- apollo enrich application (via OAuth 2)
- add applicationId to var env in logic function executor
- update `getDefaultUrl` logic
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
<img width="450" height="212" alt="Capture d’écran 2026-03-03 à 11 41
54"
src="https://github.com/user-attachments/assets/b2c29a48-7dc0-4b16-a085-8f305d21f7ca"
/>
New status `FAIL_SAFE` added. This status propagates to the following
nodes until reaching the iterator, that will start the new iteration.
The difference with `SKIP` is that, when the parent nodes have at least
one `FAIL_SAFE`, it becomes `FAIL_SAFE` too. While a parent 1 `SKIP` +
parent 2 `SUCCESS` => to be executed.
I also thought about just going back to the iterator as a break would,
but since we have branches, it may lead to inconsistent statuses with
parallel updates.
# Introduction
We need to build and validate the flat entity operation in the following
order delete update and create
For example if not, if a created field has the same name than a deleted
one than it will fail whereas it should not
This PR adds an explicit role selector to the "Invite by email" flow,
requires a role choice before sending, and stores the selected role with
each invitation. The backend now accepts and persists `roleId` on
invitations and applies it when the invite is accepted, while keeping it
optional to avoid breaking existing clients and legacy invites.
---
### Frontend
- **Settings → Members → Invite by email**
- New **Role** dropdown (same `Select` pattern as member/API key role
selectors) between the email input and Invite button.
- Roles are loaded via `SettingsRolesQueryEffect` and
`settingsAllRolesSelector`; only roles with `canBeAssignedToUsers` are
shown.
- Role is **required**: form validates `roleId` (e.g.
`z.string().min(1)`) and the Invite button is disabled until a role is
selected and emails are valid.
- `WorkspaceInviteTeam` receives `roles` as a prop from the parent;
layout is responsive (e.g. stacked on small viewports).
- **Pending invitations table**
- New **Role** column showing the invitation’s role label (or "Unknown
role" for legacy invites without `roleId`), using the same roles source
for lookup.
- **Onboarding invite step**
- When sending invites during onboarding, the workspace **default role**
is used when available (`currentWorkspace?.defaultRole?.id`), so no role
selector is added there.
- **GraphQL**
- `sendInvitations` mutation accepts optional `roleId`;
`findWorkspaceInvitations` and resend mutation responses include
`roleId` on `WorkspaceInvitation`. Frontend types (e.g.
`WorkspaceInvitation`, hook variables) updated accordingly.
---
### Backend
- **API**
- `SendInvitationsInput` has an **optional** `roleId` (UUID, nullable).
The resolver normalises `null` to `undefined` so existing callers and
legacy flows are not broken.
- **Validation (when `roleId` is provided)**
- Role checks are centralised in **RoleValidationService**
(`RoleValidationModule`, in `metadata-modules/role-validation/`). It
validates that the role exists in the workspace and has
`canBeAssignedToUsers`, and throws a permissions-style error otherwise.
This avoids circular dependencies (e.g. `RoleModule` imports
`UserWorkspaceModule`, so invite/accept flows cannot depend on
`RoleModule`).
- **Send flow:** `WorkspaceInvitationResolver` and
`WorkspaceInvitationService.sendInvitations` both call
`RoleValidationService.validateRoleAssignableToUsersOrThrow` when
`roleId` is present (resolver before calling the service; service again
before creating tokens so that **resend** also validates the stored role
and fails fast if the role was deleted or made unassignable).
- **Accept flow:**
`UserWorkspaceService.addUserToWorkspaceIfUserNotInWorkspace` uses the
same service in `resolveRoleIdForNewMember` when an invitation provides
a `roleId`, then falls back to `workspace.defaultRoleId` when not.
Role/default is resolved and validated before any user/workspace/member
creation.
- **Persistence**
- Invitation app tokens store `roleId` in `context` next to `email`
(`context: { email, roleId? }`). `generateInvitationToken` and
`createWorkspaceInvitation` accept an optional `roleId` and only add it
to `context` when defined.
- **Resend**
- Resend passes the existing invitation’s `context.roleId` into
`sendInvitations`. The service validates that role (when present) before
creating the new token, so if the role was deleted or made unassignable,
resend fails with a clear error instead of sending a broken link.
- **Response shape**
- `SendInvitationsOutput.result` remains `WorkspaceInvitation[]`. When
`usePersonalInvitation` is false we only push full invitation records
(from `castAppTokenToWorkspaceInvitationUtil`), so the result always
matches the GraphQL type (`id`, `email`, `roleId`, `expiresAt`).
