All Articles
/
AirOps Updates

What's new in AirOps 1.4 🦙🔥

Llama 2, Sitemap Import, Text -> Code all live

Published on Feb 13, 2024 by

‍

Generated with AirOps 😎 🦙

‍

New LLMs available - Meta's Llama 2 + Stability AI

AirOps now supports Meta's Llama 2 (available in 7, 13, and 70 billion parameter sizes) and the latest Stability AI Stable LM model. More model selection makes it easier for you to experiment faster and find the perfect balance between output quality, response speed, and cost.

‍

‍

Build incredible transcription workflows with Deepgram's Nova 2

We are delighted to offer Deepgram's Nova 2 transcription model which is 18% more accurate than the previous Nova model and offers a 36% reduction in errors vs OpenAI Whisper (large). We are seeing teams across Marketing Sales, Product, and Customer Support build transcription workflows to improve quality and efficiency and are excited to see what new use cases more accurate transcription models unlock.

Import your Sitemap to Memory Stores

You can now import any sitemap.xml file directly into Memory Stores. This allows you to bring either page metadata OR entire pages into your Memory Stores, unlocking entirely new use cases :

- Automated internal linking for your SEO content
- Recommendation engines for customers
- Q+A systems over your website docs
- Net new content creation based on your existing content library

‍

‍

Text to Code 🪄

We recognize that using Code within AirOps can be a challenging technical hurdle, so we now support Text to Code. Now you can use natural language to easily Generate, Edit and Explain Python and Javascript Code steps within your workflows. This helps unlock more complex use cases such as chunking and re-compiling large text inputs and outputs, iterating over multiple portions of text (e.g., content outlines), and extracting specific data from LLM outputs.  

More Scraping Improvements

Token efficiency matters in prompting. We have modified our Google Search and Web Page Scrapers to optionally return markdown, reducing overall token usage by as much as 70%, while maintaining page structure and links.

‍