✨ This week in Maia: Smarter search, Salesforce crawling in Context Engine, and new BigQuery AI components

Welcome back to the Maia New Features Blog! This week, we're rolling out updates that help you find what you need faster, bring native AI transformations to Google BigQuery, and extend your knowledge graphs beyond the warehouse. For a full list of recent changes, be sure to check our changelog updates. Let's dive into the details.

🔍 Smarter search and suggestions in Add Component and file search

Finding what you need is now faster, with search that learns from the way you work.

• Add Component now shows three suggested components, scoped to the type of pipeline you're building.
• Quick file search (Cmd+K) shows five suggested files, and the file panel shows 10.
• Search in Add Component and the file panel now matches near-misses, so a typo still finds what you meant.
• Search results are ranked for you, so the things you use most appear higher.

The file panel has also been redesigned. The "All" and "Recent" tabs have been removed: clicking the search box swaps the file tree for your suggested files, and typing swaps those for your search results.

Suggestions and result ranking are powered by a score called "frecency", a blend of frequency and recency. Each time you open a file or add a component, its score goes up, and anything you stop using slowly loses score over time. This means your suggestions and results follow what you're working on now, rather than what you used months ago.

The benefit is that your likely next step is already waiting when you open search, with nothing to pin, star, or tidy up. Suggestions and ranking adapt on their own and are personal to you. They're stored in your browser, so they're specific to each browser and device.

✨ Native AI transformation components for Google BigQuery

Six new transformation components are now available, bringing AI transformation capabilities to Google BigQuery alongside those already available for Snowflake (Cortex) and Databricks:

• Generate: Produce text, or typed structured output, from a prompt using Google Gemini.
• Embed: Turn a text column into a vector embedding, the first step in building in-warehouse RAG.
• Vector Search: Run a nearest-neighbor search against a base table of embeddings, completing the Generate, Embed, and Vector Search retrieval pattern.
• Classify: Assign each row to one (or more) of a set of user-defined labels.
• Score: Assign each row a numeric rating against a prompt-stated rubric, for ranking and ordering.
• Check Condition: Flag each row true or false against a natural-language yes/no condition, pairing with Filter for conditional routing.

Each component wraps one of Google BigQuery's own built-in AI SQL functions (AI.GENERATE, AI.EMBED, AI.CLASSIFY, AI.SCORE, AI.IF, and VECTOR_SEARCH). Google BigQuery itself makes the model call as part of the generated SQL, so your data stays inside Google Cloud.

Read the linked documentation for each component to get started.

🕸️ Context Engine: crawl your external sources, starting with Salesforce

You can now add external sources to a knowledge graph directly from the Context Engine UI. Crawlers now reach beyond the warehouse to the systems that feed it, starting with Salesforce.

Your knowledge graph already covers what's in your warehouse and which pipelines filled it. Much of your business logic lives further upstream, in systems like Salesforce. By crawling those sources, Maia Team can now answer "where did this data come from?" as well as "what's in my warehouse?", and lineage from pipeline runs reaches all the way back to the source object that feeds each table.

What's new:

• External source crawlers: In a graph, open Crawlers → Add crawler and pick an external source. Point it at an existing connection and choose the objects you want to crawl.
• No new credentials: Crawls use your existing project connections through the Runner, so secrets never leave the Runner.
• Linked into the graph: Source objects and fields become nodes, linked to the warehouse tables and pipeline runs they feed, so Maia Team can trace data end to end.
• Scheduling and access controls: These work in exactly the same way as they do for warehouse crawlers.

We're starting with Salesforce and will add connectors one at a time, so that each one produces good-quality graphs for you.

How to get started: Open Context Engine from the hub and pick a knowledge graph (or create one). Go to Crawlers → Add crawler and choose External source → Salesforce. Select a Salesforce connection from your project and choose the objects to crawl, then run it or put it on a schedule and start asking Maia Team.

Read the Creating knowledge graphs documentation to learn more.

💬 We'd love to hear from you!

Let us know how these new features are improving your workflows—we're all ears! Join the conversation in the Matillion Community.