This week’s SmythOS release packs serious capability upgrades. You can now scrape websites through country-specific proxies with a simple dropdown, access OpenAI’s latest GPT5.2 model, and take advantage of expanded vector database options in Data Pools v2. The team also shipped MCP OAuth 2.1 support, production logging controls for chatbots, and a long list of fixes that tighten up the Builder and chat experiences.
New Features
WebScrape Multiple Country Selection with Proxy Support
The WebScrape component now includes a Countries field where you can select one or more countries from a dropdown menu. Your scraping requests automatically route through proxies in those geographic locations. No more manual proxy configuration for each country you need to target.
This is useful when you need region-specific content, localized pricing data, or information from country-restricted websites. Select the countries you care about, and the proxy routing handles the rest. Your settings persist across sessions, so you don’t have to reconfigure every time.
How this helps you
- Scrape geo-targeted content without manually configuring proxies.
- Access localized pricing and region-specific data from a single component.
- Save time with persistent country selections across sessions.
GPT5.2 Model Support
GPT5.2 is now available throughout SmythOS. You can select and configure it anywhere AI models appear, including agent creation, workflow steps, and automation tasks. The integration has been tested end-to-end to make sure it works reliably across the platform.
How this helps you
- Access OpenAI’s latest model capabilities in your agents and workflows.
- Choose from an expanded set of AI models based on your specific needs.
- Build more advanced automations with GPT5.2’s improved performance.
Data Pools v2
Data Pools gets a major upgrade with v2. You now have additional configuration options and support for more vector database providers. The team built this carefully to avoid breaking existing Data Pool functionality, so your current setups continue working while you gain access to new capabilities.
How this helps you
- Connect to more vector database providers without leaving SmythOS.
- Configure Data Pools with greater flexibility for your specific use case.
- Upgrade to v2 without disrupting existing workflows.
Execution Logs Toggle for Deployed Chatbots
A new toggle in the chatbot configuration modal lets you enable or disable execution logs in deployed environments. Previously, this was only available in development and debug mode. Now you can turn on logging for production chatbots when you need visibility into what’s happening.
How this helps you
- Debug production chatbot issues without redeploying to a test environment.
- Monitor live chatbot behavior with detailed execution logs.
- Control logging overhead by toggling it on only when needed.
MCP OAuth 2.1 Authorization
SmythOS now supports OAuth 2.1-based authorization for MCP servers using HTTP transport. This implementation follows the MCP specification requirements, including Protected Resource Metadata (RFC 9728), PKCE support, token validation, and resource indicator handling.
If you’re deploying MCP servers over HTTP, this gives you proper authentication and authorization that complies with the specification. Users authenticate through familiar OAuth flows while the system maintains token audience binding and prevents unauthorized access.
How this helps you
- Secure your MCP servers with standards-compliant OAuth 2.1 authorization.
- Protect sensitive data and operations in HTTP-based MCP deployments.
- Give users familiar OAuth authentication flows when accessing MCP resources.
Improvements
Clearer Error Message for GenAI LLM Attachment Input
The GenAI LLM component used to show “model does not support files” when you passed non-binary data like strings, to the attachment input. That was misleading because the model might support files just fine. The real issue was the input format. The error message now clearly states that binary data is required.
How this helps you
- Understand immediately what’s wrong when attachment inputs fail.
- Fix input format issues without guessing about model compatibility.
- Debug attachment-related problems faster with accurate error messages.
Chatbot Architecture Improvements
The chatbot feature received comprehensive refactoring under the hood. This includes API client enhancements, better context state management, optimized file handling, and various user experience improvements. The code is cleaner now with consistent model override handling, standardized hooks, removed unused components, and stronger type safety.
How this helps you
- Experience smoother chatbot interactions with better error handling.
- Benefit from improved input focus management and message delivery.
- Trust a more reliable system built on cleaner, more maintainable code.
Chat Architecture Improvements
Similar to the chatbot work, the Chat feature got its own architecture refresh. The ChatAPIClient was enhanced, chat context was refactored, and the file management system was upgraded. Focus areas included code quality, state management, error handling, and performance optimization.
How this helps you
- Enjoy better scrolling behavior and input management in chat.
- Experience more responsive chat interactions overall.
- Rely on a chat system that’s more stable and easier to maintain.
Optimized Avatar Generation
Avatar generation now pulls from a library of pre-generated avatars instead of generating new ones via AI every time. This applies to both initial agent creation and manual regeneration through the Generate button. You still get diverse avatars, but without the AI API calls and associated costs.
How this helps you
- Get avatars instantly without waiting for AI generation.
- Reduce costs associated with avatar creation.
- Maintain avatar diversity across your agents without performance overhead.
Bug Fixes
This sprint addresses thirteen issues affecting chat context, component behavior, Builder usability, and visual consistency throughout the platform.
Chatbot Missing Conversation History Context
The chatbot was treating each message as completely isolated, forgetting everything from earlier in the conversation. If you told it your name or job, it wouldn’t remember a few messages later. Conversations felt disjointed, and you had to repeat yourself constantly. The chatbot now properly retains session history.
How this helps you
- Have natural conversations without repeating information.
- Build chatbots that remember context throughout a session.
- Deliver coherent user experiences in your deployed chatbots.
SmythOS Agent Component Extra Outputs Not Mapped to JSON
The SmythOS Agent component was returning plain text for extra custom outputs instead of structured JSON. This was inconsistent with how the GenAI LLM component handles the same situation. Workflows expecting JSON responses would break. Extra outputs now map to JSON properly.
