## Summary
- **Add `lingui:compile` to Dockerfile** before both the server and
frontend build stages, ensuring compiled translation catalogs are always
fresh regardless of git state
- **Add `repository-dispatch` to i18n workflows** (`i18n-push.yaml` and
`i18n-pull.yaml`) to trigger reactive automerge in `twenty-infra` when
the i18n PR is ready, replacing the 15-minute polling approach
## Context
Users sometimes see "Uncompiled message detected" errors because
releases can be cut from `main` before the i18n PR (with freshly
compiled translation catalogs) has been merged. This creates a race
condition between new translatable strings landing on `main` and their
compiled catalogs being available.
These changes fix this in two ways:
1. **Safety net in builds**: Every Docker build now compiles
translations before building, so even if compiled catalogs in git are
stale, the build artifact is always correct
2. **Faster i18n PR merges**: Instead of a 15-minute cron polling for
i18n PRs, the workflows now notify `twenty-infra` immediately when
translations are ready, reducing merge latency from ~15 minutes to ~1
minute
Companion PR in twenty-infra: twentyhq/twenty-infra
(feat/i18n-reactive-automerge)
## Test plan
- [ ] Verify `TWENTY_INFRA_TOKEN` secret is available to i18n workflows
- [ ] Docker build still succeeds with the added `lingui:compile` steps
- [ ] i18n-push triggers automerge in twenty-infra after pushing changes
- [ ] i18n-pull triggers automerge in twenty-infra after pulling
translations
Made with [Cursor](https://cursor.com)
## Summary
- Removes the `isBillingEnabled` guard that was hiding the Providers and
Custom Providers sections in the Admin AI settings
- The `GET_AI_PROVIDERS` query was being skipped when billing was
enabled, and the provider UI sections were conditionally hidden —
there's no reason to gate provider configuration behind billing status
- Cleans up the now-unused `billingState` and `useAtomStateValue`
imports
## Test plan
- [ ] Verify Providers and Custom Providers sections are visible in
Admin > AI settings on cloud (billing enabled)
- [ ] Verify they remain visible on self-hosted (billing disabled)
Made with [Cursor](https://cursor.com)
## Summary
This PR adds a comprehensive billing usage analytics feature that
provides detailed breakdowns of credit consumption across execution
types, users, resources, and time periods. The implementation includes a
new ClickHouse-backed analytics service, GraphQL API endpoint, and a
frontend dashboard component.
## Key Changes
### Backend
- **New BillingAnalyticsService**: Queries ClickHouse for usage
breakdowns by user, resource, execution type, and time series data
- **BillingEventWriterService**: Writes billing events to ClickHouse for
analytics while maintaining best-effort semantics (never blocks Stripe
billing)
- **ClickHouse Schema**: Added `billingEvent` table with 3-year TTL for
storing detailed billing event data
- **GraphQL Resolver**: New `getBillingAnalytics` query that aggregates
usage data for the current billing period, protected by feature flag and
billing permissions
- **Enhanced BillingUsageEvent**: Added `userWorkspaceId` field to track
per-user credit consumption
- **AI Billing Integration**: Updated AI billing service to pass
`userWorkspaceId` when recording usage events
### Frontend
- **SettingsBillingAnalyticsSection**: New component displaying:
- Usage breakdown by execution type with progress bars
- Daily usage time series chart (28-day view)
- Per-user credit consumption breakdown
- Per-resource (agent/workflow) credit consumption breakdown
- **SettingsUsage Page**: Dedicated page for viewing usage analytics
- **GraphQL Query**: `GetBillingAnalytics` query with generated hooks
- **Navigation**: Added Usage menu item in settings (feature-flagged)
- **Mock Data**: Included screenshot mock data for preview/testing
### Feature Flag
- Added `IS_USAGE_ANALYTICS_ENABLED` feature flag to control visibility
and access to analytics features
## Implementation Details
- Analytics data is queried in parallel for performance
- ClickHouse writes are non-blocking to ensure billing operations never
fail
- Progress bars use dynamic coloring from a predefined palette
- Time series visualization normalizes bar heights relative to max value
- Empty state handling when no analytics data is available
- Responsive UI with proper text truncation for long names
https://claude.ai/code/session_01Y1EqrX6PFq3EJxJq89h7DF
---------
Co-authored-by: Claude <noreply@anthropic.com>
## Summary
- The `MigrateModelIdsToCompositeFormat` migration was setting
`agent.modelId = NULL`, but the column has a `NOT NULL` constraint —
causing the migration to fail on deploy.
- Fixed by setting `modelId` to the `'default-smart-model'` sentinel
value instead of NULL. The runtime already resolves this sentinel
dynamically via `getEffectiveModelConfig` / `isDefaultModelSentinel`, so
agents correctly fall back to the admin-configured smart model.
## Test plan
- [ ] Run `npx nx run twenty-server:database:migrate:prod` — migration
should complete without errors
- [ ] Verify agents still resolve to the correct model at runtime
(sentinel → `getDefaultPerformanceModel()`)
Made with [Cursor](https://cursor.com)
- Renamed FieldConfiguration's layout field to fieldDisplayMode as it
caused issues with the layout field of BarChartConfiguration
- Create relation Field widgets for standard objects
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
## Automated fix for [bug 6848](https://sonarly.com/issue/6848?type=bug)
**Severity:** `critical`
### Summary
WorkflowEditActionCodeFields.tsx calls Object.entries(functionInput)
without a null guard on line 40, crashing when
action.settings.input.logicFunctionInput is undefined. This happens
because workflow version step data in the database may contain the old
field name serverlessFunctionInput (from before the rename in commit
da6f1bbef3), and no data migration was created to update JSON keys
inside the steps JSONB column.
