##
The command pulls the image, compares it against the one the container
was created from, and only recreates the container if the image actually
changed. Your data volumes are preserved — only the container is
replaced.
## Background
The 2026-04-26 incident saw 716M Sonnet 4.6 tokens consumed in a single
trial workspace. Two causes: failed agent executions weren't billed
(addressed by #20065) and the credit-cap gate had been removed from
`WorkflowExecutorWorkspaceService.executeStep` in #19904, leaving no
enforcement point at all.
## Why not just revert #19904#19904 was right that gating at the workflow executor is too coarse.
When one user exhausted a workspace's credits via chat, *all* workflows
hard-failed mid-run — including cheap DB/CRUD/branch automations costing
essentially nothing. Reverting would re-introduce that cliff.
## New design: gate at the AI entry points
The chat resolver already gates this way
(`agent-chat.resolver.ts:137-148`). This PR replicates the same pattern
at every other point where the workspace can incur real AI cost:
- `executeAgent` in `agent-async-executor.service.ts`
- the REST handler in `ai-generate-text.controller.ts`
- `generateThreadTitle` in `agent-title-generation.service.ts`
In each, after auth/validation: skip if `IS_BILLING_ENABLED` is false;
otherwise call `BillingService.canBillMeteredProduct(workspaceId,
BillingProductKey.WORKFLOW_NODE_EXECUTION)`; on `false`, throw
`BillingException(BILLING_CREDITS_EXHAUSTED)`. No new method, no new
exception code, no new product key.
This matches industry convention (Lovable/Replit also gate at the
expensive-operation boundary, not at every cheap step).
## Deliberately not gated
- `WorkflowExecutorWorkspaceService.executeStep` — the design choice is
now intentional, so the #19904 TODO is replaced by a one-line
absolute-behavior comment explaining why the gate isn't here. Cheap
workflow steps (DB CRUD, branching, action steps) are not gated, so a
chat-driven cap exhaustion does not block non-AI automations.
- `repair-tool-call.util` — repair is a sub-call inside an already-gated
AI flow. If the parent is gated, repair will naturally not run. Adding a
gate here adds complexity without value.
## Net effect
A workspace that exhausts credits via chat or AI agent stops making AI
calls. Its non-AI workflows continue running normally. A workflow with
both AI and non-AI steps fails at the AI step with
`BILLING_CREDITS_EXHAUSTED`, but downstream non-AI steps that don't
depend on the AI output still run.
## Conflicts
This PR overlaps with three other in-flight PRs in the same files. None
of them touch the gate logic; rebasing on top of any of them is trivial:
- #20065 (agent-async-executor): adds `workspaceId` to `executeAgent`
args and bills in `finally`. The gate at the top of `executeAgent` from
this PR sits naturally above that.
- #20066 (REST controller): adds usage billing to the controller.
- #20067 (title gen): adds usage billing to title generation and
tool-call repair.
Recommend landing #20065/#20066/#20067 first; this PR rebases trivially
on top.
## Tests
Out of scope per the PR series convention. The existing chat-resolver
gate isn't unit-tested either; this PR follows the same precedent.
Follow-up: add integration coverage that exercises a workspace at
`hasReachedCurrentPeriodCap=true` against each of the three new gates
plus the pre-existing chat-resolver gate.
## Future follow-ups
- Per-user soft cap inside a workspace (the Lovable Business-tier
pattern), so one user can't exhaust the workspace's cap.
- Pre-flight cost estimate so the user sees an "approaching cap" warning
before the hard stop.
- Rename `BillingProductKey.WORKFLOW_NODE_EXECUTION` — the name predates
this design choice and is misleading now that it gates AI entry points
rather than workflow nodes.
## Test plan
- [ ] Trigger a workspace into `hasReachedCurrentPeriodCap=true`.
- [ ] Send a chat message — expect failure with
`BILLING_CREDITS_EXHAUSTED`.
- [ ] Run a workflow whose only AI step is an `ai-agent` action — expect
that step to fail with `BILLING_CREDITS_EXHAUSTED`, downstream non-AI
steps still run.
- [ ] POST to `/rest/ai/generate-text` — expect
`BILLING_CREDITS_EXHAUSTED`.
- [ ] Create a new chat thread (which kicks off `generateThreadTitle`) —
expect `BILLING_CREDITS_EXHAUSTED`.
