SmythOS Weekly Update: Claude 4.5 Haiku, Smarter Onboarding Tests, and Faster Debugging (October 13 – October 19, 2025)

weekly release notes changelog sprint 62

This week’s SmythOS release helps you test smarter onboarding experiences, explore the new Claude 4.5 Haiku model, and build large agents faster with smoother debugging. We’ve also polished component naming, improved consistency in the Builder, and fixed several chat and UI issues to make your workflow more reliable and fluid.

New Features

A/B test for mobile number in onboarding

We launched an A/B test that shows the mobile number field to some new free users while leaving others unchanged. Paid sign ups are not affected.

How this helps you

  • Learn whether asking for a mobile number improves activation or creates friction.
  • Make data driven decisions about sign up flow without disrupting paid user experience.
  • Reduce guesswork when optimizing onboarding conversion rates.

Claude 4.5 Haiku available across SmythOS

Claude 4.5 Haiku is now available in LLM components, AI model providers, and Agent Creation.

How this helps you

  • Try a new powerful model to improve quality or creativity in your agents.
  • Compare outputs across models to pick the best mix of speed, cost, and accuracy.
  • Add new capabilities to assistants and automations without reworking flows.

Improvements

API Call component: placeholder vs pre filled defaults A/B test

We are testing two setups: placeholders only, or pre filled safe defaults to speed testing.

How this helps you

  • Find out whether you prefer guided hints or immediate working values for faster experimentation.
  • Reduce friction when mocking or testing APIs in builder flows.
  • Save time on onboarding new integrators or documentation readers.

Component naming consistency in Builder and sidebar

Names now match between the canvas and the component sidebar.

How this helps you

  • Stop hunting for components or second guessing which item you edited.
  • Make documentation and troubleshooting more reliable.
  • Improve search, filtering, and log traceability.

Faster debugger toggle for large agents

Debugger toggles respond more quickly when working with large agents.

How this helps you

  • Debug and iterate faster without waiting for UI lag.
  • Keep momentum during heavy development sessions.
  • Reduce context switching and lost productivity.

Important Bug Fixes

  • Chat works for simple queries
    You can now ask simple questions like “who are you” and get a response instead of an error. This restores basic chat reliability.
  • Tutorial Previous button shows disabled styling
    When you can not go back, the Previous button now looks disabled so you are not confused.
  • Builder connections preserved after sorting and refresh
    Connections now reload correctly so visual order and functional wiring stay in sync.
  • Validation warnings appear only after user interaction
    Warnings no longer show up before you type anything, so you are not distracted by premature alerts.
  • Edit OAuth opens the correct modal
    Editing an existing OAuth connection now opens the Edit modal, not the Add modal. You can update credentials without workarounds.
  • Input buttons in prompt fields stay available after use
    You can reuse input injectors multiple times during a single editing session without refreshing.
  • Prettify aligns components after default position change
    Prettify now respects the new agent card default and keeps your canvas tidy and readable.
  • Agent sharing invites land on the shared agent
    Accepting a team invite now takes you to the actual agent, not the home page.
  • Save button visible on small screens
    You can save edits when working on small windows or narrow devices.
  • GenAI LLM component sidebar responsiveness improved
    Sidebar lag is reduced so settings open faster during long builder sessions.
  • Onboarding no longer shows unnecessary scrollbar
    Onboarding steps are cleaner when content fits the screen.
  • Agent Weaver chat stays functional after a successful build
    After you build an agent, Weaver chat continues to accept messages so you can iterate without restarting.

Try these updates

  • Open an LLM component and pick Claude 4.5 Haiku to compare responses with your current model.
  • Toggle the debugger on a large agent to feel the faster response.
  • Reopen critical API Call components and see whether placeholders or pre filled defaults speed your testing.
  • Use Prettify after layout changes to keep your canvas clean and documentation ready.

For step by step guides and best practices, visit the SmythOS documentation.