Model One, month by month

October 2026

So far in October: bringing your own AI agent into a project, holding every agent to the app's rules, and task badges on the model.

• Connect Claude Desktop or your own AI agent

  Anyone in a project, Viewers included, can connect Claude Desktop or another MCP client with a personal access token that acts as them, with their role, in that project only. Tokens can be read-only and revoked, and their changes are audited. OneAssist →

• AI agents held to the app's own rules

  OneAssist and external agents can only make the changes the app itself allows, and no longer overwrite a colleague's newer save. Model One's server now enforces OneAssist's read-only Ask and Plan modes. OneAssist →

• See the open work linked to each object

  Entities, relationships, graphs, Flows, terms and physical models carry a badge counting their open work, and a concept, submodel or folder adds up what it holds. A click lists that work in Tasks, whose filters are remembered per project. Tasks →

• Foreign keys take their referenced column's type

  Give a column a physical type, and every foreign key that references it takes the same type in the DDL, checks, migrations and dbt export. OneAssist can align the foreign keys a model already holds. Physical model →

• Each kind of Flow step takes the connections it can use

  A step takes only as many inputs and outputs as its kind can use, so a Flow cannot hold a connection that produces no sensible dbt model, such as a second input into a Merge. A refused line turns red as you draw it. Flow →

• dbt projects declare the physical model's types

  An exported dbt project declares each column with the type the chosen physical model's script gives it, and its tests also work on columns named after reserved words, such as Key. Flow →

Also so far in October: a new logo and typeface, and about 30 smaller improvements and fixes, mostly to the canvases, shared editing and OneAssist.

September 2026

September brought history-keeping loads to Flow, Fabric Data Warehouse as a database target, concepts for reading large models, and graph model exports.

• SCD2 dimensions and hash keys in Flow

  A new Historize (SCD2) step loads dimensions that keep their history, set up by naming roles rather than by writing SQL, and exports to dbt. A new Hash step builds Data Vault keys and change hashes from a column picker. Flow →

• Fabric Data Warehouse as a database target

  Choose Fabric Data Warehouse for a physical model and its script runs as generated, while validation reports any setting the warehouse cannot hold. The dbt export follows the physical model, building a Fabric or SQL Server project that can also read tables already in the warehouse. Physical model →

• Check a Flow as you design it, and choose what to export

  A banner above the Flow canvas lists what Model One's checks find, and clicking a finding selects what it names. Exporting to dbt opens a dialog for choosing the Flows and the physical model, and it shows what blocks the export before it runs. Flow →

• SQL DDL import keeps the types your script declares

  Importing SQL DDL can now build the physical model too, keeping the types the script declares, its defaults, auto-numbering, computed columns, CHECK constraints and schema. It says which database the script came from and why, and reports anything it leaves out. Physical model →

• Read a large model as a dozen business concepts

  A concept is one box that stands in for a whole submodel on Full Model, so a model of hundreds of entities reads as a dozen named ideas. A model too large to draw in full now shows its concepts instead of a placeholder. Concept model →

• Export graph models and build supertype hierarchies

  Export a graph model as GraphML, SQL/PGQ, a Neo4j Cypher script or a Microsoft Fabric graph model. A node type can now be a subtype of another, an edge drawn once at a supertype exports for each subtype, and Focus dims all but the part you are reading. Graph models →

Also in September: OneAssist began answering how-to questions from the user manual, and about 200 smaller improvements and fixes, mostly to Flow and its dbt export, to editing alongside teammates and OneAssist, and to versions and the audit log.

August 2026

Public launch on 13 August 2026

Migration scripts, graph models, naming standards and an audit log arrived, along with steady work on reliable saving and shared editing.

• Deployable DDL, and migrations after go-live

  Generated DDL gained indexes, CHECK constraints, foreign-key actions and your own SQL. For a database already in production, Model One plans the migration from a saved version, with data-losing steps commented out until you confirm, and records which version each environment runs. Physical model →

• Graph models over your logical model

  Design node and edge types over the entities, keys and foreign keys you already model, so there is no second model to keep in step. Derive a graph from the model, or sketch one and turn it into entities and foreign keys. Graph models →

• Naming standards and approved abbreviations

  A Naming tab sets the rules a physical model's names follow, from casing to constraint names, and can require the dictionary's approved abbreviations. Renames are previewed first, and a deliberate exception can be waived with a reason. Physical model →

• An audit log of who changed what, and when

  Changes across the logical model, dictionary, Flow, physical models, sources and tasks are recorded field by field, with the values before and after. OneAssist's changes name the person who asked, and Owners can hide a sensitive value, giving a reason, and show it again. Versions & audit →

• Edit one Flow together, and watch teammates drag live

  Elements, connections and notes in a Flow save one by one, so colleagues can edit the same diagram without overwriting each other, and on every canvas you see a teammate's drag as it happens. Flow →

• Ask OneAssist without it changing anything

  A new Ask mode, beside Plan and Agent, answers questions about your project, such as which entities reference Customer, without creating, changing or deleting anything. OneAssist →

Also in August: about 130 smaller improvements and fixes, mostly to saving and shared editing, OneAssist and the logical model.