- [ ] Run a workflow with no AI step (only DB CRUD/branching/actions) —
expect it to run unaffected.
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
EDIT :
- solving auto-caching from Anthropic by updating ai-sdk/anthropic +
adding providerOption at stream level
- concerning Bedrock, it needs breakpoint
**1. Breakpoints were only on the system prompt**
The code already placed a cache marker on the system prompt (~10K
tokens). But the conversation history — which can grow to hundreds of
thousands of tokens — had no marker, so Anthropic re-read it at full
price on every turn.
The fix adds a prepareStep hook inside streamText that stamps the last
message with a cache breakpoint before every LLM call. Anthropic then
caches the entire conversation prefix, and subsequent turns read it at
$0.30/M instead of $3/M.
prepareStep is used rather than a one-shot pre-processing step because
an agentic turn makes multiple internal LLM calls as tool results
accumulate — the hook refreshes the breakpoint before each one.
**2. Bedrock was using the wrong field**
The system prompt marker for Bedrock was set as cacheControl: { type:
'ephemeral' } — which is the Anthropic wire format. The Bedrock Converse
API expects cachePoint: { type: 'default' }. The system prompt was
silently not being cached on Bedrock at all.
Both the system prompt and the new prepareStep now go through a shared
getCacheProviderOptions helper that returns the correct field per
provider.
**3. Persisted cached token usage to monitor cache strat. efficiency**
https://github.com/user-attachments/assets/fe1461c7-0d5c-4c6f-8c2e-2cf569e7de90
## What
Fix `RECORD_SELECTION` items leaking into the command menu when nothing
is selected, and unify how the menu renders in normal vs edit mode.
## The bug
`RECORD_SELECTION`-availability items were showing up even when
`numberOfSelectedRecords === 0`. New util
`doesCommandMenuItemMatchSelectionState` gates them, applied
consistently in the runtime provider and the editor.
## The refactor
`PinnedCommandMenuItemButtonsEditMode` was a 140-line near-duplicate of
`PinnedCommandMenuItemButtons` with its own (drifting) filter logic.
Killed it. Edit mode now flows through the same
`CommandMenuContextProvider` with a new `isInPreviewMode` flag — one
filter chain, one rendering path.
## Behavior in edit mode
**Header (pinned buttons in page header):**
- Runs the full filter chain — object metadata, page type, selection
state, page layout, *and the conditional availability expression*
- Buttons render at full styling but are inert via `pointer-events:
none` + `cursor: not-allowed`
- Preview now reflects exactly what users will see on the live page (not
a grayed-out approximation)
**Side panel editor:**
- New `useEditableCommandMenuItems` hook
- Same filters as runtime *minus* the conditional availability
expression and `FALLBACK` items — so it surfaces everything that's
actually configurable for this page context
- Still gates on selection state — if no records selected,
`RECORD_SELECTION` items are hidden from the editor too. Open to
feedback if we'd rather always show them so users can pin them ahead of
time.
## Misc
- `usePinnedCommandMenuItemsInlineLayout` — visible count now waits
until every item is measured before committing. Fixes a flash of wrong
counts on mount/resize
- Renamed `useCommandMenuContextApi` → `useCurrentCommandMenuContextApi`
— name now conveys it reads from the *current* scoped context store
- Copy: "Records selected" → "Record(s) selected"
Refactors the website visual runtime to make WebGL-heavy sections more
reliable and less expensive.
This adds shared image/model loading caches, safer WebGL context
recovery, staggered visual mounting, and static rendering for decorative
Helped card visuals. It also removes a large bespoke Helped renderer in
favor of the shared halftone model canvas, reduces scroll/layout work in
the Helped section, and cleans up duplicated model-loading code across
several visuals.
Hardened CalDav with new approach of wrapping axios ssrf http agent to
fetch via `@lifeomic/axios-fetch` because `tsdav` only accept `fetch`
override.
Also Hardened test endpoint
## Summary
Adds `h1`–`h6` component overrides to `LazyMarkdownRenderer` so that
`[[record:...]]` references placed inside markdown headings in the AI
chat are parsed by `processChildrenForRecordLinks` and rendered as
clickable `RecordLink` chips, matching the behavior already in place for
`p`, `li`, `td`, `th`, and `a`.