### User Impact
Users viewing or editing a workflow version containing a CODE step that
was created before the serverlessFunction-to-logicFunction rename see a
crash. The page fails to render the workflow step detail panel,
preventing them from viewing or modifying their workflow.
### Root Cause
Proximate cause: Object.entries(functionInput) on line 40 of
WorkflowEditActionCodeFields.tsx throws TypeError because functionInput
is undefined. The stack trace confirms this: the crash occurs during
React rendering (through scheduler -> react-dom render pipeline ->
WorkflowEditActionCodeFields:40:15 -> Object.entries).
1. Why did Object.entries throw? Because the functionInput prop is
undefined. It is initialized from
action.settings.input.logicFunctionInput via useState on line 142 of
WorkflowEditActionCode.tsx, with no fallback value.
2. Why is action.settings.input.logicFunctionInput undefined? Because
the workflow version step data stored in the database JSONB steps column
contains the OLD field name serverlessFunctionInput instead of
logicFunctionInput. When the frontend accesses logicFunctionInput on the
deserialized JSON object, it returns undefined.
3. Why does the database still have the old field name? Because commit
da6f1bbef3 (Rename serverlessFunction to logicFunction, Jan 28 2026)
renamed the Zod schema field from serverlessFunctionInput to
logicFunctionInput, and the database migration
1769556947746-renameServerless.ts only renames SQL tables and columns
(serverlessFunction table to logicFunction, etc.) but does NOT update
the JSON keys inside the steps JSONB column of the workflowVersion
table.
4. Why was no JSON data migration created? The rename was a large-scale
refactoring across the entire codebase. The steps column stores workflow
action data as opaque JSON, and the migration only addressed relational
schema changes (table/column renames) without updating the embedded JSON
document structure. There is no upgrade command in 1-17, 1-18, or 1-19
directories that transforms the JSON keys from serverlessFunctionInput
to logicFunctionInput.
5. Why did the component not handle this gracefully? The
WorkflowEditActionCodeFields component has no null guard on the
functionInput prop before calling Object.entries. Notably, the analogous
WorkflowEditActionLogicFunction component DOES have a null guard on line
52 (action.settings.input.logicFunctionInput ?? {}), showing the team
was aware of this possibility in one component but missed it in the
other.
**Introduced by:** martmull on 2026-02-16 in commit
[`da064d5`](https://github.com/twentyhq/twenty/commit/da064d5e88a62939e0545a37c68381822e6932ef)
### Suggested Fix
Two minimal null guards are added, matching the pattern already used in
the analogous WorkflowEditActionLogicFunction component. First, in
WorkflowEditActionCode.tsx the useState initializer is changed from
action.settings.input.logicFunctionInput to
action.settings.input.logicFunctionInput ?? {} so that functionInput is
always an empty object rather than undefined when the DB record still
uses the old serverlessFunctionInput key. Second, in
WorkflowEditActionCodeFields.tsx the Object.entries call is changed from
Object.entries(functionInput) to Object.entries(functionInput ?? {}) as
a defensive guard at the render layer, ensuring the component renders
safely even if an undefined value is passed from any caller.
### Evidence
- **Code:**
[packages/twenty-front/src/modules/workflow/workflow-steps/workflow-actions/code-action/components/WorkflowEditActionCode.tsx:142
- const [functionInput, setFunctionInput]
=](https://github.com/twentyhq/twenty/blob/main/packages/twenty-front/src/modules/workflow/workflow-steps/workflow-actions/code-action/components/WorkflowEditActionCode.tsx#L142)
- **Code:**
[packages/twenty-front/src/modules/workflow/workflow-steps/workflow-actions/code-action/components/WorkflowEditActionCodeFields.tsx:40
- {Object.entries(functionInput ?? {}).map(([inputKey, inputValue]) =>
{](https://github.com/twentyhq/twenty/blob/main/packages/twenty-front/src/modules/workflow/workflow-steps/workflow-actions/code-action/components/WorkflowEditActionCodeFields.tsx#L40)
---
*Generated by [Sonarly](https://sonarly.com)*
Co-authored-by: Sonarly Claude Code <claude-code@sonarly.com>
- Automatically create and sync command menu items for all workflows
with a manual trigger
- Refactor `useCommandMenuItemsFromBackend`
- Prefill a _Quick Lead_ workflow command menu item during workspace
setup and dev seeding
- Add a ready prop to `HeadlessEngineCommandWrapperEffect` to prevent
premature execution when async data hasn't loaded yet
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
## Summary
Follow-up to #18157. Cherry-picks the useful parts from #18481 (by
@dnplkndll), adapted to align with the current state of `main`:
- **Helm unit tests** for Redis external authentication (secret-based +
plaintext password, both server and worker)
- **Helm unit tests** for `extraEnv` injection (plain values and
`valueFrom` on both server and worker)
- **JSON schema validation** for `server.extraEnv` and `worker.extraEnv`
in `values.schema.json`
- **Fix** `server_url_test.yaml` to use JSONPath filters
(`@.name=="SERVER_URL"`) instead of brittle `env[0]` index selectors
- **Fix** worker `storageEnv` whitespace (missing `-` in `{{-
$storageEnv | nindent 12 }}`)
Stale tests from #18481 (for `disableDbMigrations`, `server.env`
pass-through, and `run-migrations` init container) were dropped since
those features were removed during the #18157 cleanup.
