# Announcing Collate 2.1: Your Agents Now Know What Your Business Means

Nov 17, 2026

![Sriharsha Chintalapani](/_next/image?url=https%3A%2F%2Fcdn.hashnode.com%2Fres%2Fhashnode%2Fimage%2Fupload%2Fv1699594258729%2F81993512-5eef-4eb8-a0c3-28c8fe0f2c4b.png&w=96&q=75)

Sriharsha Chintalapani

![Announcing Collate 2.1: Your Agents Now Know What Your Business Means](/_next/image?url=https%3A%2F%2Fcdn.hashnode.com%2Fuploads%2Fgql%2F661eafb497eb31f2d368a2bf%2F7031da82-2e17-44ed-9f29-cd277b0676d9.png&w=3840&q=75)

Ask an AI agent for last quarter's revenue, and it may find several tables with a revenue column. Without business context, it cannot tell which figure Finance recognizes or how to apply the fiscal calendar. It can return a plausible answer that is wrong.

Collate is the Agentic Data Platform, built on the open-source OpenMetadata project. It gives AI the context to deliver more accurate and trusted answers to the business. Release 2.1 deepens what AI agents understand about your business: ontologies that give agents relationships and rules, more intelligent reasoning of your knowledge graph, and workflows that hold a governance change until it is approved by the right people.

The release has five parts:

*   **Model the business:** Ontology Studio turns your concepts into classes with typed relationships, binds them to your data, and exposes them to agents through the MCP server.
*   **Ground it in unstructured data:** Files you upload automatically become memories. These memories can then become proposed terms that a reviewer approves before they enter the ontology.
*   **Use graph relationships:** Agents follow relationships in the knowledge graph and learn from conversations through the Memory Agent.
*   **Stage changes for approval:** Governance workflows keep changes staged until an approver signs off, so metadata can change often without unapproved edits going live.
*   **Measure metadata coverage and manage metrics:** Audit Hub scores your metadata against a compliance standard such as BCBS 239, and Playbooks guide people through creating governed assets. Semantic metrics from your warehouse and BI tools, plus governed charts in AI Analytics, keep definitions and answers in one place. Hierarchical metrics organize metrics into parents and children.

Metadata, meaning, and memory belong in one context platform. Metadata tells an agent what data exists. Meaning tells it what the data represents. Memory tells it what your organization has learned. If they are spread across separate tools, agents get partial views that disagree with each other. Collate keeps all three on one graph, and 2.1 gives AI agents a deeper understanding of business meaning.

## **Model the Business**

Capturing business meaning for AI starts with an ontology model. Collate’s new Ontology Studio is where you build it.

### **Ontology Studio**

![Ontology Studio](https://cdn.hashnode.com/uploads/gql/661eafb497eb31f2d368a2bf/94b250b7-a03d-448b-a431-e134436aae56.png)

An ontology is how a business writes its rules down in a form AI can reason over. Ontology Studio lets teams create, edit, and query ontologies in Collate. You model concepts as classes with their own properties, connect them with typed relationships, bind them to the data that implements them, and put the model in front of your agents. Collate previously offered Ontology Explorer, which was designed for reading an ontology model. The new Ontology Studio is built for creating one, with different capabilities to View, Edit, and Query.

*   **Concepts as Classes**: Each concept gets its own fields. An Account has an account ID; a Loan has a principal and a term. Before 2.1, a custom property added to one glossary term showed up on every term, so Revenue and Churn got an account ID too. Now each property belongs only to the concept it describes. Each field has a data type, a field can be marked as the concept's unique ID, and child concepts inherit them. So, a Checking Account picks up the account ID from an Account parent.
*   **Typed Relationships**: Concepts connect through named, directional relationships such as part of, broader and narrower, and calculated from, plus types your administrators define. Each type is backed by an OWL or SKOS predicate, so a reasoner or an agent knows what the edge means.
*   **Meaning Mapped to Data**: Switch to the Data view and each concept shows the tables and dashboards bound to it. You can see the data table that actually holds account information, so agents can know where to query the correct data.
*   **Exposed to Agents Through MCP**: Collate's MCP server gives agents graph tools such as ontology\_describe, sparql\_query, entity\_neighborhood, and find\_by\_tag, so an agent can follow the relationships your business defined and return the query behind its answer. With sparql\_query, agents can query the RDF graph directly instead of following one relationship at a time.
*   **Start From What You Have**: Import an existing ontology from OWL, Turtle, N-Triples, or JSON-LD and preview what it will create. Starter packs for FIBO, GS1, EPCIS, ISA-95, HR Open, and HL7 FHIR give teams a pre-packaged model to get started more quickly and extend to their needs.