Fixes#20072
## Test plan
- [ ] In AI chat, ask a question whose answer places a record reference
inside a markdown heading (e.g. `## Found [[person:uuid:John Doe]]`) and
confirm a clickable `RecordLink` chip renders instead of the raw
`[[...]]` text.
- [ ] Verify heading styling (sizes, weights, margins) is unchanged.
Generated with [Claude Code](https://claude.ai/code)
---------
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: nitin <ehconitin@users.noreply.github.com>
## Summary
A customer reported that after **Stop Impersonating**, the sidebar still
showed the impersonated user's pinned favorites, the AI chat tab toggle,
and the AI chat history — even though the original admin's session was
correctly restored.
## Root cause
The refactor in #19597 replaced the previous `signOut()`-based stop flow
with an in-place token swap, but only cleared Apollo cache + reloaded
the user. Several user-scoped client stores were left untouched:
- **`metadataStoreState`** is localStorage-backed
(`navigationMenuItems`,
`agentChatThreads`, `views`, `pageLayouts`, etc.) and only refreshed
by `MinimalMetadataLoadEffect`. That effect is gated by
`metadataLoadedVersion` + `desiredLoadState`, neither of which flips
on a same-workspace token swap, so the effect never re-runs.
- **In-memory AI atoms** (`currentAiChatThreadState`,
`agentChatInputState`,
`hasInitializedAgentChatThreadsState`) keep pointing at the
impersonated user's selected thread / input.
- **Session localStorage keys** (`agentChatDraftsByThreadIdState`,
`lastVisitedObjectMetadataItemIdState`,
`lastVisitedViewPerObjectMetadataItemState`,
`playgroundApiKeyState`) carry the impersonated user's drafts and
navigation state.
`clearSession()` (used by logout) avoids this because it calls
`applyMockedMetadata()` and flips `desiredLoadState` mocked↔real, which
chain-triggers a full metadata reload on next sign-in.
## Fix
Extract a `resetUserScopedClientState` helper inside
`useImpersonationSession` that:
1. Calls `clearSessionLocalStorageKeys()` to drop user-scoped
localStorage
keys.
2. Resets the in-memory AI session atoms.
3. Marks `metadataStoreState['agentChatThreads']` as `'empty'`.
`useLoadStaleMetadataEntities` does **not** handle this entity key, so
without an explicit reset to `'empty'` the
`AgentChatThreadInitializationEffect` (which only fires on `'empty'`)
would never refetch.
4. Calls `invalidateMetadataStore()` to clear all
`currentCollectionHash`
values and bump `metadataLoadedVersion`, forcing
`MinimalMetadataLoadEffect` to re-run and refetch
`navigationMenuItems`, `views`, `pageLayouts`, etc. against the new
token.
The helper is applied to both `startImpersonating` and
`stopImpersonating`
— start had the same latent bug; the impersonated user could see the
admin's favorites until the cache happened to refresh.
## Test plan
- [ ] As an admin user, pin some favorites in the sidebar
- [ ] Impersonate a user with different favorites → favorites should
switch to the impersonated user's
- [ ] Click "Stop Impersonating" → sidebar should immediately show the
admin's favorites (not the impersonated user's)
- [ ] As an admin **without** AI permission, impersonate a user **with**
AI permission, open AI chat, send a message, then stop impersonating
→ AI chat history should be empty / inaccessible (the AI tab
visibility itself is fixed in a separate PR)
- [ ] Type a draft in AI chat as the impersonated user → after stop, the
draft should be gone
- [ ] Verify regular sign-out still works while impersonating
- [ ] Verify the impersonation banner still shows / hides correctly
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Automated daily sync of `ai-providers.json` from
[models.dev](https://models.dev).
This PR updates pricing, context windows, and model availability based
on the latest data.
New models meeting inclusion criteria (tool calling, pricing data,
context limits) are added automatically.
Deprecated models are detected based on cost-efficiency within the same
model family.
**Please review before merging** — verify no critical models were
incorrectly deprecated.
