Honeydew is a semantic layer. Entities, attributes and metrics are defined once, stored as YAML under Git version control, and compiled into SQL that runs inside Snowflake, Databricks or Google BigQuery. It serves those definitions to BI tools and AI agents through SQL, XMLA, GraphQL and MCP.

2. Is a semantic layer the same thing as semantic layer governance?

No. A semantic layer is where a definition is written and computed. Governance is the decisions around it: who may author it, whose version is authoritative when two exist in different tools, what it depends on down to the column, and whether it is certified or a draft.

3. Does Honeydew have version control for metric definitions?

Yes. Honeydew stores each entity, attribute, metric, dataset and domain as a separate YAML file in a Git repository, and supports all major Git providers. Publishing a change automatically opens a pull or merge request, and branch protection rules can require review.

4. When is Honeydew alone the right answer?

When the semantic model is the only place metrics are defined, one team owns authoring, the warehouses in scope are ones it supports, and Git review is the level of change control the organization needs. That is a real and common situation, and adding a second layer to it buys very little.

5. What does a governance layer add on top of a semantic layer?

Reach and trust status. It connects definitions living outside any one model, in BI measures, transformation code and warehouse objects, records ownership and certification on each, traces a metric to the columns it reads, and delivers those signals alongside the number.

6. Can you run Honeydew and Atlan together?

Yes, and that is the common enterprise pattern. The semantic layer holds and computes the definition; the context layer records who owns it, whether it is certified, what it depends on and whether the data is current, then delivers all of it through an MCP server.

Honeydew vs Atlan for Semantic Layer Governance

Emily Winks

Data Governance Expert

Updated:09/03/2026

Published:09/03/2026

14 min read

Watch Context Layer Live Get the Context Layer Ebook

Key takeaways

Honeydew is a semantic layer: entities, attributes and metrics defined as YAML and compiled into warehouse SQL.
Semantic layer governance is a separate job that sits above any semantic layer, including a very good one.
Honeydew publishes model changes through a pull request, so change control is already part of its workflow.
The scoping question is not which tool, it is whether definitions outside the semantic layer are governed too.

Honeydew vs Atlan: what is actually being compared?

Honeydew is a semantic layer. Entities, attributes and metrics are defined once as YAML, versioned in Git, and compiled into SQL that runs inside Snowflake, Databricks or Google BigQuery. Semantic layer governance is a different job that sits above any semantic layer: who may author a definition, whose version is authoritative when two exist in different tools, what a metric depends on down to the column, and whether an agent can be told the answer is certified and current. Atlan does that second job. Most enterprises need both.

Key distinction

Honeydew — semantic layer; entities, attributes and metrics compiled to warehouse SQL
Semantic layer governance — authorship, authority, lineage and trust status above any model
Where they meet — an agent asking for a metric needs the definition and its trust signals together
Atlan — the Enterprise Data Graph and MCP server that carry the second job across tools

See how far your metric definitions actually reach

Start the Assessment

Honeydew is a semantic layer. Business entities, attributes and metrics get defined once, and its compiler turns them into SQL that runs inside Snowflake, Databricks or Google BigQuery.^[1] Semantic layer governance is a different job, and it sits above any semantic layer: who is allowed to author a definition, whose version wins when finance and product both write one for “active customer,” what that metric depends on down to the column, whether it is certified or still a draft, and whether an agent asking for it can be told the number is trustworthy right now. Atlan’s Enterprise Data Graph and MCP server carry that second job across the tools where definitions accumulate. The scoping question is not which product, it is whether that job is covered anywhere at all.

Gartner has put numbers on what happens when it is not. Organizations that prioritize semantics in AI-ready data could raise agentic AI accuracy by up to 80% and cut costs by up to 60% by 2027, with Distinguished VP Analyst Rita Sallam framing it plainly: “Agentic AI outcomes depend on context including semantic representations of data.”^[7] That is an argument for having a semantic model, and separately for knowing which one an agent should believe.