## Test plan
- [ ] `helm lint` passes
- [ ] `helm template` renders cleanly
- [ ] Unit tests pass with `helm unittest` (requires the plugin)
Made with [Cursor](https://cursor.com)
This pull request enhances the Helm chart for the Twenty application by
improving how environment variables and Redis credentials are handled
for both server and worker deployments. The main changes include support
for injecting additional environment variables, improved Redis password
management (including external secrets), and a more robust database
migration workflow.
**Environment Variable Injection:**
- Added support for specifying additional environment variables for both
the server and worker deployments via the `additionalEnv` field in
`values.yaml`. These variables are automatically injected into the
respective pods.
[[1]](diffhunk://#diff-b5d958eae48fd1919e5623bcf0144aac7abb323ae8743e6f31367e383c63c296R55)
[[2]](diffhunk://#diff-b5d958eae48fd1919e5623bcf0144aac7abb323ae8743e6f31367e383c63c296R109-R110)
[[3]](diffhunk://#diff-20bb91909627a12b50b3c165a2a027b663479c0104ed8dbf91d2b9ad8ea8a931R74-R77)
[[4]](diffhunk://#diff-20bb91909627a12b50b3c165a2a027b663479c0104ed8dbf91d2b9ad8ea8a931R157-R172)
[[5]](diffhunk://#diff-20bb91909627a12b50b3c165a2a027b663479c0104ed8dbf91d2b9ad8ea8a931R225-R229)
[[6]](diffhunk://#diff-fb612a3b7a13156aaa607b27d23025e2c6831f111b6a582fd313fad26d2fdb5bR89-R92)
**Redis Credential Management:**
- Introduced support for using external secrets for Redis passwords by
adding `secretName` and `passwordKey` fields under `redis.external` in
`values.yaml`, and logic to inject `REDIS_PASSWORD` from a Kubernetes
secret if configured.
[[1]](diffhunk://#diff-b5d958eae48fd1919e5623bcf0144aac7abb323ae8743e6f31367e383c63c296R180-R182)
[[2]](diffhunk://#diff-5c4fa358b10abd7581188995feb9b4d6be0bc4f06a95bf27bb31b5595d6693d8R92-R100)
[[3]](diffhunk://#diff-20bb91909627a12b50b3c165a2a027b663479c0104ed8dbf91d2b9ad8ea8a931R157-R172)
[[4]](diffhunk://#diff-20bb91909627a12b50b3c165a2a027b663479c0104ed8dbf91d2b9ad8ea8a931R196-R205)
[[5]](diffhunk://#diff-fb612a3b7a13156aaa607b27d23025e2c6831f111b6a582fd313fad26d2fdb5bR70-R79)
- Updated the logic for constructing the `REDIS_URL` to include
authentication information if a password is set or an external secret is
used.
**Database Migration Workflow:**
- Improved the startup command for the server deployment to optionally
skip database migrations (using `DISABLE_DB_MIGRATIONS`), check for an
existing schema before running migrations, and ensure setup scripts are
only run on empty databases.
These changes make the chart more flexible and secure, especially for
production deployments requiring externalized secrets and custom
environment configurations.
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
## Summary
Fixes#13008
Azure AD has a [known
bug](https://learn.microsoft.com/en-us/answers/questions/5570899/our-adminconsent-url-always-fails-with-aadsts65005)
where the first consent attempt for a multi-tenant app fails with
`AADSTS650051` because the service principal is mid-provisioning in the
target tenant. Microsoft's own engineers have acknowledged the issue
with no fix ETA.
This PR adds a transparent retry to the `MicrosoftOAuthGuard`:
1. When the OAuth redirect returns `AADSTS650051`, the guard parses the
`state` parameter to recover the original query params (workspaceId,
invite hash, locale, etc.)
2. Redirects the user back to `/auth/microsoft` with a `msRetry=1`
counter
3. The full OAuth flow restarts — by this point the service principal
has finished provisioning, so consent succeeds
4. Capped at 1 retry (`MAX_MICROSOFT_AUTH_RETRIES`) to prevent infinite
loops. If it still fails, falls through to the normal error page
From the user's perspective, they just see a brief extra redirect
instead of a cryptic error page.