Co-authored-by: FelixMalfait <6399865+FelixMalfait@users.noreply.github.com>
# Introduction
Prevent using both `--start-from-workspace-id` and `--workspace`
When any of the two are being passed we prevent passing to the next
instance segment, it would require an upgrade re run even if legit
When `--start-from-workspace-id` is passed we filter from all the
fetched active or suspended workspace ids and apply equivalent filter as
before
This centralizes routing/SEO ownership, replaces deprecated middleware
with proxy, adds safer lifecycle/runtime primitives, and introduces
visual error boundaries so broken WebGL/canvas-heavy visuals can fail
gracefully instead of taking down the page. It also hardens animation,
resize, visibility, cleanup, and WebGL fallback paths for a broader
range of browsers and devices.
The hero visual was split from large monolithic files into focused
domain folders for shell, pages, shared primitives, window interactions,
terminal conversation, prompt, editor, and traffic-light behavior.
Legacy unused section code was removed, visual configs were extracted,
state/geometry logic was moved into testable modules, and coverage was
added across routing, SEO, lifecycle, animation, visual runtime,
halftone behavior, and hero interactions.
More changes on the way, but this should make the website a lot more
stable - disabling WebGL on Firefox and loading the website does not
cause crashes on local any longer, will test on dev once this is merged.
# Summary
(fixes#20044)
This PR implements two fixes for the REST API to enforce stricter
validation and provide better error messages.
Issue 1: Cursor parameter silently ignored
Problem: When users provided common cursor aliases (e.g., cursor, after,
before) instead of the correct parameter names (starting_after,
ending_before), the API silently ignored them and returned page 1 on
every request.
Solution: Added strict validation to detect common cursor aliases and
throw a clear error directing users to use the correct parameter names.
Files modified:
-
packages/twenty-server/src/engine/api/rest/input-request-parsers/rest-input-request-parser.exception.ts
-
packages/twenty-server/src/engine/api/rest/input-request-parsers/starting-after-parser-utils/parse-starting-after-rest-request.util.ts
-
packages/twenty-server/src/engine/api/rest/input-request-parsers/ending-before-parser-utils/parse-ending-before-rest-request.util.ts
- Test files for both parsers
Example error:
Invalid cursor parameter 'cursor'. Use 'starting_after' for pagination.
---
Issue 2: OpportunityStageEnum not validated on REST
Problem: When creating or updating opportunities via REST with an
invalid stage value, the API either silently dropped the value or
returned a generic error without listing valid options.
Solution: Updated the SELECT field validation to include valid options
in the error message.
Files modified:
-
packages/twenty-server/src/engine/api/common/common-args-processors/data-arg-processor/validator-utils/validate-rating-and-select-field-or-throw.util.ts
- Test file
Example error:
Invalid value "BAD_VALUE" for field "stage". Valid values are: NEW,
SCREENING, MEETING, PROPOSAL, CUSTOMER
---
### Testing
- Added 5 new test cases for `parse-starting-after-rest-request.util.ts`
- Added 6 new test cases for `parse-ending-before-rest-request.util.ts`
- Added 1 new test case for
`validate-rating-and-select-field-or-throw.util.ts`
- All 21 tests passing
---
Breaking Changes
None - correct usage is unaffected.
---------
Co-authored-by: Etienne <45695613+etiennejouan@users.noreply.github.com>
- Update usageEvent clickhouse table, partitioning, indexing and
projection (auto materialized view) to optimize credit usage queries
- Add caching for available credits and billing subscription
To do in next PR: deprecate enforceCapUsage cron. Bonus : real-time on
billingSubscription
## Summary
Iterative redesign of two related areas in settings, plus a new
`pages/settings/layout/` folder for read-only entity detail pages.
### Application content tab
- **Grouped into three sections** — Data / Layout / Logic — each with
one H2 + multiple `TableSection`-wrapped sub-tables (mirrors the
role-permissions pattern). Replaces six per-category table/row
components with one uniform `<SettingsApplicationContentSubtable>` +
`ApplicationContentRow` shape (net **−~700 lines** across the refactor).
- **All 10 row categories now clickable** for installed apps:
- Objects / Fields / Logic functions / Front components → existing
detail pages
- Agents → existing `AiAgentDetail`
- Skills → existing `AiSkillDetail` (looked up by `Skill.applicationId +
name`)
- Roles → existing `RoleDetail` (looked up by
`Role.universalIdentifier`)
- Views / Page layouts / Navigation menu items → **new** detail pages
(see below)
- **Lifecycle hooks visible** — `pre-install` / `post-install` logic
functions are surfaced in the Trigger column instead of appearing as
empty/misconfigured.
