Now that all existing and new workspaces have the following syncStage:
- CALENDAR_EVENT_LIST_FETCH_PENDING
- MESSAGE_LIST_FETCH_PENDING
We can fully deprecate the old FULL_CALENDAR_EVENT_LIST_FETCH_PENDING
and PARTIAL_CALENDAR_EVENT_LIST_FETCH_PENDING (full vs partial is now
directly inferred from the presence of a cursor)
# Introduction
Fixing self relation field creation in v2
- When computing related flat field to delete on object metadata
deletion that has self relation fields
- On self relation creation validation name availability not searching
for the relation target field of the current object field if it's the
field being validated
## Coverage
- added CUD integration testing on self relation fields
close https://github.com/twentyhq/twenty/issues/15153
Resolves [Dependabot Alert
289](https://github.com/twentyhq/twenty/security/dependabot/289) and a
couple other alerts.
Removed types for `imapflow` since the package ships them internally
now. `yarn.lock` has major changes due to an upgraded AWS SDK
`@aws-sdk/client-sesv2` which is used by Nodemailer 7.
- No breaking changes were introduced in imapflow and mailparser.
- Nodemailer's breaking change was dropping the legacy SES transport; we
already use the SMTP transport + our own AWS SES client, so nothing else
needs changing.
## Description
Fixes#15348
The Vercel AI SDK serializes tool schemas with a `jsonSchema` wrapper
(`{ jsonSchema: {...} }`), but the Model Context Protocol specification
expects the schema directly (`{ type: 'object', properties: {...} }`).
This was causing MCP clients like Claude Desktop to silently reject all
tool definitions as invalid, leaving users with no available tools.
## Changes
Modified `handleToolsListing()` in `mcp.service.ts` to unwrap the
`jsonSchema` key before returning tools to MCP clients.
## Before
```json
{
"name": "create_company",
"description": "...",
"inputSchema": {
"jsonSchema": {
"type": "object",
"properties": {...}
}
}
}
```
## After
```json
{
"name": "create_company",
"description": "...",
"inputSchema": {
"type": "object",
"properties": {...}
}
}
```
## Testing
Tested with:
- Direct curl request to verify schema format
- Claude Desktop client via mcp-proxy
- Verified tools now appear correctly in MCP clients
## References
- [MCP Tools
Specification](https://modelcontextprotocol.io/specification/2025-06-18/server/tools)
- Related discussion in #12953
Legacy workspaces still hold the old stored expression, which omits
public.unaccent_immutable, so their tsvectors remain accented and can’t
match the new, unaccented queries. Metadata sync doesn’t touch
asExpression, so only a targeted drop/recreate fixes the underlying
column.
In simpler words, the search vector should contain `mader` instead of
`mäder` for the search to work properly. Therefore, this command
regenerates the search vector across every object that uses
`SEARCH_FIELDS_FOR_*`.
Note that dashboard has a searchVector, but breaks the pattern of using
`SEARCH_FIELDS_FOR_DASHBOARD`. If you look at
packages/twenty-server/src/modules/dashboard/standard-objects/dashboard.workspace-entity.ts:116,
the searchVector field is hard-coded as
```
asExpression: `to_tsvector('english', title)`
```
Therefore, the following code snippet.
```
const storedExpression = hasAsExpressionSetting(
searchVectorFieldMetadata.settings,
)
? searchVectorFieldMetadata.settings.asExpression
: undefined;
if (storedExpression) {
return storedExpression;
}
return undefined;
```
It checks whether the searchVector field already carries its own
asExpression value in metadata. If the settings object includes that
string, it returns it so the upgrade can reuse the existing expression
for objects that aren’t in our predefined lists. If not, it returns
undefined, signaling there’s no stored expression to fall back on.
---------
Co-authored-by: Félix Malfait <felix.malfait@gmail.com>
Co-authored-by: Félix Malfait <felix@twenty.com>
This PR implements Sentry's AI agent monitoring by:
- Configuring vercelAIIntegration with recordInputs and recordOutputs
options
- Adding sendDefaultPii to Sentry.init() for better debugging
- Creating a shared AI_TELEMETRY_CONFIG constant to DRY up the telemetry
configuration
- Adding experimental_telemetry to all AI SDK calls (generateText,
generateObject, streamText)
All AI operations are now fully monitored in Sentry with complete
input/output recording for debugging and performance analysis.
Note: Currently on Sentry v9.26.0, which is compatible with this
implementation. No breaking changes from the v9-to-v10 migration guide
were found in the codebase.
Instead of declaring packages at the workplace-level package.json,
moving them into their relevant package-level package.json file.
`twenty-front`, `twenty-server` and `twenty-emails` continue to build
and work fine because of hoisting, but the dependencies now follow the
internal strategy of declaring at the package-level, plus give us a
single source of truth to updating package versions.
