Product

A specification the platform is actually built from.

Wikis go stale, spreadsheets get forgotten, and the real definitions end up in people's heads. Specly keeps the model, the documentation and the intent in one place – versioned, searchable and shipped with the code.

How it fits together

The bridge between people and the code

Analyst
Modeler
Developer
Business

Specly

one shared model: tables, data products, terms, owners and status

Git
changes both ways, always reviewed
Data platformDatabricks, Microsoft Fabric, dbt
read for import and validation, never written to
AI agents
read the model over MCP
Consumers
data contracts in ODCS

Shared understanding

One model, one shared language.

From conversation to diagram – start where the business is, and end with facts, dimensions and relations everyone recognises.
The same term everywhere – the business glossary gives each term one definition and links it to the columns that mean the same thing.
Consistent by construction – naming standards, data types and required documentation are enforced as rules, not as review comments.
Work together in real time – when several people work on the same branch, you see who is there, each other's pointers in the diagram, and who is editing what, as it happens.

Specly diagram view with fact and dimension tables connected by relations

Overview

The questions your platform can finally answer.

What is happening now – see what is live, what is on its way to production, and which tables are being changed in more than one place at once.
Search across branches – find data products, tables, columns and terms, including what so far exists only in a branch.
Who to ask – every table and data product has an owner, so a question takes one click – not three chat threads.

The Specly overview with key figures, a search field and the questions the platform can answer

Build with AI

Knowledge your agents can build from.

In Git – every table is a file in the repo, and changes flow both ways with review – you own the access and choose the repos.
As a data contract – tables are delivered as data contracts in ODCS and data products in ODPS, so the people who use the data know what they can rely on.
Validate what was built – compare the platform with the design and see where they disagree. Validation suggests a status but never sets one itself.
For your agents – AI agents read the model over MCP and build from what you have agreed on, not from guesswork.