## Summary
- **Fix token renewal endpoint**: Use `/metadata` instead of `/graphql`
for token renewal in agent chat, fixing auth issues
- **Improve tool display**: Add `load_skills` support, show formatted
tool names (underscores → spaces) with finish/loading states, display
tool icons during loading, and support custom loading messages from tool
input
- **Refactor workflow agent management**: Replace direct
`AgentRepository` access with `AgentService` for create/delete/find
operations in workflow steps, improving encapsulation and consistency
- **Simplify Apollo client usage**: Remove explicit Apollo client
override in `useGetToolIndex`, add `AgentChatProvider` to
`AppRouterProviders`
- **Fix load-skill tool**: Change parameter type from `string` to `json`
for proper schema parsing
- **Update agent-chat-streaming**: Use `AgentService` for agent
resolution and tool registration instead of direct repository queries
## Test plan
- [ ] Verify AI agent chat works end-to-end (send message, receive
response)
- [ ] Verify tool steps display correctly with icons and proper messages
during loading and after completion
- [ ] Verify workflow AI agent step creation and deletion works
correctly
- [ ] Verify workflow version cloning preserves agent configuration
- [ ] Verify token renewal works when tokens expire during agent chat
Made with [Cursor](https://cursor.com)
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: cubic-dev-ai[bot] <191113872+cubic-dev-ai[bot]@users.noreply.github.com>
## Add application-scoped GraphQL schema generation
When an application token is used to authenticate, the `/graphql` schema
is now dynamically filtered to only include entities belonging to that
application (plus the Twenty Standard Application). This enables
third-party applications and the SDK to introspect a schema that is
relevant to their scope, rather than seeing the full workspace schema
with all custom objects.
### Changes
- **New `generateApplicationToken` mutation** on the `/metadata`
endpoint, allowing callers to exchange an API key for an
application-scoped JWT token
- **Schema filtering by application** in `WorkspaceSchemaFactory` — when
`request.application` is present (from an application token), flat
entity maps are filtered by `[appId, standardAppId]` before schema
generation
- **Per-app caching** — both the Yoga in-memory cache and Redis cache
now include the `appId` in their keys to avoid serving wrong schemas
- **Consolidated `getSubFlatEntityMapsByApplicationIdsOrThrow`** —
unified the single-ID and multi-ID filtering utilities into one
- **Integration tests** covering token generation (admin + API key auth)
and schema introspection filtering (standard app token excludes custom
objects)
Schema generated on seeds with applicationToken (see that pets is
missing)
<img width="782" height="994" alt="image"
src="https://github.com/user-attachments/assets/82510031-0965-435d-bc26-77c9f5d74e1f"
/>
⚠️ **AI-generated PR — not ready for review** ⚠️
cc @FelixMalfait
---
## Changes
### System prompt improvements
- Explicit skill-before-tools workflow to prevent the model from calling
tools without loading the matching skill first
- Data efficiency guidance (default small limits, use filters)
- Pluralized `load_skill` → `load_skills` for consistency with
`load_tools`
### Token usage reduction
- Output serialization layer: strips null/undefined/empty values from
tool results
- Lowered default `find_*` limit from 100 → 10, max from 1000 → 100
### System object tool generation
- System objects (calendar events, messages, etc.) now generate AI tools
- Only workflow-related and favorite-related objects are excluded
### Context window display fix
- **Bug**: UI compared cumulative tokens (sum of all turns) against
single-request context window → showed 100% after a few turns
- **Fix**: Track `conversationSize` (last step's `inputTokens`) which
represents the actual conversation history size sent to the model
- New `conversationSize` column on thread entity with migration
### Workspace AI instructions
- Support for custom workspace-level AI instructions
---------
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
## Context
Deprecating legacy ObjectMetadata from cache in favor of flat entities.
Introducing utils to build byName/byNameSingular/byNamePlural in
isolated cases
## Next
- I had to introduce a util to build from flat to legacy
objectMetadataMaps, we should instead use flat maps directly when needed
(datasource, schema generation, etc)
- Deprecate metadata version in the cache
- Use the new cache strategy for flat entities with permissions and
feature flags and inject in the global datasource context
Following https://github.com/twentyhq/twenty/pull/14869
Dynamically clearing the cache, as one entry was forgotten
It seems like a permission test deletes a role that's expected to be
existing in following tests, as now cache is getting invalidated tests
are failing
Seems to be related to
https://discord.com/channels/1130383047699738754/1423768505911869460
## Singleton local cache key collision
When set for the first time caches local keys looks like
`workspaceId:undefined", singleton is shared between several cache
instances
If the given key is undefined it will read on other entity cache entry.
## Improvements
- Add logs to all gql operations and rest calls to help debug CPU issues
on the backend. These are temporary and should be removed
- Remove nested relations from workflowVersions load (used to add manual
triggers in the side bar). On some workspaces this call result in a
response of 4MB which is heavy on CPU
- Investigated Redis Usage ==> made a few improvements, we are should
still migrate to the new cache service once available
- investigated db calls in messaging / calendar fetch list + workflow
enqueue run cron jobs. Everything seems to be properly batched
## Context
We are experiencing bad performance on Twenty. One of the root cause
hypothesis is that computing `currentUser.currentWorkspace.views` is CPU
consuming. Without views, the GetCurrentUser response is ~1000 lines.