With an ontology in place, the agent can look up which revenue table Finance signs off on and how it connects to the fiscal calendar. It also shows the SPARQL query it ran, so your team can check the answer before using it. The model is built on open standards, and you can export it to any tool that reads those standards. The [Ontology Studio blog](https://www.getcollate.io/blog/ontology-studio-business-meaning-for-ai-agents) goes deeper on how it works.

## **Ground Meaning in Unstructured Data**

![Ground Meaning in Unstructured Data](https://cdn.hashnode.com/uploads/gql/661eafb497eb31f2d368a2bf/4a82b2ad-1e65-4aa9-a165-17562c56fcc6.png)

Much of what defines your business lives in documents: a revenue recognition policy in a PDF, a metric definition in a spreadsheet, the rules in a runbook. In Collate 2.1, the Context Center parses the PDF and CSV files you upload and vectorizes their contents, so agents can use what those files say. Collate turns an uploaded file into a proposed glossary term, and a human reviewer decides whether it is approved. The workflow:

*   **Upload a file:** Add a PDF or CSV to the Context Center. The new Memory Agent parses it, shows a preview, and draws on it when answering questions, along with your articles and memories.
*   **Memories come from the file:** Each memory links back to the file it came from, so you can check the source.
*   **Memories become proposed terms:** The Memory Agent promotes memories into proposed glossary terms. Each one includes a definition and links to the memory behind it.
*   **A reviewer approves:** A proposal stays a draft until a reviewer submits it.

Through this process, a definition someone wrote in a runbook document can end up as a governed glossary term, with a trail back to the memory and the file it came from.

## **Let Agents Reason**

In 2.1, Collate agents use the knowledge graph alongside vector search to retrieve related metadata with fewer tool calls and learn from conversations.

### **Agents That Use the Knowledge Graph**

Ask an agent a question about your customers, and it starts with vector search, which finds a table named customers. Agents can already find what's connected to that table, though it takes extra tool calls. In 2.1, Collate's agents also use the RDF knowledge graph to do it in fewer steps. From that table, the agent follows its relationships in the graph to see whether it has a metric, belongs to a domain, carries data quality tests, or feeds a dashboard. Those relationships come from metadata Collate already holds, such as ownership, lineage, and test results, so your team does not need to do any additional modeling.

A new graph tool lets agents generate and run SPARQL queries against the graph, instead of traversing it step by step.

### **Automatic Memory Extraction**

![Automatic Memory Extraction](https://cdn.hashnode.com/uploads/gql/661eafb497eb31f2d368a2bf/7f263672-431e-456b-a64c-97bfb5907c5c.png)

Collate 2.1 extracts memories automatically from your conversations and from the files you upload. Correct an agent in chat, and the Memory Agent saves the correction as a memory autonomously. If two memories conflict, the Memory Reconciliation capability resolves them. Each memory can be private, shared, or scoped to a single asset, and you can edit or retire it.

### **Governance That Holds Changes Until They Are Approved**

![Review pending changes: proposed edits to a data product wait for an approver](https://cdn.hashnode.com/uploads/gql/661eafb497eb31f2d368a2bf/cfdcc262-e5e7-4a43-b48f-4121eee74643.png)

Governance workflows in earlier releases applied a change first and then sent it for approval. A new tag, for example, appeared on a table right away and was removed only if the approver rejected it. In 2.1, a workflow holds the change until an approver signs off, so an asset shows an update only after it’s approved. We made this change in response to customer feedback, and it makes metadata easier to evolve. Since changes stay staged until approval, teams can propose updates freely while their assets stay trustworthy.

## **Measure and Prove**

Audit Hub scores your metadata against a compliance standard. Enhancements to Metrics and AI Analytics keep definitions and answers in one governed place.

### **Audit Hub (Beta) and Playbooks**

![Audit Hub (Beta) and Playbooks](https://cdn.hashnode.com/uploads/gql/661eafb497eb31f2d368a2bf/ee5f71b7-8d04-4440-8b8d-5e4784865d23.png)

Audit Hub scores your metadata against a compliance standard your administrators publish and actionably shows where the gaps are for remediation. A centralized governance dashboard covers your entire data estate and automatically refreshes daily. The scoring is deterministic, so the same inputs give the same result. Each run records its score, the rule version, and those inputs, so you can track results over time.

*   **Standards and Rules**: BCBS 239 ships as the first included compliance standard, and you can also define your own custom standard. A standard evaluates different dimensions like documentation, ownership, data quality, and data contracts. Each check is an explicit condition, such as minimum description length. Changes to these checks can be dry-run before publishing them as a new rule version.
*   **Gaps Ranked by Usage**: Audit Hub ranks failing assets by downstream use, so the ones people read most come first. Assign owners in bulk or export the list.
*   **Scored by Ownership**: Domains, teams, and data products each get a compliance score. Teams are scored on the assets they own.
*   **Playbooks**: These references guide people through creating and publishing a data product: the fields to fill in, the conditions it must meet, and the tasks along the way. Playbooks now include custom intake forms.