July 2026

July introduced Tasks, work items linked to the model, and made shared editing and the version history more dependable.

• Tasks: work items linked to the model

  Work items, from epics to bugs, link to the entities, Flows and dictionary terms they concern, and each of those lists its tasks. Work them as a tree, a board or a list, and let OneAssist create, assign and link them. Tasks →

• Edit alongside teammates without losing work

  Edits made at the same moment as a teammate's are no longer lost or overwritten, locks are released reliably when someone disconnects, and OneAssist and version restores stop at work someone has open. Features →

• Version numbers that are never reused

  Each version gets a number of its own, never reused after a delete or shared when two people save at once, and an interrupted save or restore no longer leaves a version or a project half done. Versions & audit →

Also in July: seven smaller improvements and fixes, mostly to the dbt export, project deletion and model validation.

June 2026

June was mostly about Flow: a dbt export for Microsoft Fabric and richer load design, plus reliable saving of very large models.

• Export Flows as a runnable dbt project for Microsoft Fabric

  Flows export as one dbt project for Fabric Data Warehouse: staging views, a model per step and marts per target entity, with tests from the keys. Merge steps load incrementally on your merge keys and add newly mapped columns to the target table. Flow →

• Merge steps with standard SQL MERGE branches

  A Merge step has WHEN MATCHED, WHEN NOT MATCHED BY TARGET and WHEN NOT MATCHED BY SOURCE branches, each with an optional condition and an action, and the dbt export generates them as one MERGE statement. Flow →

• Import source tables into Flow from SQL DDL

  Paste a SQL DDL script to bring its tables into Flow as source objects. Scripts from SQL Server Management Studio import correctly too, with bracketed names, storage clauses and computed columns. Flow →

• See what each Flow step does, and what needs fixing

  In Definition mode a Flow step shows a short summary of what it does, such as the columns a Lookup returns, and warnings flag a step whose source is no longer connected and names used twice. Flow →

• Usage quotas for OneAssist

  Platform administrators can cap how many messages an agent answers in a rolling time window, and everyone chatting can follow the usage on a chart. Only platform administrators create or delete agents. OneAssist →

• Large models stay fast to edit

  Models with hundreds of entities and submodels save reliably, dictionaries with thousands of terms open quickly, and long lists gain a search box. Logical model →

Also in June: 18 smaller improvements and fixes, mostly to sign-in and first-time setup, security and the Flow editor.

May 2026

Model One became a team tool: sign-in and roles, live collaboration, an assistant and MCP server, richer Flow steps, and ER/Studio import.

• Sign in, share projects and edit together live

  Sign in through Microsoft Entra and share each project as Owner, Member or Viewer. Changes appear for everyone as they happen, and an object locks while someone edits it, freeing itself when they leave. Features →

• Flow steps that model real data operations

  Flow gained sources and column-level mappings, and its steps became typed operations such as Join, Lookup, Aggregate, Merge, Validate, Mask and Match, each with its own settings. A step's outputs can feed the next step, so lineage holds across stages. Flow →

• An assistant and an MCP server for every project

  Each project gained an MCP endpoint for AI agents, and an assistant on Azure AI Foundry in your own subscription, with Plan and Agent modes, reference files and optional web search. Named OneAssist at the end of the month, it can build a whole Flow from one request. OneAssist →

• Column-level control in physical models

  A Columns tab sets each column's native type, nullability and default, marks it computed or identity, and overrides table and column names. A table can take its own schema or be left out of a model. Physical model →

• Import ER/Studio models

  Import ER/Studio .DM1 files with entity and attribute definitions, notes, dictionary links, alternate keys, submodels and Flow diagrams, with progress shown and a summary of anything left out. Features →

• Undo, redo and version restore

  Undo and redo a recent change in any module, with a preview of what will happen, and restore the model from an earlier version. Versions & audit →

Also in May: a new canvas with zoom, panning and adjustable lines, entity templates, the in-app user manual, the first container image for installing Model One in your own Azure subscription, and dozens of smaller improvements and fixes.

April 2026

The first version of Model One: a logical model, data-flow diagrams, DDL for four databases, and versions you can compare.

• Logical models on a canvas

  Entities, attributes, keys and relationships on a canvas, organised into nested submodels, with display levels from names only to every attribute, and a data dictionary for the business terms behind them. Logical model →

• Data-flow diagrams for how data moves

  Draw how data moves between systems, processes and entities, in diagrams kept in folders. They were the start of what became Flow. Flow →

• Generate DDL for four databases

  Derive physical models from the logical model and generate DDL for PostgreSQL, SQL Server, Oracle and MySQL. Physical model →

• Import from SQL DDL and DBML

  Bring an existing schema in from a SQL DDL script, and import or export DBML. Features →

• Versions, comparisons and an HTML report

  Save numbered versions of the model, compare any two, and export the model as a self-contained HTML report. Versions & audit →

Also in April: copy and paste, multi-select, drag and drop, collapsible objects and a handful of fixes.