Dimension	Honeydew	Semantic layer governance
What it is	A semantic layer for AI and BI, compiled and executed in the warehouse	The decision layer above whichever models and tools hold definitions
Core objects	Entities, attributes, metrics, datasets, domains	Owners, certification status, lineage, policy, freshness
Where it runs	Inside Snowflake, Databricks or Google BigQuery	No compute of its own; it governs and delivers context
Change control	Git-backed YAML; publishing opens a pull request	Ownership and approval recorded on the definition itself
Interfaces	SQL, XMLA, GraphQL and MCP	MCP server, model agnostic
Question it answers	“What does this metric mean, and how is it computed correctly?”	“Which definition is authoritative, and can it be trusted today?”

Every Honeydew claim here is footnoted to its own documentation or to Snowflake, verified 2026-09-03. Capabilities no public source confirms are not asserted either way.

Honeydew vs semantic layer governance: what is actually being compared?

These are two different purchases that share a word. Staging them as a feature fight produces a table nobody can act on, because one is a modeling decision and the other is an operating decision.

The modeling decision is where a metric lives and how it computes, which is what a semantic layer for analytics is for. The operating decision is what happens once definitions exist in several places at once, the normal end state for any company older than its current data stack. That second question is the subject of context layer vs data catalog vs semantic layer, and a better model does not answer it.

Six questions define the governance job, and a buyer can check each against their own stack in an afternoon. Who may create or change a metric. Whose definition is authoritative when two exist. What it depends on, down to the physical columns, the work data lineage for AI describes. Whether it is certified, deprecated or somebody’s experiment. Whether the data behind it is fresh enough to answer with. And whether an agent gets all five signals in the call that returns the number.

What is Honeydew?

Honeydew is a semantic layer for AI and BI. Business logic is modeled as entities, attributes and metrics, and a semantic compiler turns every query, metric and agent instruction into SQL executed inside the customer’s warehouse, across Snowflake, Databricks and Google BigQuery.^[1] Snowflake Ventures invested in May 2025, describing it as a bridge between the business language BI tools and agents use and the join paths, granularity and metric definitions underneath.^[6]

Its change-control story is stronger than most tools in the category. Each entity, attribute, metric, dataset and domain is its own YAML file in a Git repository, all major Git providers are supported, and publishing a change opens a pull or merge request against the main branch. Honeydew’s documentation recommends branch protection rules so production changes require review.^[2] That is real governance of the authoring workflow, the same discipline context versioning for AI agents argues for.

It also treats other semantic surfaces as sources rather than rivals. Honeydew describes a Snowflake Semantic View as “a singular schema-level object for a specific purpose” and itself as an organization-wide layer over multiple models and semantic views, with support for consuming those views in private preview at the time of writing.^[5] Teams weighing Snowflake Semantic Views or Unity Catalog Metrics against a standalone product are picking a shape for the same modeling decision.

Core components of Honeydew

Semantic compiler: turns queries, metrics and agent instructions into warehouse SQL
Entities, attributes and metrics: the modeling primitives, each stored as an individual YAML file
Git version control: publishing opens a pull request; all major Git providers supported
Interfaces: SQL, XMLA, GraphQL and MCP, plus Power BI, Tableau and Excel
Security posture: SOC 2 Type II, warehouse role and row-level security, SSO via Google or Microsoft^[3]

What is semantic layer governance?

Semantic layer governance decides, records and enforces which business definitions are authoritative across an estate, and attaches enough evidence to each one that a person or an agent can judge whether to act on it. It is adjacent to metadata management for AI and overlaps with data contracts for AI, but its unit is the definition, not the dataset.

The reason it is a separate job is arithmetic. A large enterprise has no single place where “revenue” is defined. It has a warehouse view, a transformation model, three BI measures written by three teams, a finance spreadsheet older than all of them, and now a semantic model. Enterprise context silos form exactly there, which is why self-service analytics governance is its own build-or-buy decision.

Agents raise the stakes because they do not hesitate. An analyst who sees two numbers goes and asks someone. An agent queries whichever definition its tool exposes and answers with the same confidence either way, the failure mode AI agent accuracy work keeps landing on and the reason text-to-SQL for enterprise breaks without governed definitions underneath.