- **Modules**
- `WorkspaceInvitationModule` and `UserWorkspaceModule` import
**RoleValidationModule** (not `RoleModule`) and inject
**RoleValidationService** for validation. `RoleModule` imports
`RoleValidationModule` and `RoleService` delegates to
`RoleValidationService` for the same validation where the module graph
allows.
---
### Backward compatibility
- **Optional `roleId`**: Clients that don’t send `roleId` (or send
`null`) are unchanged; invitations are created without a role and the
accept flow uses the workspace default role.
- **Legacy invitations**: App tokens with only `context.email` still
work; `context.roleId` is optional and the UI can show e.g. "Unknown
role" for those in the pending-invitations table.
## Context
Part 1 of migrating gridPosition in favor of typed position
FE should now always send both values to the BE and use both.
Next steps:
- Update the backend to enforce and validate the new position field + DB
migrations gridPositon -> position (type: GRID)
- Cleanup frontend usage
- Cleanup backend
If-else branches cannot be recreated once deleted. Only else-if branches
can. On if-else branches removal, we now remplace the node by an empty
node instead of only deleting
Also fixing nested if-else.
## Summary
Follow-up to #18267. Hardens the OAuth implementation with security
fixes identified during audit:
**P0 — Critical:**
- Bind authorization codes to `client_id` in context to prevent auth
code injection (RFC 6749 §4.1.3)
- Store PKCE `code_challenge` directly in auth code context instead of a
separate `CodeChallenge` token — cryptographically binds the challenge
to its code
- Enforce `code_verifier` when `code_challenge` was used during
authorization
- Hash authorization codes (SHA-256) before storage to prevent exposure
if DB is compromised
- Add `Cache-Control: no-store` + `Pragma: no-cache` headers on token
responses (RFC 6749 §5.1)
- Add rate limiting on `/oauth/token` endpoint (20 req/min per client
via existing `ThrottlerService`)
**P1 — High:**
- Return HTTP 401 for `invalid_client` errors instead of 400 (RFC 6749
§5.2)
- Verify refresh tokens belong to the presenting client (cross-client
token theft prevention)
- Limit fields exposed by public `findApplicationRegistrationByClientId`
query to only what the frontend needs (`id`, `name`, `logoUrl`,
`websiteUrl`, `oAuthScopes`)
- Require `API_KEYS_AND_WEBHOOKS` permission for
`createApplicationRegistration` mutation
**P2/P3 — Medium/Low:**
- Add error handling and loading states to frontend Authorize page
- Rename redirect URL param from `authorizationCode` to `code` (RFC
standard)
- Add unit tests for `validateRedirectUri` utility (8 test cases)
## Test plan
- [ ] Existing OAuth integration tests updated for all changes (hashed
codes, context-based PKCE, client binding, 401 status codes, cache
headers)
- [ ] New test: auth code rejected when presented by a different client
- [ ] New test: refresh token rejected when presented by a different
client
- [ ] New test: `code_verifier` required when PKCE was used in
authorization
- [ ] New test: `Cache-Control: no-store` header present on responses
- [ ] New unit tests for `validateRedirectUri` (HTTPS, localhost,
fragments, invalid URIs)
- [ ] Verify frontend authorize page shows errors gracefully
Made with [Cursor](https://cursor.com)
- Fixes self host application
- add new telemetry information
- add serverId to identify a server instance
- remove .twenty from git tracking
- tree-shake "twenty-sdk" usage in built logic functions and front
components
- fix "twenty-sdk" version usage
- fix twenty-zapier cli
---------
Co-authored-by: cubic-dev-ai[bot] <191113872+cubic-dev-ai[bot]@users.noreply.github.com>
Adds color support for navigation menu items.
---------
Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
Co-authored-by: Devessier <baptiste@devessier.fr>
Co-authored-by: Charles Bochet <charles@twenty.com>
## Summary
Add Home/Chat tabs and a dedicated threads list in the navigation
drawer.
## Changes
- **Navbar tabs:** Tabs in the drawer to switch between Home and Chat
(with “New chat” button). Shown on desktop when expanded and on mobile
below the workspace selector.
