Bernard Aceituno
@bernaceitunoOver the past few weeks, we’ve been working on several updates: launching new models (including GPT 4.5), developing more agentic tools for your LLMs (e.g., Stripe, Jira), and introducing new export options like Google Sheets. Plus, you can now automatically import metadata tags from SharePoint.
Let’s dive into each update and explore how to use it.
Here’s a look at some of the most notable AI models that our team has recently released in Stack AI:
GPT 4.5 is OpenAI’s largest and best model for chat yet. GPT 4.5 scales up pre-training and post-training, increasing capabilities for unsupervised learning. This enhances pattern recognition, drawing connections, and generating creative insights without reasoning.
Claude 3.7 by Anthropic offers increased capabilities in coding and front-end web development. The model engages in an ‘extended thinking’ paradigm that improves performance in math, physics, and instruction-following.
DeepResearch by Perplexity performs dozens of searches, reads hundreds of sources, and reasons through the material to deliver comprehensive reports. DeepResearch excels in expert-level tasks, in domains such as finance, marketing, and product research.
Gemini 2.0 supports multimodal inputs like Google’s previous models, but now it can also perform native image generation, text-to-speech, and tool use. The model can also natively execute code, as well as third-party user defined functions.
As the saying goes: all data ends up in spreadsheets! Now you can export your AI workflow results to Google Sheets.
Simply connect your Google Drive and enter information for the spreadsheet you want to output the data into. Then manipulate your results with the power and flexibility of spreadsheets.
We are expanding the number of tools available by releasing two new integrations: Stripe and Jira.
Agentic tools allow LLMs to execute actions when triggered by the prompt, rather than automatically performing the action on each run. This makes AI agents faster and more efficient.
Here’s a look at both new tools.
Using Stripe as a tool, the LLM can issue refunds, list charges, and search for customers in Stripe. This makes it easier to automate finance applications and speed up financial transactions.
Our Jira tool allows the LLM to create Issues in Jira, including automatically setting assignees, descriptions, and custom fields via JSON. Combine this with our Zendesk tool to streamline your customer support workflow.
And a reminder: we also allow custom tools in Stack AI! You can build your own tool via an OpenAPI specification for third-party platforms.
Documents’ metadata is very useful when retrieving information from a knowledge base. It helps find documents based on topic-based tags, or prioritize the most recent ones.
We’ve implemented an automated system to import metadata tags from documents stored in SharePoint.
Click any document in your knowledge base and observe the metadata imported. You will have the option to add custom tags as well.
Don’t forget to activate the metadata filter strategy clicking your knowledge base settings.
As an admin, you can now set up a default connection to your LLMs, so users don’t have to do it manually.
Simply add your key once — and voila! Now everyone in the account is using your company’s LLM. This saves your users lots of set up time and increases security for enterprise companies.