### Context
- `AADSTS650051` is a race condition in Azure AD's service principal
provisioning during multi-tenant app consent
- It's distinct from missing consent (`AADSTS65001`) or admin consent
required (`AADSTS90094`)
- The error is transient — retrying the same flow immediately succeeds
- Microsoft Q&A threads confirm this affects many multi-tenant apps, not
just Twenty
### References
- [Microsoft Q&A: adminconsent always fails with AADSTS650051 on first
attempt](https://learn.microsoft.com/en-us/answers/questions/5570899/our-adminconsent-url-always-fails-with-aadsts65005)
- [Microsoft Q&A: AADSTS650051 for multiple
applications](https://learn.microsoft.com/en-us/answers/questions/5571098/when-customers-attempt-to-sign-in-and-grant-consen)
## Test plan
- [ ] Deploy to a staging environment with Microsoft OAuth enabled
- [ ] Have a user from a new Azure AD tenant attempt Microsoft sign-in
for the first time
- [ ] If AADSTS650051 occurs, verify the user is transparently retried
and signs in successfully
- [ ] Verify `msRetry` counter prevents infinite redirect loops (check
server logs for the warn message)
- [ ] Verify normal Microsoft sign-in flow (no error) is unaffected
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: neo773 <62795688+neo773@users.noreply.github.com>
Co-authored-by: neo773 <neo773@protonmail.com>
## Summary
- Renames `ObjectMetadataItem` to `EnrichedObjectMetadataItem` across
the entire frontend (~440 files) to clarify that this type includes
derived fields (`readableFields`, `updatableFields`, nested `fields[]`,
`indexMetadatas[]`) computed at read time from the metadata store
- Creates `splitObjectMetadataGqlResponse` that goes directly from a
GraphQL `ObjectMetadataItemsQuery` response to flat store items
(combining the old
`mapPaginatedObjectMetadataItemsToObjectMetadataItems` +
`splitObjectMetadataItemWithRelated` two-step flow into one call)
- Removes `ObjectMetadataItemWithRelated` type and all "WithRelated"
naming
- Renames `generatedMockObjectMetadataItems` to
`generateTestEnrichedObjectMetadataItemsMock` to make it clear this is
test-only enriched data
- Deletes `useLoadMockedObjectMetadataItems` hook (consolidated into
`useLoadMockedMinimalMetadata`)
- Ensures nothing destined for the metadata store computes
`readableFields`/`updatableFields` (preventing the localStorage bloat
from #18809)
## Type hierarchy (before → after)
**Before:**
```
ObjectMetadataItemsQuery → mapPaginated → ObjectMetadataItemWithRelated → enrich → ObjectMetadataItem
→ split → FlatObjectMetadataItem (store)
```
**After:**
```
ObjectMetadataItemsQuery → splitObjectMetadataGqlResponse → FlatObjectMetadataItem (store)
→ mapPaginated + enrich (tests only) → EnrichedObjectMetadataItem
```
## Test plan
- [x] `npx nx typecheck twenty-front` passes
- [x] `npx nx test twenty-front` passes (767 suites, 4505 tests)
- [x] `npx nx lint twenty-front` passes
- [ ] CI checks pass
Made with [Cursor](https://cursor.com)
## Summary
- On unauthenticated pages (login), mocked object metadata was being
hydrated into the store with `readableFields` and `updatableFields`
attached. These derived arrays duplicate every field per object,
inflating `objectMetadataItems` in localStorage from ~70 KB to ~2 MB.
Combined with other metadata keys, this exceeded Safari's 5 MB quota and
caused `QuotaExceededError`, preventing real data from replacing mock
data on login.
- Extracted flat mock data into a dedicated file
(`generatedMockObjectMetadataItemsWithRelated.ts`) and use it for store
hydration, keeping the enriched version
(`generatedMockObjectMetadataItems`) only for tests.
- Completed `splitObjectMetadataItemWithRelated`'s contract by stripping
`readableFields` and `updatableFields` at runtime (not just via
TypeScript's `Omit`), matching the `FlatObjectMetadataItem` return type
that omits all four relational properties.
## Root cause
1. `MinimalMetadataLoadEffect` loads mocked metadata on unauthenticated
pages.
2. `generatedMockObjectMetadataItems` was produced by
`enrichObjectMetadataItemsWithPermissions`, which attaches
`readableFields` and `updatableFields` (full copies of the `fields`
array).
3. `splitObjectMetadataItemWithRelated` only destructured `fields` and
`indexMetadatas` — `readableFields`/`updatableFields` leaked into
`...objectProperties` at runtime because `Omit` is a compile-time-only
guard.
4. These bloated objects were written to localStorage, consuming ~2 MB
instead of ~70 KB.
5. On login, the intermediate state (old composite mock `current` + new
flat real `draft`) exceeded the 5 MB quota, causing a
`QuotaExceededError` deadlock.
## Test plan
- [ ] Open the app in a fresh Safari private window (empty localStorage)
- [ ] Verify the login page loads without errors
- [ ] Log in and verify metadata loads correctly
- [ ] Check `localStorage` size —
`metadataStoreState__objectMetadataItems` should be ~70 KB, not ~2 MB
- [ ] Run existing tests: `npx nx test twenty-front` — no regressions
from the mock data split
Made with [Cursor](https://cursor.com)
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
## Summary
- Store SSO connections (Google, Microsoft, OIDC, SAML) as connected
accounts in the core schema during sign-in/sign-up, gated behind the
`IS_CONNECTED_ACCOUNT_MIGRATED` feature flag
- Add `OIDC` and `SAML` to `ConnectedAccountProvider` enum with
exhaustive switch handling across frontend and backend
- Add `IS_CONNECTED_ACCOUNT_MIGRATED` to `DEFAULT_FEATURE_FLAGS` for new
workspaces, with a fallback check so SSO accounts are created even
before workspace activation
- Always upsert connected accounts to both workspace and core schemas
during messaging OAuth flow, fixing FK constraint violations when
SSO-only accounts exist only in core
- Create message/calendar channels when they don't exist regardless of
new vs reconnect flow
- Filter settings accounts list to only show accounts that have message
or calendar channels
## Test plan
- [ ] Sign up with Google SSO → verify connected account is created in
core schema
- [ ] Connect messaging (Google APIs) after SSO sign-up → verify no FK
errors, channels created, configuration page renders correctly
- [ ] Reconnect an existing messaging account → verify tokens updated,
sync resets triggered
- [ ] Sign in with OIDC/SAML SSO → verify connected account created with
oidcTokenClaims
- [ ] Verify settings accounts page only shows accounts with channels
(SSO-only accounts hidden)
- [ ] Verify typecheck, lint, and unit tests pass
## Summary
Fixes#18744 — The workflow FIND_RECORDS action silently drops filter
conditions when a variable resolves to null/empty, causing the query to
return **all records** instead of erroring.