With its ~10000 lines.
As graphql is going through all fields recursively this can be quite
heavy on CPU. We had a similar issues on ObjectMetadataItems 2 years ago
and came with storing the response in redis.
Note: I thought there was also a cache in RAM but this is not the case,
so to invalidate the cache we can just empty redis.
## How
- Extract getting all views from GetCurrentUser and update frontend to
perform both queries
- Add views to cached graphql operations
- invalidate the cache manually on view or related core entities update
/ create / delete / destroy
## Tests
I have tested a lot on v1
In this PR, I'm fixing a bug introduced in recent performance work on
the cache.
Bug context: https://github.com/twentyhq/twenty/issues/12865
Related PR opened by a contributor:
https://github.com/twentyhq/twenty/pull/13003
## Root cause
We cache all objectMetadataItems at graphql level : see
`useCachedMetadata` hook:
- instead of going through the regular resolvers, we direlcty load data
from the cache. However this data must be localized regarding labels and
descriptions
In a precedent refactoring, we introduced the notion of locale in the
cache key. However, the user locale was not properly taken into account
as we did not have the information in this hook.
## Fix
1. **Introduce locale in userWorkspace entity**. The locale is stored on
workspaceMember in each postgres workspaceSchema (workspace_xxx) which
is the alter ego of userWorkspace in postgres core schema. Note that we
can't store it in user as a user can be part of multiple workspaces (the
locale already there must be seen as a default for this user), and we
cannot rely on workspaceMember as we would need to query the
workspaceSchema in the authentication layer which we want to avoid for
performance reasons.
2. During request hydration from token (containing the userWorkspaceId),
we fetch the userWorkspace and store it in the Request (this impact both
AuthContext and Request interface)
3. Leverage userWorkspace.locale in the useCachedMetadata hook
## Additional notes
There is no need to change the way we store and retrieve the
object-metadata-maps object itself which is different from the graphql
layer cache. object-metadadata-maps are not localized
In this PR:
## Improve recompute metadata cache performance. We are aiming for
~100ms
Deleting relationMetadata table and FKs pointing on it
Fetching indexMetadata and indexFieldMetadata in a separate query as
typeorm is suboptimizing
## Remove caching lock
As recomputing the metadata cache is lighter, we try to stop preventing
multiple concurrent computations. This also simplifies interfaces
## Introduce self recovery mecanisms to recompute cache automatically if
corrupted
Aka getFreshObjectMetadataMaps
## custom object resolver performance improvement: 1sec to 200ms
Double check queries and indexes used while creating a custom object
Remove the queries to db to use the cached objectMetadataMap
## reduce objectMetadataMaps to 500kb
<img width="222" alt="image"
src="https://github.com/user-attachments/assets/2370dc80-49b6-4b63-8d5e-30c5ebdaa062"
/>
We used to stored 3 fieldMetadataMaps (byId, byName, byJoinColumnName).
While this is great for devXP, this is not great for performances.
Using the same mecanisme as for objectMetadataMap: we only keep byIdMap
and introduce two otherMaps to idByName, idByJoinColumnName to make the
bridge
## Add dataloader on IndexMetadata (aka indexMetadataList in the API)
## Improve field resolver performances too
## Deprecate ClientConfig
# Introduction
Added a no-explicit-any rule to the twenty-server, not applicable to
tests and integration tests folder
Related to https://github.com/twentyhq/core-team-issues/issues/975
Discussed with Charles
## In case of conflicts
Until this is approved I won't rebased and handle conflict, just need to
drop two latest commits and re run the scripts etc
## Legacy
We decided not to handle the existing lint error occurrences and
programmatically ignored them through a disable next line rule comment
## Open question
We might wanna activate the
[no-explicit-any](https://typescript-eslint.io/rules/no-explicit-any/)
`ignoreRestArgs` for our use case ?
```
ignoreRestArgs?: boolean;
```
---------
Co-authored-by: etiennejouan <jouan.etienne@gmail.com>
This is a first PR to remove old relation logic
Next steps:
- remove relationMetadata from cache
- remove relationMetadata table content and structure
- refactor relationDefinition to leverage field.settings instead
In this PR we are
1. cleaning typeORM service by removing connectToDataSource method
2. using workspaceDataSource instead of mainDataSource when possible,
and replacing raw SQL with workspaceRepository methods to use
In this PR we are
- introducing a cached map `{ userworkspaceId: roleId } `to reduce calls
to get a userWorkspace's role (we were having N+1 around that with
combinedFindMany queries and generally having a lot of avoidable
queries)
- using the roles permissions cache (`{ roleId: { objectNameSingular:
{ canRead: bool, canUpdate: bool, ...} } `) in Permissions V1's
userHasObjectPermission, in order to 1) improve performances to avoid
calls to get roles 2) start using our permissions cache
In this PR we are
- (if permissionsV2 is enabled) executing permission checks at query
builder level. To do so we want to override the query builders methods
that are performing db calls (.execute(), .getMany(), ... etc.) For now
I have just overriden some of the query builders methods for the poc. To
do so I created custom query builder classes that extend typeorm's query
builder (selectQueryBuilder and updateQueryBuilder, for now and later I
will tackle softDeleteQueryBuilder, etc.).