### Logic function settings (Triggers + Test tabs)
- Triggers tab is now editable (HTTP / Cron / Database event / AI tool)
with a `<SettingsLogicFunctionTriggerSection>` wrapper that owns the
toggle, header, and read-only short-circuit.
- HTTP section gets a Live URL field with copy-to-clipboard.
- Each section shows a **Sample input** preview (the JSON the function
will receive) using the same payload builders the Test tab uses.
- Test tab: **Simulate trigger** buttons that prefill the JSON input
from the configured trigger's schema. Replaces an unclickable `<Select>`
(which auto-disables when there's only one option — the typical case).
- Read-only behavior for installed-app functions: explicit `<Callout>`
notice when there's no trigger; trigger sections render as disabled
controls when there is one.
- Removed the empty Environment Variables section from the Settings tab
(it just told the user to go elsewhere).
### New `pages/settings/layout/` folder
Three new app-scoped detail pages so users can drill into entities the
GraphQL `Application` type doesn't expose by id (keyed by manifest
`universalIdentifier`):
- `ApplicationViewDetail` — type, object, visibility + Fields / Filters
/ Sorts subsections (field UIDs resolved to readable labels via
`useFieldLabelByUid`)
- `ApplicationPageLayoutDetail` — type, object + per-tab subsections
listing widgets
- `ApplicationNavigationMenuItemDetail` — type, destination (resolved),
icon, color, position
Each page reads from the marketplace manifest the parent app page
already loads (no extra queries). Folder set up so a future "Layout"
settings tab can grow here (analogous to the existing `data-model/`
folder under the Data tab).
### Other consistency fixes
- Breadcrumbs on every app-scoped entity detail page now include a
category crumb so users know what they're looking at: `Workspace /
Applications / Timely / Navigation menu items / Time entry`.
- Title fallback for nav menu items uses the resolved destination
(`"Time entry"`) instead of the raw enum (`"OBJECT"`).
- New shared utils: `getNavigationMenuItemDestination`,
`resolveManifestObjectLabel`, `getLogicFunctionTriggerLabel`,
`<MonoText>`.
## Backend changes
Only one minor schema-shape change (additive): added `applicationId` to
the `SkillFields` GraphQL fragment and `universalIdentifier` to the
`RoleFragment` so the new lookups have what they need. Generated
metadata schema patched in-tree to match — regenerate with `nx run
twenty-front:graphql:generate --configuration=metadata` if it drifts.
## Test plan
- [ ] Application content tab on an installed app shows the 3 grouped
sections; rows in each section are clickable
- [ ] Click an Object → existing object detail page
- [ ] Click a Field → existing field-edit page
- [ ] Click an Agent / Skill / Role → existing detail page
- [ ] Click a View / Page layout / Navigation menu item → new read-only
detail page; subsections (Fields/Filters/Sorts for views, per-tab
widgets for page layouts) populate correctly
- [ ] Breadcrumbs on every entity detail page have 5 crumbs ending in
`<Category> / <Entity name>`
- [ ] Logic function Triggers tab: toggle each trigger type on/off, see
the Sample input preview update; for installed apps, sections render as
read-only
- [ ] Test tab: each "Simulate trigger" button prefills the JSON editor
with the matching payload shape
- [ ] Functions list: a function configured as `post-install` shows
"Post-install" in the Trigger column
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: claude[bot] <claude[bot]@users.noreply.github.com>
## Summary
- Rename safeParseRelativeDateFilterJSONStringified to
safeParseRelativeDateFilterJsonStringified
- Update the matching utility file, exports, tests, and workflow usages
Part of #19839.
## Validation
- CI passed
## Summary
`AgentAsyncExecutorService.executeAgent` consumes Anthropic tokens at
two points (the main `generateText` and the optional structured-output
sub-call). Billing was previously the **caller's** responsibility,
executed only after `executeAgent` returned. If `executeAgent` threw —
e.g. when `structuredResult.output == null` for a schema-mismatched
response, or anything caught by the catch-and-rethrow — we paid
Anthropic but never recorded a `usageEvent`. Likely the dominant source
of the 716M-token-vs-3.27-credits discrepancy seen on the affected
workspace in the 2026-04-26 incident.
