Changelog

What changed in Eddytor, week by week - new features, fixes, and doc updates.

Every notable change to Eddytor, newest first. Entries are grouped by week and split into Added, Changed, and Fixed. For endpoint-level details, see the API reference.

2026-11-01

Engine v2.10.0

Before you upgrade (self-host): snapshot Postgres first. The server runs new database migrations at startup, and it now refuses to start when [mail] holds app settings but names no mail provider. See Upgrades.

Added

  • Reference a plain column, like a foreign key - a column can take its allowed values from a column in another table, and that column needs no domain of its own. Values are compared by type, so a number references a number and a date references a date. See Reference (cross-table).
  • Referenced values are protected - Eddytor refuses to delete or change a value that another table still references.
  • Label column on references - show a readable label, such as a name, next to each code. Only the code is stored.
  • Send mail through Microsoft Graph with an Entra app, using a client secret, workload identity, or managed identity. See Mail via Microsoft Graph.
  • SMTP with OAuth - sign in to Microsoft 365 SMTP with Entra app tokens instead of a password. See SMTP.
  • Entra SSO for accounts without an email address - a connection can accept the user principal name instead. See SSO with Microsoft Entra ID.
  • Roll back a release (self-host) - eddytoradm migrate status shows the database schema version, and eddytoradm migrate undo reverts the migrations of a newer release.

Changed

  • Azure and Google account links are per organisation. If you belong to several organisations, link your account once in each. Existing links move to one of your organisations during the upgrade. See Link cloud accounts.
  • After an SSO or invite sign-in, you land in the organisation that sign-in came from.
  • Linking an Azure account asks for storage access first. Access to browse subscriptions is optional.
  • A table that cannot be loaded now says why: an expired storage link asks you to link again, a storage outage is a server error, and "not found" means the table is gone.
  • Unresponsive object storage fails a request after about 10 seconds for metadata and about 30 seconds for data, instead of hanging for minutes.
  • merge_rows refuses the whole batch when a row's _operation is not exactly INSERT, UPDATE, or DELETE. Before, those rows were skipped without an error.
  • MCP tools refuse argument names they do not declare, so a misspelled argument fails instead of being ignored.
  • A create-table schema can no longer carry reference or table-lookup domains. Set them after you create the table.
  • Creating a table without a storage connection ID is refused when more than one connection could match.
  • Re-registering an AWS bucket with a corrected region updates the existing storage connection instead of failing.
  • The Go SDK is discontinued.

Fixed

  • A table deleted or recreated outside Eddytor is reported as unavailable instead of showing its old contents.
  • Tables with required (NOT NULL) columns are marked the way Databricks expects. Older tables are repaired on their next write. A write that collides with that repair returns a retryable 409.
  • Registering the same bucket twice at the same time works.
  • A row can reference another row in the same batch, text columns from Polars and ADBC clients pass reference checks, and reference cycles are refused when you set them.
  • A slow Microsoft sign-in endpoint no longer holds up every outgoing mail.
  • eddytoradm create-api-key picks a unique default name, so running it or the installer again works. A duplicate API key name over REST returns 400 instead of 500.
  • MCP profile_table includes columns another user just added.
  • Security hardening: references and table lookups respect row- and column-level security on every path, error messages no longer show storage endpoints or names of hidden columns, and storage credentials are never shared between organisations or workspaces. A rejected storage registration no longer leaves its credentials in use. Reported dependency advisories are patched.

2026-10-18

Engine v2.9.0 – v2.9.1 · Web UI v2.6.0

Added

  • Column comments and display names - describe a column and give it a readable label without renaming it, from the CLI, REST, or MCP. The Web UI shows the display name in the grid. Comments use the field Databricks reads, so text you write in one shows in the other. See column settings.

Fixed

  • Every read sees the latest committed data. A change from the Web UI, another session, or an external pipeline is visible to the next query on every interface, including MCP and Flight SQL. Before, it could take up to an hour.
  • If you belong to several organisations, linking a Microsoft or Google account uses your current organisation's provider app, and a new sign-in returns you to the organisation you last switched to.
  • The Web UI opens tables again after the v2.9.0 upgrade (Unrecognized type: "undefined" (24)).

2026-10-11

Engine v2.8.15 · Web UI v2.5.14 – v2.5.15

Changed

  • The Web UI storage browser shows each connection's account and container.

Fixed

  • Two storage connections with the same container or bucket name no longer get mixed up. A new table is written to the connection you chose.
  • Registering storage no longer overwrites an existing connection that happens to have the same name, and registering a renamed connection again no longer creates a duplicate.