What the governance job has to answer

Authorship: who may create or change a definition, and through what review
Authority: which of several definitions to use, recorded rather than assumed
Dependency: what the metric reads from, traced to columns, as automated SQL lineage does it
Status: certified, deprecated or draft, visible to whoever asks
Trust at query time: freshness and quality returned with the answer, not looked up after

WTF Is the Context Layer?

The plain-language ebook on what sits above your semantic models, and why agents need it before they need a better model.

Get the 5x Accuracy Ebook

Honeydew and a governance layer: head-to-head

The comparison that helps is scope against scope, not capability against capability. The two layers are accountable for different failures.

Dimension	Honeydew	Governance layer (Atlan)
Primary focus	Modeling and computing business logic correctly	Deciding which logic is authoritative and whether it is safe to use
Boundary	The semantic model and the warehouses it runs on	The estate, including definitions written in BI tools and pipelines
Key stakeholder	Analytics engineering	Governance, analytics engineering and AI platform teams jointly
Change control	Git pull request on the YAML definition	Ownership, certification and policy recorded on the definition
Delivery to agents	MCP, SQL, XMLA and GraphQL from the compiled model	MCP server carrying definition, owner, lineage and trust status
Measured by	Query correctness, consistency across BI and AI	Share of definitions certified, owned and traceable to source
Failure mode	A correct model that a second team’s dashboard contradicts	Governance metadata with no real model to attach it to

Each column does something the other does not. Teams comparing options inside the left column will get more from Lightdash vs dbt Semantic Layer or Cube than from this page. Teams sizing the right column want metadata tooling build vs buy evaluation criteria.

Do you need both a semantic layer and a governance layer?

Most enterprises run both, for the same reason they run a database and a catalog. One holds the logic; the other holds the record of what it is for and who stands behind it.

How they work together

The division is clean. The semantic layer owns the definition and its execution, so a metric computes the same way whether the request comes from Power BI, a notebook or an agent. The governance layer owns what a consumer needs in order to decide whether to use it: the owner, the certification state, the lineage back to physical columns, and the freshness of the tables underneath, where data observability for AI pipelines feeds in.

Honeydew compiles each MCP request into explainable steps and optimized SQL, with every action passing through the compiler so it follows the modeled business rules and access controls.^[4] That handles correctness within the model. A governance layer adds the surrounding judgment on the same call: this definition is the certified one, this person owns it, it reads these columns, the source table last loaded four hours ago. MCP delivering business context is the mechanism, and when to use MCP vs an API covers the plumbing choice.

The same layering shows up in adjacent stacks: Genie ontology and the Atlan context layer, and OpenAI Frontier vs semantic layers, both land on needing both.

When is Honeydew alone the right answer?

Often enough that this deserves a straight answer rather than a hedge.

Honeydew alone is the right call when the semantic model is genuinely the only place metrics are defined, one team owns authoring, the warehouses in scope are Snowflake, Databricks or BigQuery, and Git review is the change control the organization needs. Pizza Hut’s reported results, 50% faster development of insights and a 30% cut in cost and engineering effort, came from that kind of focused deployment.^[6] Adding a governance layer to one well-run model buys very little.

Add the second layer when definitions already sit in BI measures and transformation code no single model will absorb, when an auditor asks who approved a number, when agents from several platforms query the same metrics, or when “is this figure current” needs a person to answer. That is the threshold context layer for data governance teams describes, and context layer role-based access control covers the access half.

Plan for both from the start when the program is federated or data mesh shaped, where domains model independently and something has to arbitrate later.

Find the gap before an agent does

Score how much of your metric estate has an owner, a certification state and lineage an agent could actually read.

Run the Gap Calculator

How Atlan approaches semantic layer governance

Atlan’s position is that a definition and the evidence around it should travel together. The Enterprise Data Graph connects glossary terms, ownership, certification state, policy and column-level lineage into one graph over whatever holds the logic: a semantic layer product, a warehouse object or transformation code. The context layer is the name for that job, and data catalog vs context layer explains why a catalog built for human browsing is different.

