Entry Points: Dedicated feedback channels have been added: Bug reports Suggestions Billing issues Business inquiries

4. Model Labels and Descriptions: Labels and descriptions have been added for each model. Users can make informed model choices instead of guessing blindly.

5. Revenue and User Growth Dashboards: Internal dashboards now cover revenue and user growth metrics.

Improvements

1. In-app Interaction Experience: The in-app interaction model has been fully redesigned, drawing on CodeX and Claude-style workflows. Operations feel more responsive, and flows are more coherent.

2. Prompt Quality: Chinese and English prompts have been optimized, raising the domestic model pass rate from 40%+ to 80%+. For the same task, the Agent's probability of getting it right the first time has doubled.

3. Model Cache Hit Rate: The model cache hit rate has improved to 90%. Responses are faster, and the cost of repeated computation drops accordingly.

Billing

1. Higher-tier Plans and Credit Add-ons: Higher-tier subscription plans and credit add-on packages are now available.

2. First-month Subscription Discounts: New users enjoy a first-month subscription discount.

3. Manual Credit Top-ups: Manual credit top-ups are now supported on the backend.

v0.3.12 — Multi-platform Support, Permission Modes, and Custom Skills

Release Date: 2026-04-30

Summary: KroWork steps beyond macOS, covering Windows and Intel Mac. Permission modes let users control Agent behavior at a granular level, custom Skills open the door to extensibility, and the toolchain plus model-side stability are reinforced in this release.

New

1. Multi-platform Support: Windows support and Apple Intel chip support have been added, along with Chinese and English interfaces and domestic base model integration. KroWork's user base expands from a single platform to a broader range of usage scenarios.

2. Permission Modes: Users can now control the Agent's execution permissions at a granular level. When running sensitive tasks, every step stays within the approval chain — no room for unintended actions.

3. User-defined Skills: Users can extend KroWork's capabilities on their own terms. No need to wait for the official roadmap — users can fill missing capabilities themselves.

4. Instructions: Instructions are now accessible from both the input bar and the settings page. Users can set persistent behavioral guidelines for the Agent, without re-instructing at the start of every conversation.

5. Conversation Status Indicators: Status indicators cover:

• New message
• Pending approval

Users always know which stage the conversation is in — no need to guess whether a silent period is normal.

6. Channel Invite Codes and Short-link Tracking: Invite codes and short-link tracking are now supported.

7. Early Web Features: New additions:

• Contact Us
• Account information viewing

8. File and Image Upload Improvements: Drag-and-drop upload and copy-and-paste upload are now supported. Uploading is no longer limited to clicking a button.

9. Editable Deployed Apps: Deployed apps can still be edited after creation. Making changes no longer requires starting from scratch.

Improvements

1. Tool Capabilities: Read, write, search, and error fallback have all been strengthened. The Agent's fault tolerance and success rate improve noticeably when handling complex file operations.

2. Browser Use: Browser automation capabilities have been upgraded. Task completion rates for web-based operations are higher.

3. Interaction Model: The interaction model has been redesigned with "apps" as the central object. Users' attention stays focused on what they build and deploy, rather than being distracted by the tool's own interface.

4. Long-context Handling: Long-context processing has been optimized. In extended conversations, the Agent no longer loses track of critical earlier information.

5. Prompt-leak Protection: Prompt-leak defenses have been hardened.

6. Model Cache Efficiency and Auto-fallback: Model cache optimization reduces costs by approximately 70%, and automatic fallback is added when a model becomes unavailable. Users spend less and won't notice an interruption when models fail.

Billing

Commercialization Features: New billing infrastructure:

• Payments
• Credits
• Frontend subscription display

v0.2.3 — First Core Release of KroWork

Release Date: 2026-03-31

Summary: The first core release of KroWork makes its debut. The integration of Agent with local computer control, file and Office processing, web search and browser automation, app runtime management, and Deep Research — all foundational capabilities are built in one go, providing a complete skeleton for future evolution.

New

1. Registration and Login: Invite-code registration and email login are supported.

2. Agent and Local Computer Control: The Agent now integrates local computer control capabilities — it can organize the desktop, batch-rename, and batch-move files. Routine local operations can now be delegated to the Agent, sparing users from manual repetition.

3. File-level Operations: Supports:

• Reading
• Rewriting
• Revising

4. Office Document Generation: Supports direct generation of PPT, Doc, and Excel documents. The time spent on formatting and layout between an idea and a finished document is eliminated.

5. AI-generated Apps and Scripts: The Agent can generate software applications and scripts directly from requirements.

6. Persistent App Runtime and Startup Launch: Apps can run persistently and auto-launch at system startup. Created apps no longer need manual starting and monitoring.

7. Deployed App Management: Apps support:

• Start
• Stop
• Add to system apps

8. Intelligent Data Analysis: Intelligent data analysis is available out of the box.

9. Image and Text Input: Screenshots and pasted images are supported as input methods.

10. Web Search and Chrome Browser Automation: Web search and Chrome browser automation are integrated directly into the Agent's workflow. The Agent doesn't just generate text — it can genuinely access and operate on web pages.

11. Deep Research (First Version): The first version of Deep Research is live, supporting multi-step investigation and synthesized output.

12. IM Integrations: Integrations include:

• Lark
• Discord