**Root cause (three compounding layers):**
1. **`variable-resolver.ts`** — `resolveString` returns `undefined` when
a variable lookup fails (e.g., `{{steps.trigger.output.userId}}` where
`userId` doesn't exist in context). The return type says `string` but
`evalFromContext` actually returns `undefined` at runtime.
2. **`checkIfShouldSkipFiltering.ts`** — Treats `undefined`/`null`/`""`
values as "skip this filter." This is correct for the **UI filter
builder** (user hasn't finished typing), but wrong for **workflow
execution** (variable resolution failed = misconfigured workflow).
3. **`find-records.workflow-action.ts`** — When all filters are silently
skipped, `computeRecordGqlOperationFilter` returns `{}` (match
everything). The query runs with no filter, returning all records —
silently succeeding with wrong results.
## Fix
Added validation in `find-records.workflow-action.ts` **after**
`resolveInput` but **before** `computeRecordGqlOperationFilter`. For
each filter with a value-requiring operand (i.e., not IS_EMPTY,
IS_NOT_EMPTY, IS_IN_PAST, IS_IN_FUTURE, IS_TODAY), if the resolved value
is `undefined`, `null`, or `""`, throw `INVALID_STEP_INPUT` with a
descriptive error message.
**Why this approach:**
- Scoped to the workflow executor — does **not** break the UI filter
builder's intentional skip-on-empty behavior
- Does not change shared utilities (`checkIfShouldSkipFiltering`,
`resolveInput`) used across the app
- Fails fast with a clear error instead of silently returning wrong data
- 1 file changed, 23 lines added
## Test plan
- [x] Backend typecheck passes
- [x] oxlint passes (0 warnings, 0 errors)
- [x] Prettier passes
- [ ] Manual: Create a workflow with FIND_RECORDS using a variable that
doesn't exist → should error with "Filter condition has an empty value
after variable resolution" instead of returning all records
- [ ] Manual: Create a workflow with FIND_RECORDS using IS_EMPTY operand
(no value needed) → should still work correctly
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Charles Bochet <charles@twenty.com>
## Summary
- The record board (Kanban view) only loaded the first 10 records per
column, then showed skeleton placeholder cards for the rest without ever
fetching the remaining data
- **Root cause**: The `RecordBoardFetchMoreInViewTriggerComponent`
(IntersectionObserver) is positioned at the bottom of the entire board.
With 10 real cards + 10 skeleton cards per column (~3500px of content),
the trigger div was pushed far beyond the 1600px `rootMargin` detection
zone, so `recordBoardShouldFetchMore` never became `true` and fetch-more
was never triggered
- **Fix**: The initial query now sets `recordBoardShouldFetchMore =
true` when columns need more data, and `triggerRecordBoardFetchMore`
uses an internal `while` loop to load all remaining pages in a single
invocation — making it immune to the InView component racing to reset
the flag between React render cycles
## Summary
- Replaces per-provider TypeScript constant files
(`openai-models.const.ts`, `anthropic-models.const.ts`, etc.) with a
single `ai-providers.json` catalog as the source of truth
- Adds runtime model discovery via AI SDK for self-hosted providers,
with `models.dev` enrichment for pricing/capabilities
- Introduces composite model IDs (`provider/modelId`) for canonical,
conflict-free identification
- Simplifies provider configuration: API keys are injected from
environment variables (e.g., `OPENAI_API_KEY`)
- Adds admin panel UI for provider management (add/remove/test), model
discovery, recommended model configuration, and default fast/smart model
selection per workspace
- Removes deprecated config variables (`AI_DISABLED_MODEL_IDS`,
`AUTO_ENABLE_NEW_AI_MODELS`, etc.)
- Adds database migration for composite model ID format
## Test plan
- [ ] Server typecheck passes
- [ ] Frontend typecheck passes
- [ ] Server unit tests pass
- [ ] Frontend unit tests pass
- [ ] CI pipeline green
- [ ] Admin panel AI tab loads correctly
- [ ] Provider discovery works for configured providers
- [ ] Model recommendation toggles persist
- [ ] Default fast/smart model selection works
Made with [Cursor](https://cursor.com)
## Summary
Fixes#17941 — Saving a blank subdomain causes a redirect to
`.website.com`, effectively breaking the workspace.
**Root cause:** Three layers all fail to reject an empty string `""`:
1. **Frontend (`SettingsDomain.tsx`):** `SaveButton` has both
`onClick={onSave}` and `type="submit"`. The `onClick` fires first,
calling `handleSave()` directly without running Zod validation. So
`isDefined("")` returns `true`, the confirmation modal opens, and the
blank subdomain is submitted.
