Sphinx 0.7.5 — Speed and Steerability

Sphinx 0.7.5 is here! This update has focused on the two biggest asks from our users: speed and steerability. Here is what’s new:

• Increased control over Sphinx’s core flow. Sphinx previously had a fixed “stopping point” when it deemed a task sufficiently complete. We have made this much more configurable, so you can now ask Sphinx to work on a task as long as you’d like via prompting. As an example, all of these prompts will get Sphinx to build model with different levels of depth:
  • Make me a regression model to predict sales, but stop if any of the data looks unusual
  • Make me a regression model to predict sales, and use your best judgement to deal with any data issues
  • Make me a regression model to predict sales. I do not want you to defer under any circumstances, do it fully yourself
  • We recommend that you add a short sentence telling Sphinx how autonomous to be, especially when using the Sphinx CLI. If you want this instruction to persist, add it to your Sphinx Rules.
• Faster generation when working through a task. Sphinx is around 50% faster when working through multi-step tasks, with the most noticeable performance gains on step 5 and beyond – we’ve made major improvements to how context is managed and summarized by the agent.