Doram vs Cube
Cube is an open-source semantic layer that serves one model to every tool. Doram learns that model from how your company works.
DoramChecked on 17 October 2026. Every fact about Cube comes from its public website, docs and GitHub. The links are at the end.
The short answer
Cube is an open-source semantic layer. Your data team writes the model as code, and Cube serves that one model to almost any tool: BI tools, spreadsheets, APIs and AI agents. It now has its own AI analyst on top.
Doram starts from the other end. It learns the model from how your company already works, keeps the evidence for each rule, and asks your team to approve what matters. Both keep the AI away from raw tables.
At a glance
| Question | Cube | Doram |
|---|---|---|
| What it is | An open-source semantic layer with an AI analyst | A context layer with its own AI analyst |
| Who builds the model | Your data team, in YAML, JavaScript or Python. AI can draft the first version. | Doram drafts it. Your team approves what matters. |
| What it reads | Schema, dbt, LookML, Snowflake semantic views | Schema, query history, dashboards, docs, Slack, Jira, CRM |
| How the AI gets the number | The AI writes SQL against Cube’s views. Cube writes the real SQL. | The AI picks approved facts. A semantic compiler writes the SQL. |
| Review before changes go live | Git branches and pull requests | Review by risk, with the evidence and impact of each change |
| Other AI tools | MCP, SQL, REST, GraphQL, DAX, Excel, Sheets | MCP and API |
| Where it runs | Cube Cloud, your cloud account, or self-hosted open source | Doram Cloud, or your own cloud account |
What Cube does well
- Open source. The core is free under Apache 2.0, and you can run it yourself with Docker. The paid product can also run in your own AWS, Azure or GCP account.
- It reaches every tool. One model is served over a SQL API, REST, GraphQL, DAX, MCP, Excel and Google Sheets. The same access rules apply on every path.
- Everything as code. The model, the agent rules and the trusted example queries all live in Git. Row rules and column masking sit in the model too.
Where they differ
Who writes the model
In Cube, the model is code that your data team owns. AI helps: it can draft the first model from your tables, turn LookML into Cube code, and pull in your dbt project. After that, a person writes and keeps up every change.
As far as its docs show, Cube does not build the model from query history, documents, Slack or tickets. Its agent keeps memories from past chats, but those memories do not go back into the model.
Doram reads those places on purpose. Query history shows which filter every finance query adds. A Jira ticket or a Slack thread often holds the reason behind a rule. Doram turns these into rules, with their evidence, for your team to review.
How the number is made
Both tools keep the AI off the raw tables. In Cube, the AI writes a query in Cube’s own SQL against governed views, and Cube turns it into the warehouse SQL. Cube can also give the AI trusted example queries to copy or adapt.
In Doram, the AI does not write a query at all. It picks approved facts, and the semantic compiler writes the SQL. The same question gives the same answer, and any past answer can be replayed on the version that made it.
How changes are approved
Cube treats the model like software. Edits happen on a branch, and someone merges them, often through a pull request. Changes made by AI over MCP also land on a branch first.
Doram routes each change by risk. A new word for an existing metric goes live on its own. A change to how net sales is counted waits for its owner. Each suggestion shows its evidence and every metric it would change.
Accuracy claims
Cube published a test in April 2026 where adding a short hand-written document lifted three AI models from about 50% to about 68%. That test measured a document in the prompt, not Cube’s own product. We found no accuracy number for Cube’s current agent. Doram does not publish an accuracy number yet.
Who should pick which
Pick Cube if you want open source you can host yourself, your data team wants to own the model as code, or you need one model to feed many BI tools, spreadsheets and APIs.
Pick Doram if nobody has time to write the whole model, your rules live in queries, documents and tools, and you want each rule kept with its reason and approved by its owner.