How this helps you
- Get consistent JSON output from SmythOS Agent components.
- Build reliable workflows that depend on structured responses.
- Avoid integration failures from unexpected output formats.
Missing Variable Helper Buttons in RAG Remember Metadata Field
The Metadata field in RAG Remember component settings was missing the helper buttons for injecting input and global variables. Other fields had them, but this one forced you to type variable syntax manually. The backend worked fine, it was just the UI that was lacking. The helper buttons are now present.
How this helps you
- Insert variables into RAG Remember metadata with button clicks.
- Avoid syntax errors from manual variable entry.
- Configure RAG components as efficiently as other components.
Builder Sidebar Navigation Icons Missing Tooltips
The main navigation icons in the Builder sidebar stopped showing tooltips on hover. The Agent Weaver icon still worked, but everything below it, clipboard, organization, settings, and others, gave you nothing when you hovered. Tooltips are back across all sidebar icons.
How this helps you
- Identify sidebar icon functions without clicking to find out.
- Navigate the Builder confidently with consistent tooltip behavior.
- Avoid confusion when looking for specific features.
Gemini 3 Empty Response Error
Gemini 3.0 and 2.5 models were sometimes returning HTTP 200 responses with empty content despite generating reasoning tokens. The API said finish_reason: STOP but provided no actual response text. This happened intermittently across multiple use cases including grounded search and JSON generation. The issue has been addressed.
How this helps you
- Get actual responses from Gemini models instead of empty content.
- Run production workflows without mysterious silent failures.
- Avoid manual retries and workarounds for Gemini API calls.
GenAI LLM Echo Mode Treats Integer Zero as Null
In Echo mode, passing the integer 0 as input caused the GenAI LLM component to treat it as null. Instead of substituting the value 0, it would inject literal variable syntax like “undefined” into the prompt. Other numbers worked correctly, just zero was broken. Zero now substitutes properly.
How this helps you
- Use zero values in your workflows without broken variable substitution.
- Get accurate prompt outputs when working with numeric data.
- Debug and run production workflows that involve zero values.
Resizable Left Sidebar Layout Issues
Widening the resizable left sidebar in the Builder didn’t trigger proper responsive behavior. Text got truncated in sections like Zapier, SmythOS Agents, and Hugging Face. Delete buttons stayed fixed in position instead of moving with the expanded width. The sidebar now adapts correctly when resized.
How this helps you
- See full component names when you widen the sidebar.
- Access delete buttons in their expected positions.
- Work with the sidebar at whatever width suits your workflow.
Component Settings Expanded Text Area Missing Scroll Bar
When you expanded a text area modal in component settings, long content that exceeded the visible area had no scrollbar. You couldn’t access or edit the full text. The scrollbar now appears when content overflows.
How this helps you
- View and edit complete text in expanded modals.
- Work with long content in component settings without limitations.
- Access everything you’ve written without missing hidden portions.
Debugger Incorrectly Attaching to Default Attachment Fallback Skill
In debug mode, the default attachment fallback skill would execute successfully, but then trigger a “No debug session to attach” toast after a few seconds. The system was wrongly trying to attach a system-level skill to the debugger. This misleading error no longer appears.
How this helps you
- Debug agents without confusing error messages after successful executions.
- Focus on your custom skills without system-level noise.
- Trust that error messages reflect actual problems.
Remixing Template Defaults to Position 0,0
Remixing a template was dropping you at canvas position 0,0 instead of using the saved template position or defaulting to a view centered on the agent card. You’d have to manually navigate to find your components. Remixed templates now open at sensible positions.
How this helps you
- Start working immediately after remixing without hunting for components.
- See your agent card and components right away.
- Experience smoother template remixing workflows.
Multiple File Upload Fails with Comma-Separated String
Uploading multiple files in Chat and Chatbot interfaces failed when files were passed as a comma-separated string instead of an array. This broke document and image processing workflows that needed to handle multiple files at once. Multiple file uploads now work reliably.
How this helps you
- Upload multiple files in chat without failures.
- Build agent workflows that process batches of documents or images.
- Trust that multi-file handling works as expected.
Inconsistent Button Border Radius
Buttons throughout SmythOS had varying border radius values. Some were rounded, some were sharper, and the inconsistency showed up in the Builder, Agent Settings, modals, and navigation elements. All buttons now use a consistent 4px border radius.
How this helps you
- Experience a polished, visually consistent interface.
- Work in an application that looks intentionally designed.
- Avoid the distraction of mismatched UI elements.
Global Variables Default Values Not Resolving
Global variables set as default values in component inputs weren’t being resolved during execution. The system passed the raw variable reference instead of the actual value. Workflows that depended on these defaults would fail or process incorrect data. Default values now resolve properly.
How this helps you
- Use global variable defaults in components without failures.
- Build workflows that rely on default values with confidence.
- Avoid unexpected data processing errors from unresolved variables.
Try These Updates
- Add a WebScrape component and select multiple countries from the new dropdown.
- Create an agent using GPT5.2 to test the new model integration.
- Check out Data Pools v2 and explore the expanded vector database options.
- Enable execution logs on a deployed chatbot and monitor its behavior.
- Have a multi-turn conversation with a chatbot to confirm context retention works.
- Resize the Builder left sidebar and verify text and buttons adapt properly.
For a deeper understanding and additional examples, refer to the official SmythOS documentation, which provides a more detailed exploration of these changes.To schedule a meeting or request a demo, click here.