2. **Backend DTO (`update-workspace-input.ts`):** The `subdomain` field
has `@IsString()` + `@IsOptional()` but no pattern validation, so an
empty string passes the DTO layer.
3. **Backend service (`workspace.service.ts:152`):** `if
(payload.subdomain && ...)` — empty string is falsy in JS, so it skips
`validateSubdomainOrThrow()` entirely and writes `subdomain: ""` to the
database.
**The crash:** After save, the redirect logic does
`"myworkspace.website.com".replace("myworkspace", "")` →
`".website.com"`, sending the user to an invalid URL.
## Fix
- **Frontend:** Call `form.trigger()` at the start of `handleSave` to
run Zod validation regardless of whether the function was invoked via
`onClick` or `form.handleSubmit`. Returns early with validation error if
invalid.
- **Backend DTO:** Add `@Matches(/^[a-z0-9][a-z0-9-]{1,28}[a-z0-9]$/)`
to reject invalid subdomains at the request validation layer
(defense-in-depth).
- **Backend service:** Change `if (payload.subdomain && ...)` to `if
(isDefined(payload.subdomain) && ...)` so empty strings route through
`validateSubdomainOrThrow()` instead of being silently skipped.
## Test plan
- [x] Existing `is-subdomain-valid.util.spec.ts` tests pass (36/36)
- [x] TypeScript type checks pass for both `twenty-server` and
`twenty-front`
- [x] oxlint passes on all changed files
- [x] Prettier passes on all changed files
- [ ] Manual: Navigate to Settings > Domains, clear the subdomain field,
click Save — should show validation error, not redirect
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Charles Bochet <charles@twenty.com>
Co-authored-by: Charles Bochet <charlesBochet@users.noreply.github.com>
This PR:
- Breaks useAgentChatData into focused effect components (streaming,
fetch,
init, auto-scroll, diff sync)
- Splits message list into non-last (stable) + last (streaming/error) to
prevent full re-renders on each stream chunk
- Adds scroll-to-bottom button and MutationObserver-based auto-scroll on
thread switch
- Lifts loading state from context to atoms
- Adds areEqual to selector
factories
We could improve further but this sets up a robust architecture for
further refactoring.
## Messages flow
The flow of messages loading and streaming is now more solid.
Everything goes out from `AgentChatAiSdkStreamEffect`, whether loaded
from the DB or streaming directly, and every consumers is using only one
atom `agentChatMessagesComponentFamilyState`
## Data sync effect with callbacks new hook
See
`packages/twenty-front/src/modules/apollo/hooks/useQueryWithCallbacks.ts`
which allows to fix Apollo v4 migration leftovers and is an
implementation of the pattern we talked about with @charlesBochet
We could refine this pattern in another PR.
# Before
https://github.com/user-attachments/assets/84e7a96f-6790-405d-8a73-2dacbf783be5
# After
https://github.com/user-attachments/assets/4c692e3a-2413-4513-abcc-44d0da311203
Co-authored-by: Charles Bochet <charles@twenty.com>
## Summary
- Restores the original migration timestamp (`1773945207801`) that was
accidentally changed to `1774005903909` during PR #18787
- Keeps the content improvements (default column values) from that PR
intact
## Test plan
- [ ] Verify migration runs correctly with `npx nx run
twenty-server:database:migrate:prod`
Made with [Cursor](https://cursor.com)
## Summary
This preserves percent-encoded payloads when normalizing links fields.
`lowercaseUrlOriginAndRemoveTrailingSlash` was decoding the path and
query string while lowercasing the URL origin. That changes URLs where
encoded payloads are semantically significant, such as Google Maps links
containing `%2F` segments.
Closes#18698.
## Changes
- stop decoding the path/query payload in
`lowercaseUrlOriginAndRemoveTrailingSlash`
- preserve the raw path, query, and hash while still lowercasing the
origin and trimming a trailing slash
- update shared URL normalization tests to assert encoded payloads stay
encoded
- add a server-side regression test covering imported links field
normalization
## Validation
- `corepack yarn jest --config packages/twenty-shared/jest.config.mjs
packages/twenty-shared/src/utils/url/__tests__/lowercaseUrlOriginAndRemoveTrailingSlash.test.ts
--runInBand`
- `corepack yarn jest --config packages/twenty-server/jest.config.mjs
packages/twenty-server/src/engine/core-modules/record-transformer/utils/__tests__/transform-links-value.util.spec.ts
--runInBand`
- `corepack yarn nx test twenty-server --runInBand
--testFile=src/engine/core-modules/record-transformer/utils/__tests__/transform-links-value.util.spec.ts`
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
## Summary
- Reverts the `useSortedNavigationMenuItems` coupling in
`ViewPickerOptionDropdown` introduced by #18791
- The viewPicker now uses `useNavigationMenuItemsData` directly for the
`isFavorite` check, keeping view ordering and navigation menu item
ordering independent
## Why
PR #18791 replaced `useNavigationMenuItemsData` with
`useSortedNavigationMenuItems` in the viewPicker for favorite detection.
