Describe your use case in Context Engineering Studio, approve the assets CES proposes, and review the semantic model it assembles. The repository references your Atlan assets instead of copying them.
Create your context repository
Describe your use case in Context Engineering Studio (CES). CES searches your Atlan data graph, proposes a set of assets, and waits. Once you approve the selection, it assembles a semantic model from those assets.
The repository references your Atlan assets instead of copying them, so every definition traces back to its governed source in Atlan. CES builds the semantic model from those assets; when a source changes, regenerate the repository to pull the update.
Before you begin
- Context Engineering Studio is enabled on your workspace and your team has the CES persona. See Enable Context Engineering Studio.
- You have a use case to start with: the specific job one agent has to do, such as answering pipeline questions for the sales ops team. Scope it to that job rather than to a team or a domain. A good first repository covers a single dashboard or a small set of closely related datasets.
Step 1: Describe your use case
In Context Engineering Studio, describe the repository you want in the prompt box.
To start from a common starting point instead, pick one of the suggestions under Create a context repository for:
- Sales metrics and revenue analysis
- Marketing campaign performance
- Customer health and support analysis
- Financial reporting and budget analysis
The more specific your prompt, the better the result. A good prompt covers the questions the agent must answer, who asks them, and the tables to include or exclude.
Example 1, broad:
Create context for an agent that answers pipeline questions for sales ops.
Example 2, specific:
Create context for a finance analyst to answer revenue and budget questions
using the invoice, budget_actuals, and cost_center tables. Exclude
staging tables.
Step 2: Approve your asset selection
The build stops here until you decide.
Two tabs show what CES chose:
- Selected assets lists what goes into the repository. The Rationale column explains each choice and names the signals behind it, such as verification status, row count, column count, and owner.
- Not selected lists the assets it considered and left out.
Then choose one:
- View / refine selection to add or remove assets before the build starts.
- Proceed with selection to build with what CES proposed.
Step 3: Review your repository
The repository opens in the right pane. Overview shows the model and its linked assets, Activity shows what changed upstream since the last build, and Portability is where you deploy.
Overview has two cards:
- Semantic model is the canonical definition for this use case. It lists your tables with their dimensions, time dimensions, and facts, plus the derived metrics built on top of them.
- Linked assets lists every Atlan asset the repository is built from. Each one carries a Reference badge: the repository links to the asset in Atlan for provenance rather than storing a copy. Regenerate to pull the current state.
If the derived metrics count carries a not emitted badge, the model held some metrics back. See Metrics that aren't emitted for why, and where those metrics are still available.
Step 4: Update the context repository
The semantic model draws from two kinds of fields.
- Source-owned fields, such as descriptions, come from the linked Atlan asset and show a pencil icon. Select one to open that asset's preview drawer and edit it at the source. The edit follows the asset's normal path, including any governance workflow configured on it, and the repository doesn't change until you regenerate.
- Repo-defined fields, such as names, expressions, relationships, and derived metrics, are edited directly in the model.
To change what the repository covers, use Edit linked assets in the footer of the semantic model view (add or remove an asset, or bring in a new column), then click Regenerate. Regenerating pulls the current state of each linked asset from Atlan and rebuilds the model, keeping the custom instructions you've written.
To add a dimension, fact, or time dimension, use the + control next to that section. It opens the linked table's column picker scoped to that kind; columns already modeled elsewhere are locked so you can't add them twice.
To edit relationships, use the relationships editor. Table and column selectors are scoped to what's in the model.
For business rules that aren't in the data, such as which source to prefer or what a fiscal quarter means, edit Custom instructions on the Semantic model card. Regenerating preserves them.
For mined context that's wrong or thin, such as a join, filter, or business question, fix it on the asset's Usage & Intelligence tab, then regenerate. The correction reaches every repository built from that asset.
To describe a change instead, use Ask AI or the chat window.
When upstream assets change
A banner at the top of the semantic model shows when a linked asset has changed since the model was last built: a description edited at the source, a column added or removed, a glossary term updated. The banner shows a count; select View changes for a breakdown grouped by asset, with who made each change and what changed. Routine metadata such as owners, certificates, popularity, and usage statistics doesn't trigger it.
Select Regenerate from the banner or the model header to rebuild the model from the current state of its linked assets. The Activity tab keeps the full history. See What's in a context repository.
Step 5: Activate your repository
A new repository starts in Draft. When the model looks right, activate it. CES offers Activate this repo (DRAFT → ACTIVE) as a follow-up, or you can ask for it in the chat window.
Active repositories appear under Explore context repositories on the Context Engineering Studio home page.
Next steps
- Deploy to Snowflake: push your context repository to Snowflake Cortex Analyst.
- Deploy to Databricks: push your context repository to Databricks Genie.