Atlan

Atlan is a context layer for AI. It catalogs the assets across your data landscape — along with their glossary terms, lineage, and documentation — and makes that context available to AI tools and agents. The integration works in both directions. Honeydew publishes its semantic definitions into Atlan, and Honeydew’s AI can read governed definitions back out of Atlan as context.

​

Two Context Layers

Honeydew and Atlan are both context layers, at different levels of the stack: The two are complementary rather than competing. Atlan describes assets across the estate; Honeydew defines the logic that turns them into governed metrics, and generates the SQL. Publishing Honeydew definitions into Atlan makes them discoverable next to everything else, and reading Atlan definitions into Honeydew avoids restating business knowledge that is already governed.

​

What Gets Published

Each workspace branch becomes one semantic model in Atlan, holding the entities and fields it defines: Every asset carries its display name, description, and business metadata — folder, labels, and tags, published together as the asset’s labels. Attributes and metrics also carry their SQL expression, data type, and any synonyms and sample values curated for the AI. An asset is addressed by a qualified name built from the assets above it in the hierarchy, in the form <connection>/<model>/<entity>/<field>. Honeydew also writes lineage from the warehouse columns each field reads and, when you publish into an Atlan domain, a subdomain for the model with a data product per Honeydew domain. Honeydew writes only in its own connection and the subdomains it creates, and does not modify warehouse assets or attributes curated in Atlan, such as owners, certificates, and tags.

​

Setup

Create an Atlan API token, then configure the integration in Honeydew.

​

Create an Atlan API token

In Atlan:

1. From the left menu, click Admin, then under Workspace, click API tokens (admin access is required).
2. Click Generate new and select Generate API token.
3. Enter a name for the token, such as honeydew, and an optional description.
4. Copy the generated token.

Atlan grants API access and asset access separately, so give the token personas with:

• Honeydew connection — read, create, update, and delete
• Warehouse connection — read, for lineage
• Parent domain — create subdomains, and create, update, and delete data products, when publishing into a domain

See Atlan’s API tokens documentation for token management, rotation, and revocation.

​

Configure Honeydew

1. Go to Settings → Semantic integrations.
2. Under Add new integration, select Atlan.
3. Enter a Connection identifier — a name for this credential set, for example atlan_prod.
4. Enter the Tenant base URL of your Atlan tenant, in the form https://<tenant>.atlan.com.
5. Paste the API token and save.

​

Publish a Workspace

Publishing requires the Editor role.

1. Click the branch, and in the menu click Branch integrations → Atlan → Publish to Atlan…
2. Select the Integration connection — the Atlan credential set Honeydew publishes with (only selectable when more than one exists).
3. Select the Connection to publish into — the Atlan connection that holds the assets.
4. Select the Warehouse connection — the Atlan connection the source tables were crawled into, which is what lineage is drawn against. Leave it empty to publish without lineage.
5. Optionally, select an Atlan domain to add the model to a subdomain under it.
6. Set the Model name, or select an existing model to republish into it. The default is the workspace name, with the branch appended.
7. Click Publish, then open the published model from the link in the result.

The result reports how many assets were pushed, archived, and linked. It also reports source columns that no crawled column matched: those fields are published without lineage, so a count above zero usually means the warehouse connection is the wrong one, the token cannot read it, or the tables have not been crawled yet.

​

Where Lineage Appears

Open an attribute or a metric in Atlan and select Lineage: it appears downstream of the warehouse columns it reads. A calculated field resolves through the fields it depends on, down to the physical columns underneath.

​

Republishing

Publishing again updates the model in place — assets are keyed on their qualified name, so republishing an unchanged workspace changes nothing. A field deleted in Honeydew is archived in Atlan on the next publish, and stays recoverable. Changing the model name publishes a second model rather than renaming the first, since the name is part of every asset’s identity. The data product of a deleted Honeydew domain is archived on the next publish. Publishing under a different Atlan domain moves the data products to a new subdomain and leaves the old subdomain in place. Publishing with no Atlan domain leaves the subdomain and its data products as they are.

​

Atlan as a Context Source

Atlan is also supported as an External Knowledge source, so the AI can fetch governed definitions and glossary terms from Atlan over MCP when answering questions. Use this to reuse business knowledge that Atlan already governs — glossary terms, policies, and asset documentation — instead of restating it as Honeydew context. Definitions stay owned in one place, and Honeydew reads the current version at query time. This is configured separately from the catalog integration above:

1. Go to Settings → MCP connections and connect Atlan. Each user authenticates individually, so content is fetched with their own permissions.
2. Create an External Knowledge context item whose external_source.tool is an Atlan tool, such as mcp__atlan__get_assets, and whose resource_id is the identifier of the asset or glossary term to fetch.

See External Knowledge for the full field reference.