Delivery is the part that matters for agents. Atlan’s MCP server returns the governed definition together with its owner, its status and where it came from, so the receiving agent has grounds to act or to stop. In Atlan’s AI Labs benchmark, adding that layer improved AI’s text-to-SQL accuracy by 38%, a first-party result measured on context rather than model choice. The underlying claim is the one systems of semantics makes: semantics is becoming its own enterprise system, not a BI feature.

None of that models or computes a metric. That is the semantic layer’s work, and a good semantic layer makes governance easier by offering one well-formed definition to certify instead of six informal ones to reconcile. Teams starting from zero should read types of metadata for AI agents and what an AI context platform does before buying either layer.

Real stories from real customers: One vocabulary, many tools

"Atlan captures Workday's shared language to be leveraged by AI via its MCP server. As part of Atlan's AI labs, we're co-building the semantic layer that AI needs."

— Joe DosSantos, VP Enterprise Data & Analytics, Workday

Watch Now →

"Atlan is much more than a catalog of catalogs. It's more of a context operating system…Atlan enabled us to easily activate metadata for everything from discovery in the marketplace to AI governance to data quality to an MCP server delivering context to AI models."

— Sridher Arumugham, Chief Data & Analytics Officer, DigiKey

Watch Now →

Neither company has published anything about its semantic layer vendor, and no case study here names Honeydew. Both appear because they describe the second job in their own words: a shared language captured once, then delivered to agents with governance attached. Context layer for data analytics teams documents that pattern at length.

See the layer working on real definitions

A live walkthrough of governed context reaching an agent: the definition, its owner, its lineage and its trust status in one call.

Watch the Live Demos

The question a metric definition cannot answer by itself

A well-modeled metric tells you what a number means and how it was computed. It cannot tell you that another team shipped a conflicting version last quarter, that the table underneath has not loaded since Tuesday, or that the person who owns the definition left in March. Those are facts about the estate, not facts inside the model, and an agent needs them before it puts a figure in front of an executive. Teams getting agents into production without incident have answered both questions, not the modeling one twice.

Book a Demo

FAQs about Honeydew and semantic layer governance

What is Honeydew?
Honeydew is a semantic layer. Entities, attributes and metrics are defined once, stored as YAML under Git version control, and compiled into SQL that runs inside Snowflake, Databricks or Google BigQuery. It serves those definitions to BI tools and AI agents through SQL, XMLA, GraphQL and MCP.
Is a semantic layer the same thing as semantic layer governance?
No. A semantic layer is where a definition is written and computed. Governance is the decisions around it: who may author it, whose version is authoritative when two exist in different tools, what it depends on down to the column, and whether it is certified or a draft.
Does Honeydew have version control for metric definitions?
Yes. Honeydew stores each entity, attribute, metric, dataset and domain as a separate YAML file in a Git repository, and supports all major Git providers. Publishing a change automatically opens a pull or merge request, and branch protection rules can require review.
When is Honeydew alone the right answer?
When the semantic model is the only place metrics are defined, one team owns authoring, the warehouses in scope are ones it supports, and Git review is the level of change control the organization needs. That is a real and common situation, and adding a second layer to it buys very little.
What does a governance layer add on top of a semantic layer?
Reach and trust status. It connects definitions living outside any one model, in BI measures, transformation code and warehouse objects, records ownership and certification on each, traces a metric to the columns it reads, and delivers those signals alongside the number.
Can you run Honeydew and Atlan together?
Yes, and that is the common enterprise pattern. The semantic layer holds and computes the definition; the context layer records who owns it, whether it is certified, what it depends on and whether the data is current, then delivers all of it through an MCP server.

Sources

Share this article

Atlan is the Context Layer for AI. It translates business knowledge, including data definitions, working procedures, and governance policies, into context AI can actually use. This knowledge lives in a single Enterprise Data Graph that every team and AI agent can reach.

In Atlan's AI Labs benchmark, adding this context improved AI's text-to-SQL accuracy by 38%.

Atlan is recognized as a Leader across multiple Gartner reports and Forrester Waves, and is trusted by over 400 enterprises representing $10T+ in market cap, including Mastercard, Workday, General Motors, CME Group, HubSpot, FOX, Virgin Media O2, and Elastic.

Get a Custom Demo See How It Works