Doram vs Hex
Hex is a workspace for data teams, with notebooks and chat. Doram is the layer of approved rules that every AI reads.
DoramChecked on 17 October 2026. Every fact about Hex comes from its public website, docs and changelog. The links are at the end.
The short answer
Hex is a workspace for data work. It has SQL and Python notebooks, apps, dashboards, a chat called Threads and a Slack bot. Its agent writes SQL itself, and it is told to use your semantic models and approved tables first.
Doram is the layer under that kind of work. It learns the rules from your data, documents and tools, your team approves them, and a compiler writes the SQL. Its own analyst and every other AI read the same rules.
At a glance
| Question | Hex | Doram |
|---|---|---|
| What it is | A data workspace with notebooks, apps and an AI agent | A context layer with its own AI analyst |
| Who builds the model | Admins write YAML, with an AI helper. It can import dbt, Cube and Snowflake models. | Doram drafts it. Your team approves what matters. |
| What it reads | Schema, dbt Cloud, guides you write, code repos, Notion, Linear, Slack threads | Schema, query history, dashboards, docs, Slack, Jira, CRM |
| How the AI gets the number | The AI writes SQL. It prefers modeled and approved data. | The AI picks approved facts. A semantic compiler writes the SQL. |
| Review before changes go live | A preview and checks before publish. No approval step. | Review by risk, with the evidence and impact of each change |
| Other AI tools | MCP that starts a Hex chat, plus an API | MCP and API that read the model itself |
| Where it runs | Hex cloud, or a single-tenant cloud run by Hex | Doram Cloud, or your own cloud account |
What Hex does well
- One place for data work. SQL and Python notebooks, drag-and-drop exploring, apps, dashboards, chat and Slack all live together. Any chat can be opened as a notebook to check the logic.
- Fast-moving agent tools. Tests for the agent, an “approved data only” mode, written guides, and suggestions drawn from past chats. Its 2026 changelog shows a release almost every week.
- A flexible semantic layer. Write your own models in YAML with an AI helper, or import them from dbt MetricFlow, Cube or Snowflake. When you explore a model, Hex writes the SQL and checks joins for double counting.
Where they differ
How the number is made
In Hex chat, the agent writes the SQL. Its docs say it “heavily prioritizes semantic models and endorsed data”, and falls back to raw tables when it has no good modeled data. An admin can turn on a mode that limits it to approved data only.
In Doram, the AI never writes the SQL. It picks approved facts, and the semantic compiler writes the query. The same question gives the same answer, and any past answer can be replayed on the version that made it.
Where the rules come from
In Hex, people write the model and the guides. Its review agent reads past chats and drafts updates to the guides, and the data team publishes them. The agent can also read Notion, Linear and Slack threads while it answers.
Doram builds the model from what it reads. Query history shows which filter every finance query adds, and a Jira ticket often holds the reason behind a rule. Doram turns these into rules, with their evidence, for your team to review.
How changes are approved
Hex shows a preview of each model change, with checks, before you publish. As far as its docs show, there is no step where a second person must approve it.
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.
What other AI tools get
Hex’s MCP server lets Claude, Cursor or ChatGPT start a Hex chat and read the answer. We found no MCP tool that queries the semantic model directly.
Doram serves the context layer itself. Claude, Cursor and ChatGPT read the same approved definitions through MCP, and your own agents read them through the API.
Who should pick which
Pick Hex if your data team lives in notebooks, you want SQL, Python, apps and chat in one tool, and public per-seat pricing matters to you.
Pick Doram if you want SQL compiled only from approved rules, rules found for you with their reasons, and every AI tool reading those rules directly.