2026-09-27

Web UI v2.5.13

Fixed

  • Typing into a numeric cell to edit it no longer crashes the grid.

2026-09-20

Engine v2.8.14 · Web UI v2.5.12

Added

  • eddytor login works on SSO domains - the CLI sign-in sends you to your organisation's identity provider.
  • First SSO sign-in creates the account - a new user gets the connection's defaultRole without an invite. See SSO (OIDC).
  • The Web UI constraint builder offers IS NULL and IS NOT NULL.

Changed

  • Table listings and refreshes are faster: only tables that changed are read again.
  • POST /v1/auth/sso/login and POST /v1/auth/sso/callback are removed. SSO goes through the standard sign-in flow.

Fixed

  • A third-party app that signs you in through SSO now shows the consent screen that names the app, like every other sign-in.
  • Constraint validation counts NULL results as violations, so it finds the same rows that a write rejects.
  • Copying cells from the Web UI grid keeps their decimals.

2026-09-13

Engine v2.8.11 – v2.8.13 · Web UI v2.5.10 – v2.5.11

Added

  • Unmapped values on hierarchical domains - converting a fixed domain to a dependent one keeps every value. A value you have not mapped to a parent yet stays valid under any parent until you map it. See Set a domain.

Fixed

  • Deleting rows from a table with a composite primary key deletes only the rows whose full key matches. Before, it could delete every row that shared one key column's value.
  • Tables with a date primary key accept inserts and merges again.
  • An S3 bucket and an Azure container with the same name work side by side.
  • Table history is complete for tables older than 30 days. Before, it could come back empty.
  • The Web UI shows dates and timestamps as stored, whatever your time zone.
  • Saving a dependent dropdown in the Web UI keeps unmapped values.

2026-09-06

Web UI v2.5.8 – v2.5.9

Changed

  • A self-hosted Web UI loads no analytics unless the deployment turns them on.
  • A new Web UI version is announced in a short notification instead of an inline banner.

Fixed

  • AI fix nulls suggestions display for every response shape.

2026-08-30

Engine v2.8.7 – v2.8.10 · Web UI v2.5.5 – v2.5.7

Added

  • AI analysis reports its coverage - each run says how many rows it covered, and you can continue a cut-off run where it stopped. In the Web UI, suggestions land in a queue where you accept or reject each one. See AI analysis.
  • AI results stream in - a streaming endpoint reports progress and partial results while the analysis runs.
  • AI suggestions name their key columns, composite keys included, and AI filters can include empty values.
  • Seed a fixed domain from existing data - build the allowed-values list from the column's current values. See Set a domain.
  • Row security by email - a policy builder condition limits each user to rows that carry their own email address, also on large tables. See Create policies.
  • Unity Catalog with your linked Azure identity - no service principal or stored token. Each user sees Unity Catalog with their own Databricks permissions.
  • Member overview - admins see which workspaces each member can reach and the storage shared into them, in REST, eddytor get users, and the Web UI members view.
  • MCP gains add_constraints and drop_constraint.

Changed

  • A fixed domain with an empty value list is refused.
  • An AI analysis samples at most 10,000 rows. Larger requests are refused.
  • The Web UI labels OneLake connections as Microsoft Fabric.

Fixed

  • Adding a CHECK constraint to a table with violating rows returns a clear error with the number of violating rows, and no row contents.
  • Long-running requests such as AI analysis no longer time out at the gateway.
  • Accepting an AI suggestion in the Web UI no longer damages read-only tables or tables with duplicate keys.
  • The published OpenAPI spec validates as OpenAPI 3.0.
  • eddytor domain hierarchy commands send the correct values again.

2026-08-26

Web UI v2.5.4

Added

  • Microsoft Fabric is its own connection provider - a three-step wizard (credentials → lakehouse → options) registers a OneLake lakehouse without you typing the account name or base path. See Microsoft Fabric.

Changed

  • The Azure Blob connection form is a three-step wizard too - credentials, location, options, and a summary before you connect. Its Use Microsoft Fabric endpoint checkbox is gone; pick the Fabric provider instead.
  • Unity Catalog linking is now offered only where it applies - on Azure Blob's options step, never on Fabric.