- **Navbar threads list:** New `NavigationDrawerAIChatThreadsList` for
the Chat tab with date groups (Today / Yesterday / Older), thread rows
as `NavigationDrawerItem` (IconComment, title, timestamp). Shared
`useAIChatThreadClick` hook used by navbar and command menu; navbar
passes `resetNavigationStack: true`.
- **NavigationDrawerItem:** New `alwaysShowRightOptions` prop so the
timestamp is always visible (no hover-only).
---------
Co-authored-by: Etienne <45695613+etiennejouan@users.noreply.github.com>
Co-authored-by: Félix Malfait <felix@twenty.com>
Co-authored-by: Charles Bochet <charles@twenty.com>
This PR pgrades all BlockNote packages (@blocknote/core,
@blocknote/react, @blocknote/mantine, @blocknote/server-util,
@blocknote/xl-docx-exporter, @blocknote/xl-pdf-exporter) to 0.47.0 and
adapts the codebase to the new API.
### Changes
- Dependency upgrades: Bumped all BlockNote packages to 0.47.0, added
required Mantine v8 peer dependencies, removed unnecessary prosemirror
resolutions
- Formatting toolbar: Replaced the manual reimplementation of
FormattingToolbarController (which handled visibility, positioning,
portal rendering, text-alignment-based placement, and a
dangerouslySetInnerHTML transition trick) with BlockNote's built-in
FormattingToolbarController. The toolbar buttons themselves are
unchanged.
- Side menu: Replaced manual drag handle menu positioning and rendering
(DashboardBlockDragHandleMenu, DashboardBlockColorPicker, and their
floating configs) with BlockNote's built-in SideMenuController,
DragHandleButton, and DragHandleMenu components. Deleted 4 files that
became dead code.
- Extension API migration: Replaced deprecated editor.suggestionMenus
and editor.formattingToolbar APIs with the new extension system
(SuggestionMenu, useExtensionState, editor.getExtension())
- Slash menu fixes: Filtered out BlockNote's new default "File" item
(added in 0.47) to avoid duplicates with our custom one; added icon
mappings for new block types (Toggle List, Divider, Toggle Headings,
Headings 4-6)
- Server-side: Switched @blocknote/server-util to dynamic import() to
handle ESM-only transitive dependencies in CJS context
# Introduction
Resolving each create|updated|deleted entities related entities
application through their universal foreingKey ( silently failing if not
found leaving the validator handling that )
In order to compute all the required application in the dependency flat
entity maps
## Tested
Creating a field on a custom local twenty-apps installed on the
workspace + view field
## Out of scope
- updated snapshot
- update graphql generated front
## Automated fix for [bug 5360](https://sonarly.com/issue/5360?type=bug)
**Severity:** `critical`
### Summary
When an existing user created via OAuth (with no password hash) attempts
to sign in through the signInUp password flow, their null passwordHash
is passed directly to bcrypt.compare, which throws an unhandled 'data
and hash arguments required' error instead of a user-friendly message.
### User Impact
Users who originally registered via Google or Microsoft OAuth and then
attempt to sign in using email/password through the signInUp flow cannot
authenticate. They see a server error instead of a helpful message like
'User was not created with email/password'. This blocks the user from
accessing the CRM entirely.
### Root Cause
Proximate cause: bcrypt.compare() at auth.util.ts:19 throws 'data and
hash arguments required' because the passwordHash argument is null.
1. Why did bcrypt throw? Because compareHash() was called with a null
passwordHash value. The compareHash function at auth.util.ts:19 passes
its arguments directly to bcrypt.compare without any null check.
2. Why was passwordHash null? Because SignInUpService.validatePassword()
at sign-in-up.service.ts:154 received a null passwordHash from
AuthService.validatePassword() at auth.service.ts:232. The value
userData.existingUser.passwordHash is null for users who were originally
created via an OAuth provider (Google, Microsoft) and never set a
password.
3. Why was there no null check before calling bcrypt? Because
AuthService.validatePassword() at line 229-233 does not check whether
userData.existingUser.passwordHash is null before passing it to
signInUpService.validatePassword(). The TypeScript type annotation says
passwordHash is string, but the UserEntity column is declared nullable:
true (user.entity.ts:73), so the runtime value can be null despite the
type system.
4. Why does this code path lack the null check when another path has it?
The validateLoginWithPassword() method at auth.service.ts:177 correctly
checks if (!user.passwordHash) and throws a user-friendly AuthException
'Incorrect login method'. This guard was added by Thomas Trompette on
2024-08-07 (commit 2abb6adb61). However, the signInUp flow's
validatePassword() method was introduced later by Antoine Moreaux on
2025-03-17 (commit bda835b9f8) without replicating this same defensive
check.