While the intent was to filter out stale/orphaned navigation items, this
created an unnecessary coupling: `useSortedNavigationMenuItems`
subscribes to `viewsSelector` and `objectMetadataItemsSelector`, making
the viewPicker transitively dependent on the navigation menu item
ordering system. View ordering in the picker (driven by `view.position`)
and navigation menu item ordering (driven by
`navigationMenuItem.position`) should remain decorrelated.
## Test plan
- [ ] Open the viewPicker dropdown and verify views are listed in
correct order
- [ ] Drag-and-drop to reorder views in the viewPicker — confirm it
works
- [ ] Verify the "Add to Favorite" / "Manage favorite" label still
correctly reflects favorite state
- [ ] Reorder navigation menu items in the sidebar — confirm viewPicker
order is unaffected
Made with [Cursor](https://cursor.com)
## Summary
Fixes#18757
This fixes a set of Favorites / navigation-menu-item integrity problems
related to deleted views, stale hidden items, and upgraded workspaces
with orphaned navigation items.
## What changed
- delete `navigationMenuItem` entries when their favorited view is
deleted
- keep the client metadata store in sync immediately when a view is
deleted
- determine whether a view is already favorited from visible valid
navigation items instead of raw stale items
- add a `1.20.0` upgrade repair command that deletes orphan navigation
menu items and normalizes positions
- add regression coverage for deletion of both record-based and
view-based navigation menu items
## Details
Server:
- extend `NavigationMenuItemDeletionService` so cleanup applies to
deleted views as well as deleted records
- add regression tests covering record-based deletion, view-based
deletion, and no-op behavior
- add `DeleteOrphanNavigationMenuItemsCommand` to remove orphaned items
pointing to:
- deleted views
- deleted records
- missing folders
- normalize positions per scope (`userWorkspaceId + folderId`) after
repair
- wire the new repair command into the `1.20.0` upgrade flow
Frontend:
- add `useRemoveNavigationMenuItemByViewId`
- remove the related navigation item from client metadata immediately
when deleting a view
- use sorted / visible navigation items for favorite detection so stale
hidden rows do not block re-adding a favorite
## Why
Issue `#18757` reports mismatches between Favorites shown in the UI and
rows users can still find in the database. We found that current
Favorites behavior is driven by `navigationMenuItem`, not the legacy
`favorite` table, and that stale / orphaned `navigationMenuItem` rows
could:
- remain after deleting a favorited view
- stay hidden from the UI if they point to invalid targets
- still cause the UI to think a view was already favorited
- persist in workspaces with migration damage from skipped sequential
upgrades
This patch addresses those cases directly and adds an upgrade-time
repair path for older corrupted workspaces.
## Validation
Passed:
- `./node_modules/.bin/jest --config
packages/twenty-server/jest.config.mjs --runInBand
packages/twenty-server/src/engine/metadata-modules/navigation-menu-item/services/__tests__/navigation-menu-item-deletion.service.spec.ts`
- `./node_modules/.bin/tsc -p packages/twenty-front/tsconfig.json
--noEmit --pretty false`
Known unrelated existing failure:
- `./node_modules/.bin/tsc -p packages/twenty-server/tsconfig.json
--noEmit --pretty false`
The server typecheck failure is pre-existing and unrelated to this
branch. Current errors are around `@file-type/pdf` module resolution and
`is-psl-parsed-domain.type.ts`.
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
Co-authored-by: Charles Bochet <charlesBochet@users.noreply.github.com>
On top of [previous closed
PR](https://github.com/twentyhq/twenty/pull/18713) from @FelixMalfait :
- add a schema-creation-skipping optimization
- extract a handler-per-operation pattern,
- add runtime input validation guards,
- integrate with the standard workspace cache
- add gql-style error handling
To do/optimize/check :
- gql parsing and null backfilling
## Intro
This PR introduces a **direct GraphQL execution path** that bypasses
per-workspace GraphQL schema generation for workspace data queries (CRUD
on user-defined objects like companies, people, tasks, etc.).
## Why
In the current architecture, every workspace gets its own
dynamically-generated GraphQL schema reflecting its custom objects and
fields. This costs **~20MB of RAM per workspace per pod** and takes time
to build. For a multi-tenant SaaS with thousands of workspaces, this is
a significant infrastructure cost and a latency bottleneck (especially
on cold starts or cache misses).
The insight is that most workspace queries (`findMany`, `createOne`,
`updateOne`, etc.) don't actually *need* the full schema — they can be
routed directly to the existing Common API query runners by parsing the
GraphQL AST and matching resolver names against object metadata. The
schema is only truly needed for introspection, subscriptions, or queries
that mix core and workspace resolvers.
## How It Works
1. A Yoga `onRequest` plugin intercepts incoming GraphQL requests
2. It parses the query AST and checks if all top-level fields map to
generated workspace resolvers (e.g. `findManyCompanies`,
`createOnePerson`)
3. If yes, it executes them directly against the query runners, skipping
schema generation entirely
4. If the query contains introspection, subscriptions, or core-only
resolvers, it falls through to the normal path
5. Even for mixed queries it can't fully handle, it sets
`skipWorkspaceSchemaCreation` to avoid building the schema when
unnecessary
The whole thing is gated behind the
`IS_DIRECT_GRAPHQL_EXECUTION_ENABLED` feature flag for safe incremental
rollout.
**Net effect**: dramatically lower memory footprint and faster response
times for the vast majority of workspace API calls.