2026-08-23

Engine v2.8.0 – v2.8.6 · Web UI v2.5.0 – v2.5.3

Added

  • Workspace roles now govern per-request authority - what a request may do is decided by the caller's role in the workspace it targets.
  • Multi-level domain chains - domains can nest parent/child hierarchies deeper than one level.
  • Seed domain mappings from table data - auto-detect reads existing values and creates editable inline mappings in the Web UI.
  • Composite primary keys - an atomic endpoint sets a multi-column primary key, and the Web UI designate-PK modal supports it.
  • Microsoft Fabric endpoints - eddytor az storage register gains Fabric endpoint and base-path flags, mirrored in the Web UI register form.
  • AI Magic overhaul - scoped analysis, filters, refreshed model catalogs, seven new MDM actions, and a redesigned result drawer. See AI analysis.
  • Unity Catalog managed tables resolve read-only during storage discovery.
  • Storage discovery exclude patterns and Azure tag-filtered browsing.
  • Resizable file explorer - drag the Web UI sidebar's right edge to any width between 280 and 600 px, double-click the handle to reset to 320 px. The width is remembered per browser. See Navigating the Web UI.

Fixed

  • Setting a single-column primary key now validates for duplicate values first.
  • OAuth refresh-token rotation allows a 60-second grace window for replayed requests, so parallel clients no longer get signed out.
  • Removing an organisation member now revokes their pending invites and tokens.
  • SSO: sign-in is refused when an email domain maps to more than one organisation, invite links grant access only to the organisations they were issued for, and invite matching and re-invite acceptance work reliably.
  • Permission guards that name legacy scopes no longer grant roles.
  • OAuth client registration rejections now name the reserved client_name fragments.

2026-08-16

Engine v2.7.6 · Web UI v2.4.3 – v2.4.4

Added

  • Rotate storage credentials without re-registering - swap keys on an existing storage connection from CLI, REST, or the Web UI.
  • Every REST operation now carries a stable operationId in the OpenAPI spec, so generated clients keep their method names across releases.
  • The Azure admin-consent flow in the Web UI is redesigned into three explicit steps.

Changed

  • The account Security tab is renamed Identity.

Fixed

  • Tables on Google Cloud Storage survive discovery - the gs:// logstore factory is registered at startup.
  • The CLI defaults its host to api.eddytor.com.
  • OpenAPI spec accuracy: typed 200 responses, correct server URL, security wiring, and media types.
  • Row- and column-level security enforcement now also covers Delta merge and delete write paths and additional read paths. See row/column security.

2026-08-09

Engine v2.7.0 – v2.7.5 · Web UI v2.4.0 – v2.4.2

Added

  • Workspaces arrive on Eddytor Cloud - a dedicated v2 cluster with workspace-scoped storage. Learn more under Workspaces.
  • Device-flow signup - eddytor login can create your account and organisation in one step.
  • OAuth consent screen, dynamic-client-registration hardening, and token-scope enforcement.
  • Sidebar workspace switcher in the Web UI.
  • SSO connections take a provider-driven issuer URL and can reuse an existing SSO app registration. See SSO (OIDC).
  • Storage-account discovery shows visible progress in the Web UI.

Changed

  • Auth emails and identity-provider error pages carry Eddytor branding.

Fixed

  • Security hardening: stricter rate limiting and request-body size limits, sign-in responses no longer reveal whether an account exists, and sensitive token values no longer appear in logs.
  • The magic-link verify page requires POST, so mail scanners can no longer burn the token before you click it.
  • The Web UI origin is allowed in the provider-link return_to allow-list.
  • The request body cap no longer rejects bodyless requests with 411.

2026-08-02

Engine v2.6.0 · Web UI v2.3.0

Added

  • Workspace-owned storage connections - register storage that belongs to a workspace instead of a single user. See shared storage.
  • Azure tenant-wide admin consent is surfaced on provider apps.

Fixed

  • Policies are grouped and resolved by table UUID instead of fully-qualified name, so renames no longer detach them.
  • Workspace share-by dates render in the correct format.

2026-07-26

Engine (pre-v2.6.0) · Web UI v2.2.0

Added

  • Row- and column-level security - Cedar-based policies restrict which rows and columns each principal can read or write, manageable from the Web UI. Start at row/column security.

2026-07-19

Added

  • Multi-org accounts - one account can belong to several organisations, with credentials bound to the organisation they were issued in.
  • Workspaces - a sub-organisation tier with workspace-bound credentials. See switching and credentials.

2026-06-28

Web UI v2.1.1

Added

  • This help center launched - guides for deploying, configuring, and operating Eddytor, plus the full table lifecycle.

Fixed

  • Explorer: clicking an already-open table no longer leaves it stuck loading.

On this page