5. Why was this not caught? The signInUp validatePassword method was
written assuming the existingUser would always have a passwordHash when
the auth provider is Password. This assumption is incorrect because the
signInUp flow can match an existing OAuth-created user (who has no
passwordHash) when someone tries to sign up with email/password using an
email already registered via OAuth. No test covers this cross-provider
scenario in the signInUp path.
Root cause: The AuthService.validatePassword() method
(auth.service.ts:229-233) in the signInUp code path is missing a null
guard on userData.existingUser.passwordHash before passing it to
bcrypt.compare. The fix should check for null/undefined passwordHash and
throw a descriptive AuthException, mirroring the existing guard in
validateLoginWithPassword at line 177.
**Introduced by:** Antoine Moreaux on 2025-03-17 in commit
[`bda835b`](https://github.com/twentyhq/twenty/commit/bda835b9f82)
### Suggested Fix
Added a null check on userData.existingUser.passwordHash inside the
validatePassword private method of AuthService, immediately before
calling signInUpService.validatePassword. When passwordHash is null or
undefined (i.e. the user was created via OAuth and never set a
password), an AuthException with code INVALID_INPUT is thrown with the
user-friendly message 'User was not created with email/password'. This
mirrors the identical guard that already exists in
validateLoginWithPassword at line 177, preventing bcrypt.compare from
receiving a null argument and crashing with an unhandled error.
### Evidence
- **Code:**
[packages/twenty-server/src/engine/core-modules/auth/services/auth.service.ts:229
- if (userData.type === 'existingUser')
{](https://github.com/twentyhq/twenty/blob/main/packages/twenty-server/src/engine/core-modules/auth/services/auth.service.ts#L229)
---
*Generated by [Sonarly](https://sonarly.com)*
Co-authored-by: Sonarly Claude Code <claude-code@sonarly.com>
## Summary
- **Admin-level filtering**: New AI tab in admin panel with server-wide
model availability controls (whitelist/blacklist via
`AI_AUTO_ENABLE_NEW_MODELS`, `AI_DISABLED_MODEL_IDS`,
`AI_ENABLED_MODEL_IDS` config variables). Dedicated
`setAdminAiModelEnabled` mutation replaces frontend config-variable
manipulation. Filter dropdown to show/hide unconfigured and deprecated
models.
- **Workspace-level filtering**: Per-workspace controls with "Use best
models only" mode (curated list backed by `isRecommended` flag), or
custom whitelist/blacklist. Separate Smart/Fast model selectors with
"Best (...)" virtual options.
- **Security enforcement**: Both layers enforced at every backend
execution point — workspace update, agent create/update, chat execution.
Model ID validated against known models before config mutation. All
admin endpoints protected by `AdminPanelGuard`.
## Changes
### Backend (`twenty-server`)
- New config variables for admin-level model filtering
- `AiModelRegistryService`: `getAllModelsWithStatus()`,
`setModelAdminEnabled()` with model ID validation,
`isModelAdminAllowed()`
- `AdminPanelResolver`: `getAdminAiModels` query,
`setAdminAiModelEnabled` mutation
- `WorkspaceEntity`: new fields (`autoEnableNewAiModels`,
`disabledAiModelIds`, `enabledAiModelIds`, `useRecommendedModels`)
- `WorkspaceService`: model validation on `smartModel`/`fastModel`
updates
- `AgentResolver`: model availability checks on create/update
- `isModelAllowedByWorkspace` centralized utility
- `isRecommended` flag on model definitions
- Two TypeORM migrations
### Frontend (`twenty-front`)
- New `SettingsAdminAI` component with search, filter dropdown
(unconfigured/deprecated), and model toggle cards
- AI tab added to admin panel navigation
- `useWorkspaceAiModelAvailability` hook for workspace-level filtering
- `SettingsAIModelsTab` redesigned: merged sections, "Use best models
only" toggle, conditional available models list
- `getModelIcon`/`getModelProviderLabel` shared utilities with GraphQL
enum casing normalization
- Updated generated GraphQL types and mock data
## Test plan
- [ ] Toggle models on/off in admin panel AI tab and verify they
appear/disappear in workspace settings
- [ ] Enable "Use best models only" in workspace settings and verify
only recommended models are selectable
- [ ] Disable recommended mode and verify whitelist/blacklist toggles
work correctly
- [ ] Verify deprecated models hidden by default, shown greyed out when
filter enabled
- [ ] Verify unconfigured models hidden by default, shown disabled when
filter enabled
- [ ] Try setting a disabled model as Smart/Fast model — should be
rejected
- [ ] Try creating an agent with a disabled model — should be rejected
- [ ] Verify admin panel AI tab requires admin access
Made with [Cursor](https://cursor.com)
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
## Summary
- **Refactor frontend metadata loading architecture**: Split the
monolithic `EagerMetadataLoadEffect` into focused provider effects
(`UserMetadataProviderEffect`, `ObjectMetadataProviderEffect`,
`ViewMetadataProviderEffect`) orchestrated by `MetadataProviderEffects`.
