New updates and improvements
V7 Go

Reasoning models, from Deepseek to o3, have taken the world by storm.

Compared to traditional LLMs like GPT4o and Claude Sonnet 3.5, o3 has explicit step-by-step deduction baked into the training process, making it superior for tasks requiring complicated logical reasoning.

o3-mini can now power all property types in V7 Go, including Collections, Numbers, JSON and more.

In addition, we’ve added Gemini Flash 2.0, a powerful workhorse model with low latency and a large (1m) context window.

The Number Property enforces the output of numeric values, making it easier to work with data such as integers, percentages, and currency values.

Common prompts might be:

• Extract the revenue from @company in the @pitch deck. Include only the revenue for the last year. Output in $.

• What percentage of costs in the @CIM come from salaries?

• What is the net income in the @earnings report?

In addition, you can format decimals, negatives, separators, and alignment from the property configuration menu.

Stay tuned, as over the next few days, we’ll be adding both reasoning and AI citations to the Number Property.

Users can now recompute specific property values for individual fields without recalculating an entire property column. If a field produces an error or the output isn’t satisfactory, simply click on the field and select the Recompute option. You can also select a range of values, right-click, and recalculate multiple fields at once.

Until now, you could access and analyze web data in V7 Go by submitting specific URLs. With the new Web Search feature, it is now possible to ask questions directly in your prompt and pull in relevant information from the web.

• V7 Go will automatically use the most effective search queries and visit multiple pages.

• Perfect for data enrichment and accessing current information like stock prices.

• Get results as text, structured JSON, or Select/Multi-select properties.

Note: Source URLs and citations will be automatically displayed in the reasoning tooltip for full transparency in an upcoming update. In the meantime, include "list sources" in your prompt to view reference links in the output.

You can now upload Word documents directly into V7 Go without converting them to PDFs or CSVs. Use AI Citations, preview mode, and other functionalities with your Word documents just as you would with PDFs.

V7 Go supports both .doc and .docx formats.

Note: This feature requires new file properties created after this release.

You can now create collections within collections to get more granular insights into your data. This unlocks powerful workflows like splitting a PDF into pages as a top-level collection, then extracting structured metrics for each page in your nested collections.

With nested collections, you can still enjoy all the functionalities of standard collections, including filtering, AI data extraction, and dynamic references.

This new property type lets you link data between projects. You can pull information from V7 Go source projects based on conditions you set.

References unlock:

• Dynamic referencing. Pull in matching data from other projects automatically. For example, you can analyze Gong call transcripts involving a specific participant in one project, and then sum the deal values attached to those calls in a different one.

• Manual matching. Need complete precision? You can manually filter and select specific matches for nuanced use cases.

• Quick previews: Hover over matches for instant previews and snap to the source project in one click.

Whether it's enriching a project with relevant customer interactions or linking supporting documents to financial models, References help you connect all necessary data points with ease. Best of all, your data stays in sync automatically.

To see Reference Properties in action, check out this 90 second walkthrough.

File Bundles are now live in Go. This functionality is built for use cases where you need to work with a unit of data (a company, a checklist, an email with multiple attachments) which includes multiple files, such as:

• Company analysis where a single company consists of a pitch deck, accounts files, and interview transcripts

• Financial reports with supporting documentation and regulatory filings

• Insurance claims with policies, damage photos, written reports, and scanned forms

For instance, say you’re working with a company profile that consists of multiple files: simply add the pitch deck PDF, financial CSVs, and any other relevant documents into a single bundle/collection. This allows you to group multiple files under one cohesive unit.

Additionally, you can apply filters to specify which files within a collection/file bundle should be analyzed with AI. For a quick walkthrough, watch this 90 second Loom.

If you have any questions or a relevant use case, as always, reach out to us.