For banks working toward BCBS 239, the built-in standard gives you a starting point: a regular audit of documentation, ownership, and data quality coverage across the estate. When an auditor asks how a score was produced, your team can rerun it.

### **Semantic Metrics From Your Systems, in One Place**

The same metric is often defined across different systems, and can be scattered across different warehouses, dashboards, and pipelines. Collate 2.1 brings those definitions into the context graph as first-class metrics.

*   **Snowflake**: Metrics from semantic views, with their expressions, dimensions, and underlying facts.
*   **Databricks Unity Catalog**: Measures from metric views, with their dimensions and links to the source view.
*   **Looker**: LookML measures, carrying their expressions, aggregations, filters, and dimensions.
*   **Power BI**: Measures from semantic models, preserving their DAX definitions and relationships to other measures.
*   **Tableau**: Calculated measures, with their formulas, dimensions, and dependencies.
*   **dbt**: Builds on semantic-layer metric ingestion already available in 2.0, including lineage from metrics back to source tables and columns.

Imported definitions can be governed with ownership and tags, linked to their source assets, and explored through available lineage. Chat, dashboards, and agents that each read a different definition of revenue give different answers to one question. Bringing those definitions into one governed place is the first step toward one definition of revenue for all of them.

### **Hierarchical Metrics**

![Hierarchical Metrics](https://cdn.hashnode.com/uploads/gql/661eafb497eb31f2d368a2bf/cd77d578-a6f6-4fec-a276-a6c5f30772be.png)

A company-level measure is rarely one number. It rolls up from measures that belong to a region, a product, or a team. In Release 2.1, you can organize metrics into hierarchies, similar to how a glossary organizes terms, so the definition at the top and the definitions beneath it sit in one structure.

### **Query More of Your Data in SQL Studio**

Collate 2.1 adds Amazon Athena and MySQL support to SQL Studio. Run queries against these services directly in the Collate UI, alongside existing support for Amazon Redshift, BigQuery, Snowflake, PostgreSQL, and Trino.

### **AI Analytics Enhancements**

![AI Analytics Enhancements](https://cdn.hashnode.com/uploads/gql/661eafb497eb31f2d368a2bf/0eb6df74-e056-4701-91be-15599071774f.png)

AI Analytics turns a plain-English question into a governed chart or dashboard, referenced back to the definitions already in Collate. 2.1 adds:

*   **New Advanced Charts:** More chart types, including country maps, scatter plots, treemaps horizontal/stacked, and Sankey.
*   **Analyze Your Own Files**: Upload a CSV or Excel file directly in the chat and ask questions about it in plain English. AI Analytics can cross-reference the file with data in your warehouse, so you can export a list of customer IDs from Snowflake, upload it, and drill into your analysis.
*   **Athena, MySQL, Databricks, ClickHouse, and Microsoft Fabric in AI Analytics**: AI Analytics now supports 13 services through the same query execution backend that powers SQL Studio.
*   **Governed Like Other Assets**: Dashboards keep their owners, access controls, and the chart-level build history that shows how each chart was made.

## **Also in Collate 2.1**

*   **Hierarchical Lineage**: Start at the service level and zoom in through the database, schema, and asset to a single column. This helps teams navigate large lineage graphs.
*   **New Landing Page:** The new home screen works like an AI-powered inbox: widgets that each had their own job now sit on one page, and you can add widgets from Classic mode UI. Collate adds AI-based widgets, including "actions required" and "recommended actions." Administrators can set a starting page for each persona: Home, My Data, Data Quality, Data Observability, or Data Marketplace. Data consumers can land where they find data products first.
*   **AI Governance Studio, Beta:** AI Governance Studio moves from preview to beta in 2.1. AI applications, LLMs, MCP servers and agents get owners, tags, lineage and certification like other governed assets, and an agent can be packaged into a data product with its MCP servers and tables behind a single access request.
*   **Team and user AI credit limits:** Admins set AI credit limits for each user and each team. Collate alerts people as usage approaches a limit. The billing page shows credit use by user, with monthly and weekly breakdowns, so admins can check limits against actual use.
*   **Conversational Data Insights:** Data Insights charts of key health metrics like description and quality coverage now come from asking AI Analytics, replacing the manual chart builder and dashboards.
*   **AI Mode UI:** The user interface now defaults to the new AI Mode UI for streamlined navigation, improved productivity, and a conversational AI-oriented chat experience. Dark mode and the other 2.1 additions are built for the new UI only, and the Classic mode UI is deprecated and will be removed in a future release.
*   **Dark Mode:** Available across the new AI Mode UI.
*   **Support Agent:** Ask product questions in the chat and get the steps, with links to the docs, without leaving Collate. When a connector run fails, open it and diagnose the failure with AI.
*   **Collate Desktop:** A desktop app for Mac and Windows that opens your Collate instance in its own window, with deep links, a command palette, and desktop notifications.  
    ![Also in Collate 2.1](https://cdn.hashnode.com/uploads/gql/661eafb497eb31f2d368a2bf/e6fad257-0bc8-49ff-bed7-90741f4ad181.png)