## What changed
- Inject `AiBillingService` into `AgentAsyncExecutorService`. Add
`workspaceId` (required), `userWorkspaceId`, and `operationType`
(default `AI_WORKFLOW_TOKEN`) to `executeAgent`'s args.
- Capture `accumulatedUsage`, `cacheCreationTokens`, and
`nativeWebSearchCallCount` into mutable locals as each `generateText`
resolves. A throw between the main and structured-output calls still
bills the first call's tokens; the schema-validation throw still bills
the merged usage.
- Wrap the body in `try { ... } finally { ... }`. The finally calls
`calculateAndBillUsage` and `billNativeWebSearchUsage`, each guarded by
its own `try/catch + logger.error` so a billing exception can't mask the
original execution error or block the second emit.
- `ai-agent.workflow-action.ts`: pass the new args; drop the
now-redundant billing calls and `AiBillingService` injection.
`AiBillingModule` removed from this action's module imports.
- `run-evaluation-input.job.ts`: pass `workspaceId` (already in `data`)
and `userWorkspaceId: null`. **As a side effect, the eval pipeline now
bills correctly** — closing an additional billing leak from the audit
(`RunEvaluationInputJob` previously called `executeAgent` and discarded
`executionResult.usage`).
## Behavior change worth calling out
Previously, failed agent executions were silently free. They will now be
billed for the tokens Anthropic charged us. This is intentional and
correct.
## Test plan
- [ ] Trigger a workflow agent action that succeeds — `usageEvent` count
should match what was previously emitted.
- [ ] Trigger a workflow agent with a JSON response schema and ambiguous
input that produces a non-schema-conforming output
(`structuredResult.output == null`) — verify a `usageEvent` row is now
written for the consumed tokens (was 0 rows previously).
- [ ] Trigger a `runEvaluationInput` GraphQL mutation — verify a
`usageEvent` row is written (was 0 rows previously).
## Notes for review
- Conflicts trivially with the Sentry-context PR on
`run-evaluation-input.job.ts`. Recommend merging Sentry-context first;
this PR's 2-line argument addition rebases inside that PR's
`aiCallContextService.run(...)` callback wrapper.
- A small follow-up after this lands: thread `billingContext` through
the `experimental_repairToolCall` callback in
`agent-async-executor.service.ts:198` (currently marked with a TODO from
the title-gen+repair-tool PR).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Automated daily sync of `ai-providers.json` from
[models.dev](https://models.dev).
This PR updates pricing, context windows, and model availability based
on the latest data.
New models meeting inclusion criteria (tool calling, pricing data,
context limits) are added automatically.
Deprecated models are detected based on cost-efficiency within the same
model family.
**Please review before merging** — verify no critical models were
incorrectly deprecated.
Co-authored-by: FelixMalfait <6399865+FelixMalfait@users.noreply.github.com>
## Summary
`POST /rest/ai/generate-text` calls `generateText` and returns `usage`
to the client without emitting a `usageEvent`. Authenticated, gated only
by `PermissionFlagType.AI` — any workspace user with that permission
could call it in a loop without billing. Identified during the
2026-04-26 incident audit.
## What changed
- Inject `AiBillingService` into `AiGenerateTextController`.
- Add `@AuthUserWorkspaceId() userWorkspaceId: string` to source the
user-workspace identifier.
- Wrap the `generateText` call in `try { ... return ... } finally { ...
}` so billing fires even if the controller throws after Anthropic was
paid.
- Bill with `UsageOperationType.AI_WORKFLOW_TOKEN` and
`cacheCreationTokens: result.usage.inputTokenDetails?.cacheWriteTokens
?? 0`.
- Inner `try/catch` around the billing emit so a billing error can't
break the response.
- One-line module change: `AiGenerateTextModule` imports
`AiBillingModule` (NestJS DI requirement).
## Test plan
- [ ] Call `POST /rest/ai/generate-text` with a small prompt; verify a
`usageEvent` row appears in ClickHouse for the workspace with the
correct token count and `operationType = AI_WORKFLOW_TOKEN`.
- [ ] Call with a malformed model id that throws after the API key is
validated — verify no spurious billing call occurs (no Anthropic call
was made).
## Notes for review
- Response shape unchanged.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Weiko