Your company’s second brain.
Doram learns how your business works, so every AI tool gives your team the same trusted answer.
Ask in plain words. Doram pulls context from the second brain and shows each step.
Your company’s context is everywhere.
Spread across your databases, Slack threads, old docs, spreadsheets and people’s heads. Your AI sees the data, but not what it means.
What’s our churn rate?
6.8% in September
Churn: no order in 60 days · guessed from the tables
4.2% in September 6.8%
Churn: no order in 90 days · approved by the CFO
Slack#finance-opsRohan 11:42 AM
Reminder: test orders never count in sales.
NotionFinance / GlossaryNet sales
SheetsQ3 close“We report quarters in IST. Everyone knows that.”
Said in a meeting. Not written down.
dbtmodels / fct_orders.sql
GmailInboxChurn is no order in 90 days, not 60.
LookerSales overview
JiraDATA-412Who owns the region mapping?
DrivePricing policy v3.pdfB2B discounts are booked net of tax.
Page 14
DocsQBR notesActive means paid in the last 30 days.
AAna shared this with SalesDoram binds it into one place.
Net sales excludes returns
Test orders never count
Active: paid in the last 30 days
Churn: no order in 90 days
Order value minus refunds
PQuarters by order date, IST
B2B discounts net of tax
Internal orders left out
What’s our churn rate?
6.8% in September
Churn: no order in 60 days · guessed from the tables
4.2% in September 6.8%
Churn: no order in 90 days · approved by the CFO
P0/8 rulesLunch at 1?Test orders never countRe: Fwd: invoice copyFix the checkout bugInternal orders left outChurn: no order in 90 daysUntitled dashboard 4Order value minus refundsOffsite agenda, draftCopy of targets v3Net sales excludes returnsActive: paid in the last 30 daysMeeting notes, 12 Sepstg_orders_oldB2B discounts net of taxDoram binds it into one place.
Net sales excludes returns
Test orders never count
Active: paid in the last 30 days
Churn: no order in 90 days
Order value minus refunds
PQuarters by order date, IST
B2B discounts net of tax
Internal orders left out
Three layers. One trusted answer.
What it means, how it connects, and how it is counted. Every answer Doram gives reads all three, so a number always traces back to a rule.
Ontology
Turns raw columns into the words your team uses.
Context graph
How things link up, and the reason behind each rule.
Semantic layer
One approved formula for every number.
What changes when your AI knows your business.
Ask AI, not the data team.
Right the first time.
Same question, same answer.
Every rule you approve stays approved.
How Doram compares.
How Doram differs from Cube, dbt Semantic Layer, Looker, Honeydew, Omni, ThoughtSpot, Zenlytic, Veezoo, Tellius, WisdomAI, ColRows and Atlan: where the meaning comes from, what happens to your documents, and how each answer shows its work.
| Capability | Semantic layers, AI analysts and context layers | Doram |
|---|---|---|
| Shared company brain | A model or catalog in each tool, shared with agents through MCP | One governed brain that every AI agent reads: approved rules, meaning and reasons |
| Semantic discovery | Written by hand, or drafted by AI from your tables or catalog | Finds metrics, dimensions, joins, grain and rules from your schema, data profiles and query history |
| Context ingestion | Documents left out, read while answering, or used to draft the model | Docs, wikis and Slack, turned into rules your team approves |
| Business ontology | Metrics, joins and tables, or a catalog of domains, owners and lineage | Departments, teams, owners and KPIs, with logic linked across teams, for example spend to revenue to profit |
| Rule rationale | In descriptions, notes or wikis, if someone writes them | Stored with each rule, and shown with each answer |
| Semantic compilation | Metric queries compiled to SQL, or SQL written by the AI. Tests written as code | Compiles each question at run time against approved rules. Each new rule is replayed on past queries first |
| Reviews and conflicts | Code review in Git, admin edits, or owner sign-off. Conflicts left to the person asking | Each rule is approved before any AI uses it. Conflicts and missing context go to review, and Doram asks instead of guessing |
| Compounding and drift | Grows when someone edits the model. Changes found when a query fails | Each question and correction becomes a reviewed rule. Schema and business changes get a drafted fix |
| Explainability | Shows the SQL, a chart, and sometimes its reasoning | Each step: the rule, the formula and who approved it |
- Cube
A semantic layer written in YAML or JavaScript vs. context learned from your data and documents.
Semantic layersCompare → - dbt Semantic Layer
Metrics written in YAML inside a dbt project vs. context learned from your data and documents.
Semantic layers - Looker
A BI platform modeled in LookML vs. context learned from your data and documents.
Semantic layers - Honeydew
A YAML semantic layer that looks up wiki pages vs. one that turns those pages into rules.
Semantic layersCompare → - Omni
A BI model that learns from chats vs. one that also learns from your documents.
AI analystsCompare → - ThoughtSpot
Reads Slack and Confluence while it answers vs. turns them into rules your team approves.
AI analysts - Zenlytic
An AI analyst on a YAML model in Git vs. context learned from your wikis and documents.
AI analystsCompare → - Veezoo
A Knowledge Graph built from your tables vs. context learned from tables and documents.
AI analystsCompare → - Tellius
An agent that models your warehouse for you vs. context learned from your data and documents.
AI analystsCompare → - WisdomAI
Context that sets itself up vs. context your team reviews and approves.
AI analystsCompare → - ColRows
A semantic layer that builds itself vs. rules your team approves, each with its reason.
Context layersCompare → - Atlan
A data catalog that grew a context layer vs. a context layer built to answer questions.
Context layersCompare →
Questions, answered.
Still unsure? Ask the playground →
How is Doram different from a semantic layer?
A semantic layer holds formulas that someone writes by hand. Doram holds the formulas, what each column means, and the reason behind each rule. It drafts all three from your data and documents, and your team approves them.
Does my data leave my database?
No. Doram is read-only. It reads your schema and runs queries you can see, and you can run it in our cloud or in yours.
Do we have to write the rules by hand?
No. Doram drafts each rule from your database, your documents and the questions people ask. A person on your team reviews it before any AI uses it.
Can we use it in the AI tools we already have?
Yes. Claude, ChatGPT and Cursor connect through MCP, and your own agents connect through the API. They all read the same context.
What does it connect to?
Your databases, your CRM and ERP, your docs and wikis, and Slack and email. Each source adds to one context layer.
How do I know an answer is right?
Every answer shows its steps: the rule it used, the formula, and who approved it. You can follow any number back to where it came from.
Ask it yourself.
The playground is a whole company, already connected.