A new type of user requires a new type of architecture

In February, we announced Universal Context, our next generation system of record that combined structured and unstructured data into a single, transactional data layer built for AI. It’s been incredible to watch our customers adopt Universal Context both in Attio directly and through connected AI tools like ChatGPT and Claude.

Since then, we’ve learned a lot about how the frontier of agentic revenue is being built. The systems that our most advanced customers are building demand incredible scale and reliability, whilst also needing increasingly powerful interfaces through which to access and consume data.

Historically, software has existed in two broad categories: application software that traded speed and scale for user experience, and infrastructure tools that traded increased complexity for higher speed and scale. CRM has always existed in the application software camp, but the demands of agents are reshaping our market, evolving the CRM into a hybrid platform that brings together the user experience for humans with the scale, reliability, and complexity of data access for agents.

Today, we’re taking the covers off our next leap in capability to help revenue agents go further, faster with Attio. Known internally as Project Centaur, this update brings the unprecedented connectivity and scale that was previously reserved for infrastructure like Data Warehouses directly into the CRM.

Centaur unlocks full PostgreSQL compatible SQL support for every Attio workspace with no extra steps. It offers a transactionally consistent query layer over the 100’s of millions of structured and unstructured data points that already live in Universal Context, and it does all of this in a fully hosted model that’s ideal for agents.

Introducing our SQL Engine

Making our hybrid architecture vision a reality required us to bring together the computational principles of a data warehouse with the storage techniques of a modern transactional system. We wanted to ensure that agents had a transactional view of the world, so that increasingly large numbers of concurrent running agents could avoid conflicts, whilst also allowing them to run queries and discover information from a broad corpus of data.

The result is Centaur - our Postgres compatible, horizontally scalable SQL query engine. At first glance, Centaur might look like a Postgres instance, but underneath it doesn’t share a single line of code with Postgres. Instead, Centaur operates as a translation and query planning layer, receiving queries over the Postgres protocol and translating them into a logical query plan that executes against Universal Context using our existing pathways.

This translation layer is free of the burdens of Postgres architectural decisions, meaning it doesn’t require any connection pools or party tricks to work. Your agents and applications connect directly to data.attio.com and interact with them using existing tooling designed for Postgres. As your agentic workload scales, Attio will handle the scaling story for you seamlessly.

This incredible scalability is only possible because of our existing investments in Particle and Universal Context which means we can separate out the work of planning and executing queries from the underlying storage capability underneath. Using the same techniques as a data warehouse like Snowflake or BigQuery, Particle is able to dynamically scale up the number of compute workers whilst still working on top of the same transactionally consistent storage layer.

The native language of agents

Expanding Attio with native SQL support also revolutionises the ability of agents to understand the CRM. Our system supports all common PostgreSQL query operators as well as commonly used INFORMATION_SCHEMA and pg_catalog utilities that agents and applications can use to dynamically understand the unique layout of an Attio workspace.

We translate each of the objects, attributes, records, and values that you have configured into a virtual SQL schema that aligns perfectly with what frontier models are already trained on.

Once the agent has used these familiar tools to explore the shape of the workspace, it is able to use all of the power of modern SQL including a wide variety of JOIN operations, complex aggregations, and pipeline operators. We expose easy to consume representations of the current state of the workspace’s data, as well as all of the historic data in a more complex tabular form.

All of this data allows agents running on Attio to answer questions that were previously expensive and unreliable with ease. For example, if you want to report on how many calls each member of your team made last week: Ask Attio can now deliver that in a single prompt and a single query ensuring you get an accurate answer every time.

One language, many modes

We’ve designed this feature from the ground up for every modality of agent running in the revenue stack today.

First, we offer tools that can be used directly in Ask Attio and via MCP that allow agents to execute read-only SQL queries against the data in Attio. We launched this functionality quietly a couple of months ago and have seen an incredible increase in agent speed, performance and accuracy. For example, Ask Attio is now able to answer specific questions like “How much revenue is in my pipeline for Q4?” using a deterministic query that is right every time.

Second, we have added support for SQL queries to our API so that workloads that are already using the developer API to perform complex aggregations can now execute these directly on Attio using our SQL engine instead. This saves potentially hundreds of API tools that would previously have been needed to fetch or replicate the underlying source data to drive aggregations.

Finally, we have introduced Attio Data Connectors. This innovative new product offering exposes a hosted, global, PostgreSQL wire protocol compatible SQL endpoint at data.attio.com . Customers are now able to provision unique credentials for their workspace which allow them to connect agents and tools that already support PostgreSQL directly to Attio. This also means that, for the first time, BI tools can run directly against the CRM without an ETL tool or data warehouse.

In each of these layers, every query runs against the latest, transactionally consistent state of your data. There is no replication delay between data being updated in the CRM and the result being visible in the SQL engine thanks to our advanced, transactionally consistent storage layer. This marks a significant improvement from agents that are running over data warehouses powered by ETL tools. These tools often run using scheduled sync that are either daily or weekly, meaning data used by an agent could be a week or more out of date at the time the agent consumes it.

New capabilities

Alongside this exciting new form factor for data in Attio, we are also introducing a powerful upgrade to our data model in the form of Interfaces. Interfaces bring the familiar software engineering concept to Attio, allowing users to express a shared set of attributes and configuration between multiple different Objects in our data model. Once combined into an Interface, the underlying Objects can be queried as a single logical table in the SQL engine, allowing you to report on account activity or deal status across a heterogenous family of objects.

We’re also releasing two system Interfaces that take advantage of this new functionality from day one: Activities and Interactions. These two concepts, that we’re exposing to users directly for the first time, have always existed in Attio under the surface. Now we’re bringing them directly to users as Interfaces that can be extended and modified.

Activity allows you to connect relevant data directly to your Objects (like Companies and Deals) directly - making sure that every touchpoint is recorded and available for your agents to review and report on.

Interactions extend Activity, bringing with them a concept of direction and membership ideal for things like meetings, calls, and conversations.

These two new standard interfaces connect directly to the existing systems you use in Attio every day such as Communication Intelligence and the record timeline without any extra steps.

The foundations of Agentic Revenue

As the shift towards Agentic Revenue begins, we’re going to see agents run at scales that would never have been reached by even the largest human revenue organisations. To support that, CRMs will need to evolve from applications that provide UX to humans, into hybrid platforms that also provide the technical infrastructure for agents.

The extraordinary new abilities we’re announcing today lay the foundations for the next generation of agents in the revenue stack. Whether running inside of Attio or in external platforms, agents are now able to use the familiar and universal language of SQL to access all of your data in Attio.

We believe that commerce is at the start of a tectonic shift and that the demands placed on software by this agentic workload will only continue to accelerate. Today marks a significant step in Attio’s journey along that path, and we cannot wait to see what our customers and their agents build with this next iteration of our platform.