---------
Co-authored-by: Félix Malfait <felix@twenty.com>
Fixes https://github.com/twentyhq/private-issues/issues/432
## Problem
When a user's invoice goes unpaid, Stripe moves their subscription to
`unpaid` status, and Twenty suspends the workspace. But if the user pays
that invoice while a new billing period has started, Stripe has already
generated a new **draft** invoice for that period. Since the draft isn't
finalized or paid, Stripe doesn't reactivate the subscription — the
workspace stays suspended indefinitely.
## What was missing
- No handler for the `invoice.paid` Stripe webhook event
- No mechanism to finalize draft invoices that accumulated during the
`unpaid` period
- No way to reset the workspace deletion countdown (`suspendedAt`) when
a user shows payment intent
## What was added
### 1. `StripeInvoiceService` — new Stripe SDK wrapper
- `listDraftInvoices(stripeSubscriptionId)` — lists all draft invoices
for a subscription
- `finalizeInvoice(invoiceId)` — finalizes a draft with `auto_advance:
true` so Stripe auto-charges it
### 2. `INVOICE_PAID` enum value
Added to `BillingWebhookEvent` so the controller can route it.
### 3. `processInvoicePaid()` in `BillingWebhookInvoiceService`
New private handler that:
- Fetches all draft invoices for the subscription
- Filters to only those whose `period_end` is in the past (already
overdue)
- Finalizes each one (with error handling per invoice to avoid blocking
the webhook)
- If the workspace is suspended, resets `suspendedAt` to now (buys time
before deletion)
### 4. Controller routing
`INVOICE_FINALIZED` and `INVOICE_PAID` are now grouped in the same
switch case, both calling `processStripeEvent(data, eventType)`, which
forks internally in the service.
## Expected recovery flow
User pays overdue invoice
→ Stripe fires invoice.paid
→ Handler finalizes past-due draft invoices (auto_advance: true)
→ Stripe auto-charges them
→ Subscription becomes active
→ customer.subscription.updated fires
→ Existing logic unsuspends workspace
→ suspendedAt refreshed (resets deletion countdown while payments
cascade)
## Summary
Fixes#18610
After completing the CreateProfile onboarding step,
`currentWorkspaceMembersState` (plural, used by `useActorFieldDisplay`
to resolve "Created by" names) was never updated with the new name —
only `currentWorkspaceMemberState` (singular) was. This caused "Created
by" fields to display "Untitled" for the remainder of the session.
**Root cause analysis:**
The issue reporter attributed the bug to empty
`core.user.firstName/lastName`, but `createdBy` display doesn't use
`core.user` at all. The actual flow:
1. Sign-up creates workspace member with empty names (copied from empty
`core.user`)
2. `GetCurrentUserDocument` loads `currentWorkspaceMembersState` with
empty names
3. CreateProfile step updates workspace member in DB +
`currentWorkspaceMemberState` (singular) ✓
4. `currentWorkspaceMembersState` (plural) is **never refreshed** — the
query is skipped once `currentUser` exists
5. `useActorFieldDisplay` looks up from the stale plural state → empty
name → "Untitled"
**Fix:** Update `currentWorkspaceMembersState` alongside
`currentWorkspaceMemberState` in the CreateProfile submit handler.
## Summary
Fixes "missing FROM-clause entry for table" SQL error when using
`orderByForRecords` with a relation field (e.g. `{ company: { name:
"AscNullsFirst" } }`) in a `groupBy` query.
### Root cause
This is a regression from #18005 (Feb 17). That PR correctly removed a
duplicate `applyOrderToBuilder()` call on the groupBy subquery (which
was conflicting with the `ROW_NUMBER() OVER (... ORDER BY ...)` window
function), but in doing so it also removed the LEFT JOINs that
`applyOrderToBuilder` was adding for relation fields.
After that change, ordering relied solely on `getOrderByRawSQL()` inside
`applyPartitionByToBuilder()`, which builds raw SQL for the window
function but never added the required JOINs. Scalar field ordering (e.g.
`name`, `position`) kept working since those don't need JOINs, but
relation field ordering (e.g. `company.name`) broke.
### Fix
- `getOrderByRawSQL` now returns `relationJoins` alongside the SQL
string so callers get the join info they need
- `applyPartitionByToBuilder` in `GroupByWithRecordsService` adds the
required LEFT JOINs before building the `ROW_NUMBER()` window function,
mirroring the pattern used by `applyOrderToBuilder` in the `findMany`
path
## Test plan
- [x] Manually tested locally with a GraphQL query matching the
customer's failing pattern (`orderByForRecords: [{ company: { name:
"AscNullsFirst" } }]`)
- [x] Added integration tests for ascending and descending relation
field ordering in `group-by-with-records-resolver.integration-spec.ts`
- [x] Typecheck passes
- [x] Lint passes
- [x] CI green
## Summary
- Some production workspaces have `SELECT` or `MULTI_SELECT`
fieldMetadata options that are missing the `id` property (e.g.
`[{"color":"yellow","label":"Draft","value":"DRAFT","position":0},
...]`).
- Adds a new upgrade command
`upgrade:1-20:backfill-select-field-option-ids` that queries all
SELECT/MULTI_SELECT fieldMetadata per workspace, detects options missing
an `id`, and backfills them with a UUID v4.
- The command is idempotent (no-op when all options already have ids),
supports `--dry-run`, and invalidates workspace caches after patching.