## **Feature Availability**

Collate is built on OpenMetadata, and provides these enhancements for context and data teams at scale as part of the 2.1 release.

Capability

OpenMetadata

Collate

Ontology Studio

No

Yes

Knowledge graph tools through MCP

Yes

Yes

Hierarchical metrics

Yes

Yes

Semantic metrics

Yes

Yes

Hierarchical lineage

Yes

Yes

New landing page, regular widgets

Yes

Yes

New landing page, AI widgets

No

Yes

Persona starting pages

Yes

Yes

Agents that use the knowledge graph

No

Yes

Memory Agent

No

Yes

Approval before changes apply

No

Yes

Audit Hub (beta)

No

Yes

Playbooks

No

Yes

Context Center file parsing

Yes

Yes

New UI and dark mode

Yes

Yes

AI Analytics

No

Yes

AI Governance Studio (beta)

No

Yes

Desktop App

No

Yes

Support Agent

No

Yes

Budget and Billing Enforcement

No

Yes

## **New Connectors**

Collate 2.1 adds eight new connectors, covering customer data platforms, graph and analytical databases, messaging systems, BI tools, and data-prep workflows.

*   **Salesforce Data 360 Database**: Catalogs data spaces and their Data Lake Objects, Data Model Objects, and Calculated Insights in Salesforce's customer data platform, formerly Data Cloud.
*   **Salesforce Data 360 Pipeline**: Brings Data Streams, Calculated Insights, and Data Transforms into the catalog as pipelines, with lineage connecting the data they read, transform, and produce.
*   **Neo4j (beta)**: Catalogs node labels, relationship types, and their typed properties from Neo4j graph databases.
*   **Tableau Pipeline (beta)**: Catalogs Tableau Prep flows and extract refreshes as pipelines, with run history. Prep task graphs show inputs, processing, and outputs, with lineage between source and output assets. Extract refreshes link to the data models they refresh; refresh history requires a Tableau site or server administrator.
*   **Rill**: Ingests Explores and Canvases as dashboards, Canvas components as charts, and models and metrics views as dashboard data models, with lineage back to their physical source tables.
*   **NATS JetStream (beta)**: Catalogs JetStream streams as topics, with optional schema extraction from a configured key-value bucket and message sampling. Supports TLS, username/password, token, or NKey authentication.
*   **ClickZetta (beta)**: Adds metadata ingestion, query usage and lineage, profiling, data diff, and dbt artifact ingestion for the ClickZetta Lakehouse database.
*   **Databend (beta)**: Catalogs tables, views, and columns in Databend's default catalog, with table and column profiling.

## **SSO and Authentication**

Collate 2.1 makes sign-in simpler to manage, with fewer interruptions for users and more control for administrators.

*   **Preview Your SSO Setup**: Try a real sign-in with Test Login before saving OIDC, SAML, or LDAP settings, so administrators can confirm the experience before applying changes.
*   **Stay Signed In**: SSO sessions follow your identity provider's active session and stay consistent across open browser tabs, helping people work with fewer interruptions.
*   **Keep Access in Step**: When roles are managed through your identity provider, each sign-in reflects current role assignments and administrator access.
*   **Manage SSO From AI Mode**: Administrators can manage authentication settings directly from their profile in AI Mode.
*   **Stronger Security, Built In**: Sign-in includes additional safeguards, and revoking a bot or personal access token takes effect immediately across every server.

## **Get Started**

Collate 2.1 is rolling out in November, including on the [Free Tier](https://www.getcollate.io/free) of our managed OpenMetadata service.

[Contact sales to talk to a product specialist.](https://www.getcollate.io/contact-sales)

[#ontology](/blog/tag/ontology)[#ai-agents](/blog/tag/ai-agents)[#ai-analytics](/blog/tag/ai-analytics)[#data-governance](/blog/tag/data-governance)[#metrics](/blog/tag/metrics)[#data-lineage](/blog/tag/data-lineage)

Ready for trusted intelligence?

See how Collate helps teams work smarter with trusted data

[Get Started](/welcome)[Contact Us](/contact-sales)

## Keep Reading

[View All](/blog)