- adding a notion of roles permissions version and roles permissions
object to datasources. We will now use one datasource per roleId and
rolePermissionVersion. Both rolesPermissionsVersion and rolesPermissions
objects are stored in redis and recomputed at role update or if queried
and found empty. Unlike for metadata version we don't need to store a
version in the db that stands for the source of truth. We also don't
need to destroy and recreate the datasource if the rolesPermissions
version changes, but only to update the value for rolesPermissions and
rolesPermissionsVersions on the existing datasource.
What this PR misses
- computing of roles permissions should take into account
objectPermissions table (for now it only looks at what's on the roles
table)
- pursue extension of query builder classes and overriding of their db
calling-methods
- what should the behaviour be for calls from twentyOrmGlobalManager
that don't have a roleId?
## Context
This fix ensures that even if a datasource creation promise throws and
is cached, subsequent requests won't return that cached exception.
Also adding a TTL on MetadataObjectMetadataOngoingCachingLock, this is
not something that should stay in the cache forever and could
potentially unlock some race conditions (the origin of the issue is
probably due to performances where the lock is not removed as it should
be after metadata computation and caching)
Fix https://github.com/twentyhq/core-team-issues/issues/295
Based on the feature-flag `IsNewRelationEnabled` the schema will be
marked as outdated and regenerated, this will cause an error on the
front-end on the first request on the following ones schema will be well
generated ans request will work.
## Context
We recently added an infinite ttl to metadata version however other keys
such as the object metadata map are also linked to the version so
keeping the version in cache without the rest breaks the app. I'm
editing all engine related keys with infinite ttl
In this PR:
- removing ugprade-0.24 commands as we are releasing 0.30
- introducing cache:flush command
- refactoring upgrade command and sync-metadata command to use the
ActiveWorkspacesCommand so they consistently run on all workspaces or
selected workspaces
Fixes:
- clear localStorage on sign out
- fix missing workspaceMember in verify resolver
- do not throw on datasource already destroyed exception which can
happen with race condition when several resolvers are resolving in
parallel
In this PR:
1. Refactor guards to avoid duplicated queries: WorkspaceAuthGuard and
UserAuthGuard only check for existence of workspace and user in the
request without querying the database
This PR introduces the following changes:
- add the metadataVersion to all our metadata cache keys to ease
troubleshooting:
<img width="1146" alt="image"
src="https://github.com/user-attachments/assets/8427805b-e07f-465e-9e69-1403652c8b12">
- introduce a cache recompute lock to avoid overloading the database to
recompute the cache many time
The code removed in the PR was flagged as unused by the JetBrains
inspector.
I did a QA on the dev environment but other checks are highly
recommended.
There is one commit by scope to make the review easier.
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
## Context
As we grow, the messaging scripts are experiencing performance issues
forcing us to temporarily disable them on the cloud.
While investigating the performance, I have noticed that generating the
entity schema (for twentyORM) in the repository is taking ~500ms locally
on my Mac M2 so likely more on pods. Caching the entitySchema then!
I'm also clarifying naming around schemaVersion and cacheVersions ==>
both are renamed workspaceMetadataVersion and migrated to the workspace
table (the workspaceCacheVersion table is dropped).
Calling `getObjectMetadata` from `WorkspaceCacheStorageService` in every
query was causing big performance issues. The `objectMetadataCollection`
is now part of the `WorkspaceInternalContext` so we only instance it
once in the `WorkspaceDatasourceFactory`.
Queries are now much faster, for instance for TimelineCalendar, it went
from ~450ms to 80ms.
### Overview
This PR builds upon #5153, adding the ability to get a repository for
custom objects. The `entitySchema` is now generated for both standard
and custom objects based on metadata stored in the database instead of
the decorated `WorkspaceEntity` in the code. This change ensures that
standard objects with custom fields and relations can also support
custom objects.
### Implementation Details
#### Key Changes:
- **Dynamic Schema Generation:** The `entitySchema` for standard and
custom objects is now dynamically generated from the metadata stored in
the database. This shift allows for greater flexibility and
adaptability, particularly for standard objects with custom fields and
relations.
- **Custom Object Repository Retrieval:** A repository for a custom
object can be retrieved using `TwentyORMManager` based on the object's
name. Here's an example of how this can be achieved:
```typescript
const repository = await this.twentyORMManager.getRepository('custom');
/*
* `repository` variable will be typed as follows, ensuring that standard
fields and relations are properly typed:
* const repository: WorkspaceRepository<CustomWorkspaceEntity & {
* [key: string]: any;
* }>
*/
const res = await repository.find({});
```
Fix#6179
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
Co-authored-by: Weiko <corentin@twenty.com>
Charles Bochet