`twenty-front` and `twenty-emails` only use `@react-email/components`
while `twenty-server` only depends on `@react-email/render`.
# Introduction
Fixing composite field update by computing field column type for each of
its properties instead of globally
## Coverage
Added integration tests for each composite field on both successful
`create` and `update`
```ts
Test Suites: 17 passed, 17 total
Tests: 104 passed, 104 total
Snapshots: 14 passed, 14 total
Time: 135.431 s, estimated 143 s
```
## Conlusion
Related to https://github.com/twentyhq/core-team-issues/issues/1753
## Context
When updating both enum options and defaultValue, the old default might
not be in the new options (or vice versa), causing PostgreSQL constraint
violations regardless of update order.
## Solution
Sort updates to process defaultValue last; before updating options,
temporarily set the new defaultValue in metadata so alterEnumValues
creates the column with the correct default, then skip the redundant
defaultValue update handler.
# Introduction
Adding a view field to a view in v2 would be optimistically rendered by
the front but on refresh would not get persisted.
That's because we cache both:
```ts
useCachedMetadata({
cacheGetter: cacheStorageService.get.bind(cacheStorageService),
cacheSetter: cacheStorageService.set.bind(cacheStorageService),
operationsToCache: ['ObjectMetadataItems', 'FindAllCoreViews'],
}),
```
With keys that look like:
```ts
return `graphql:operations:${operationName}:${workspace.id}:${workspaceMetadataVersion}:${locale}:${queryHash}`;
```
It was functional in v1 as we would be incrementing metadata version
often.
In v2 it gets incremented only if implies an interaction to metadata
object or fields ( will be deprecated in the future though, until we
finish the // run )
The fix was to check if an `view` or related has been processed in the
workspace migration or if we incremented the metadata in order to
manually flush the `findAllCoreViews` redis cache.
# Introduction
Refactoring the standard overrides dispatcher to only pass over fields
to has to be dispatched in the standardOverrides entry and let the other
side effects resulting from out of standard overrides mutation trigger
Related to https://github.com/twentyhq/core-team-issues/issues/1753
## This allows
- standard field settings, options etc updates and so on
## Remark
- Determine what we should do on object deactivation ( right now in
production we can still access deactivated object relation properties
and so on e.g deactivate opportunities still accessible from a view
field on company ( still have to re-create it as it has been deleted )
=> decided to leave as it is right now, `isActive` could be considered
as uiDeactivated in the end
- We should also add forbidden standard field mutations validation
inside the builder itself ( here we want to early return in the api
input transpiler too as we don't want to spread invalid side effects )
=> or in the end we could just centralize both but it will generate
several errors
## Coverage
```ts
PASS test/integration/metadata/suites/object-metadata/successful-update-one-standard-object-metadata.integration-spec.ts
PASS test/integration/metadata/suites/field-metadata/successful-update-one-standard-field-metadata.integration-spec.ts
PASS test/integration/metadata/suites/object-metadata/failing-update-one-standard-object-metadata.integration-spec.ts
PASS test/integration/metadata/suites/field-metadata/failing-update-one-standard-field-metadata.integration-spec.ts
Test Suites: 4 passed, 4 total
Tests: 18 passed, 18 total
Snapshots: 16 passed, 16 total
Time: 8.721 s, estimated 10 s
```
## Update post review
Faced a behavior where updating back the company label to its original
value would result in storing this value in the standard overrides
Refactored both field and object transpilation behavior to rather remove
the standard override value instead and let fallback on original value
Yes it's quite duplicated will factorize once we move this inside the
builder
Done ⬇️
Gql : move delete and destroy logic to common
Rest :
- rename 'delete' handler to 'destroy'
- create soft delete handlers (named 'delete') : one and many
- create destroy many handler
- update doc
--> Rest api gains NEW soft delete one/many + destroy many capabilities
closes : https://github.com/twentyhq/core-team-issues/issues/1579
Adds intelligent routing system that automatically selects the best
agent for user queries based on conversation context.
### Changes:
- Added `routerModel` column to workspace table for configurable router
LLM selection
- Implemented `RouterService` with conversation history analysis and
agent matching logic
- Created router settings UI in AI Settings page with model dropdown
- Removed agent-specific thread associations - threads are now
agent-agnostic
- Added real-time routing status notification in chat UI with shimmer
effect
- Removed automatic default assistant agent creation
- Renamed GraphQL operations from agent-specific to generic (e.g.,
`agentChatThreads` → `chatThreads`)
---------
Co-authored-by: Félix Malfait <felix.malfait@gmail.com>
Co-authored-by: Félix Malfait <felix@twenty.com>
- Use aggregate operations in the widget configuration instead of
extended aggregate operations
- Use aggregate operation from generated graphql in the frontend