Replaced `UserProvider` + `ObjectMetadataItemsProvider` with a single
`MetadataGater` that gates rendering on `isAppMetadataReadyState`. The
metadata store now validates view-object consistency before promoting
views, and `updateDraft` skips no-op updates via deep equality checks.
- **SDK CLI improvements**: Added `app:typecheck` command, improved
error handling in API sync (extracts GraphQL error messages), added
`serializeError` utility for human-readable error output, added `error`
file status to dev mode orchestrator with UI support, and fixed
ClickHouse migration/seed commands to use `transpile-only`.
The `useSaveFieldsWidgetGroups` hook was making multiple sequential API
calls per widget (create groups, delete groups, update groups, update
fields) — non-atomic, chatty, and with heavy diff computation on the
frontend.
## Backend
- **New input DTOs**: `UpsertFieldsWidgetInput` /
`UpsertFieldsWidgetGroupInput` / `UpsertFieldsWidgetFieldInput` — caller
passes the full desired state of groups + fields for a widget
- **`FieldsWidgetUpsertService`**: looks up widget → resolves `viewId`,
diffs against existing flat entity maps, builds optimistic group maps so
newly-created groups can be referenced by field updates in the same
pass, then runs creates/updates/deletes in a single
`validateBuildAndRunWorkspaceMigration` call
- **`upsertFieldsWidget` mutation** added to `ViewFieldGroupResolver`;
new `FIELDS_WIDGET_NOT_FOUND` exception code added and wired into the
GraphQL exception filter
## Frontend
- New `UPSERT_FIELDS_WIDGET` GQL document
- `useSaveFieldsWidgetGroups` replaces all per-operation calls with a
single mutation per widget, passing the full draft state — diff logic
moves entirely to the backend
```graphql
mutation UpsertFieldsWidget($input: UpsertFieldsWidgetInput!) {
upsertFieldsWidget(input: $input) {
id
name
position
isVisible
viewId
viewFields { id isVisible position ... }
}
}
```
Widget IDs are collected from both draft **and** persisted state so
widgets whose draft groups were cleared still trigger deletion of their
server-side groups.
<!-- START COPILOT ORIGINAL PROMPT -->
<details>
<summary>Original prompt</summary>
Start implementation
The user has attached the following file paths as relevant context:
- CLAUDE.md
<analysis>
[Chronological Review: The conversation began with the user requesting a
review of how updates on views are stored in the frontend of record page
layouts. The user expressed a desire to migrate heavy computations and
multiple requests from the frontend to the backend. They proposed
creating a custom upsertFieldsWidget mutation to handle updates
atomically. The conversation has focused on this migration strategy and
its implementation.]
[Intent Mapping: The user explicitly requested to "check how we store
updates on views in the frontend of record page layouts" and indicated a
need to "migrate this task to the backend" by creating a custom mutation
for atomic updates.]
[Technical Inventory: The conversation involves the following technical
concepts:
- Frontend and backend architecture for handling page layout updates.
- Custom GraphQL mutation (upsertFieldsWidget) for atomic updates.
- The file in focus is page-layout-update.service.ts, which is likely
responsible for handling updates related to page layouts in the
backend.]
[Code Archaeology: The specific file being worked on is
/Users/devessier/Web/twenty/packages/twenty-server/src/engine/metadata-modules/page-layout/services/page-layout-update.service.ts.
The user is likely looking to modify or enhance this service to
accommodate the new mutation for handling updates.]
[Progress Assessment: The user has identified the need for a backend
migration strategy but has not yet implemented any changes. The focus is
on planning the transition from frontend to backend processing.]
[Context Validation: All critical information for continuation is
captured, including the user's intent to migrate computations to the
backend and the specific file being modified.]
[Recent Commands Analysis: The last agent commands involved checking the
current file context and the user's request to summarize the
conversation. No specific tool results were generated as the focus was
on summarizing the conversation history. The agent was actively working
on summarizing the conversation when the token budget was exceeded,
which was triggered by the user's request for a summary.]
</analysis>
<summary>
1. Conversation Overview:
- Primary Objectives: The user requested to "check how we store updates
on views in the frontend of record page layouts" and expressed a desire
to "migrate this task to the backend" by creating a custom
upsertFieldsWidget mutation for atomic updates.
- Session Context: The conversation has focused on the need to shift
heavy computations from the frontend to the backend, with an emphasis on
implementing a new mutation to streamline updates.
- User Intent Evolution: The user's needs have remained consistent,
focusing on backend migration for efficiency and atomicity in updates.
2. Technical Foundation:
- Core Technology: The conversation involves a backend service likely
built with Node.js and TypeScript, as indicated by the file path.
- Framework/Library: The use of GraphQL for handling mutations is
implied through the mention of a custom mutation.
- Architectural Pattern: The user is looking to implement a
service-oriented architecture where the backend handles complex
computations instead of the frontend.
- Environment Detail: The current working file is located in a server
package of the twenty repository.
3. Codebase Status:
- File Name: page-layout-update.service.ts
- Purpose: This file is responsible for managing updates related to page
layouts in the backend.
- Current State: The user is considering modifications to implement a
new mutation for handling updates.
- Key Code Segments: Specific functions or classes have not been
detailed yet, as the focus is on planning changes.
- Dependencies: This service likely interacts with other components in
the metadata-modules related to page layouts.
4. Problem Resolution:
- Issues Encountered: The current challenge is the inefficiency of
handling updates in the frontend.
- Solutions Implemented: The proposed solution is to create a custom
mutation to handle updates atomically in the backend.
- Debugging Context: No ongoing troubleshooting efforts have been
mentioned yet.
- Lessons Learned: The need for backend processing to improve
performance has been highlighted.
5. Progress Tracking:
- Completed Tasks: No tasks have been completed yet; the user is in the
planning phase.
- Partially Complete Work: The user is preparing to implement a new
mutation for backend updates.
- Validated Outcomes: No features have been confirmed working through
testing at this stage.
6. Active Work State:
- Current Focus: The user is focused on modifying the
page-layout-update.service.ts to implement the new mutation.
- Recent Context: The last few exchanges involved discussing the
migration of update tasks from the frontend to the backend.
- Working Code: No specific code snippets have been modified yet, as the
co...
</details>
<!-- START COPILOT CODING AGENT SUFFIX -->
Created from [VS
Code](https://code.visualstudio.com/docs/copilot/copilot-coding-agent).
<!-- START COPILOT CODING AGENT TIPS -->
---
💡 You can make Copilot smarter by setting up custom instructions,
customizing its development environment and configuring Model Context
Protocol (MCP) servers. Learn more [Copilot coding agent
tips](https://gh.io/copilot-coding-agent-tips) in the docs.
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: Devessier <29370468+Devessier@users.noreply.github.com>
Co-authored-by: Baptiste Devessier <baptiste@devessier.fr>
## Summary
- **Model pricing overhaul**: All model constants updated with accurate
pricing in dollars per 1M tokens, including cached input rates, cache
creation rates, and tiered >200k context pricing
- **New providers**: Added Google (Gemini 3.x), Mistral, and AWS Bedrock
as inference providers. Bedrock serves Claude Opus 4.6 and Sonnet 4.6
via AWS, with proper credential handling following the existing S3/SES
pattern
- **InferenceProvider/ModelFamily split**: Refactored `ModelProvider`
into two orthogonal enums — `InferenceProvider` (who serves the model:
auth, SDK, metadata format) and `ModelFamily` (who created it: token
counting semantics). This eliminates growing `||` chains for token
normalization checks like `excludesCachedTokens`
- **Billing improvements**: Reasoning tokens charged at output rate,
cache token discounts applied accurately, real errors thrown to Sentry
on billing failures
## Test plan
- [x] All existing unit tests updated and passing (23 tests across 3
test files)
- [x] Lint passes for both twenty-server and twenty-front
- [ ] CI checks pass
Made with [Cursor](https://cursor.com)
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
