feat(hosted key): Add exa hosted key by TheodoreSpeaks · Pull Request #3221 · simstudioai/sim

TheodoreSpeaks · 2026-02-14T07:03:44Z

Summary

Added hosted api key integration for exa search.
For hosted sim, api sub block no longer appears for exa search. Users can use the BYOK modal to pass an api key if needed. Non-hosted sim behaves unchanged.

Added throttling per billing actor. Requests are throttled proactively, failing fast if they exceed limits (users can switch to BYOK if they need higher limits). Can be either a limit based on requests per minute or a custom limit using the api response. If custom, rate limiting is optimistic and only applies after execution completes.

Moved cost modification behavior to knowledge block directly instead of transforming each block separately.

Hosted key calls retry 3 times with exponential backoff to handle throttles from increased load, with alarms.

Important: API keys need to be stored in prod and staging before merging:

EXA_API_KEY_COUNT number
EXA_API_KEY_{number} api keys (traffic spread evenly via round robin)

Type of Change

Testing

Tested non-host still has required api subblock.
Tested hosted sim does not have api subblock
Tested hosted key updates user stats and usage log tables correctly
Unit tested throttling handles correctly
Tested rate limiting works successfully for exa search with hosted key and does not apply when BYOK is applied.

Checklist

Code follows project style guidelines
Self-reviewed my changes
Tests added/updated and passing
No new warnings introduced
I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

Screenshots/Videos

vercel · 2026-02-14T07:03:49Z

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Mar 7, 2026 3:56am

greptile-apps · 2026-02-14T07:09:30Z

Greptile Summary

This PR adds hosted Exa API key support to the Sim platform, allowing users on hosted Sim to run Exa search/content/answer operations without providing their own API key. It introduces a HostedKeyRateLimiter with per-billing-actor token-bucket limiting and round-robin key selection across a pool of hosted keys, a two-phase pricing system (per_request or custom based on Exa's cost response), and exponential backoff retry for upstream rate limits. Knowledge block cost restructuring is also moved from the generic handler to each tool's transformResponse, which is a cleaner approach.

Key changes:

tools/index.ts: Adds injectHostedKeyIfNeeded, executeWithRetry, and applyHostedKeyCostToResult — the core hosted key pipeline
lib/core/rate-limiter/hosted-key/: New HostedKeyRateLimiter class with per-actor token buckets backed by the existing RateLimitStorageAdapter
blocks/blocks/exa.ts & lib/workflows/subblocks/visibility.ts: Declarative hideWhenHosted flag hides the apiKey subblock in the UI and serializer on hosted deployments
tools/exa/*.ts: Each Exa tool gains a hosting config with pricing and rate limit settings; __costDollars internal field carries Exa's cost data through the pipeline before being stripped
tools/knowledge/search.ts & upload_chunk.ts: Cost restructuring moved from the generic handler into transformResponse
Notable open issue: The costDollars? fields declared in the four Exa response interfaces (ExaSearchResponse, etc.) are never populated — the data lives under the untyped __costDollars key — creating a false type contract
Notable risk: HostedKeyRateLimiter fails open on DB storage errors, meaning all per-actor rate limits and custom dimension budgets are silently bypassed if the storage layer is unavailable

Confidence Score: 3/5

This PR is functional but has a few unresolved concerns that warrant caution before merging to production.
The core hosted-key pipeline (injection, retry, billing) is well-structured and unit tested. However, three issues reduce confidence: (1) The HostedKeyRateLimiter fails open on storage errors, completely bypassing rate limiting during any DB outage — a risk in a billing-sensitive path. (2) stripInternalFields is applied unconditionally to all tool outputs rather than being scoped to hosted-key executions, which is a latent regression risk for other tools. (3) The in-memory roundRobinCounters won't distribute load across serverless instances, so the key-pool distribution guarantee is only per-instance. These are known trade-offs (some discussed in prior threads), but the fail-open behaviour in particular should be acknowledged before going to production.
Pay close attention to apps/sim/lib/core/rate-limiter/hosted-key/hosted-key-rate-limiter.ts (fail-open on storage errors) and apps/sim/tools/index.ts (unconditional stripInternalFields).

Important Files Changed

Filename	Overview
apps/sim/tools/index.ts	Core change: adds `injectHostedKeyIfNeeded`, `executeWithRetry`, `processHostedKeyCost`, and `stripInternalFields` for hosted Exa key support. The unconditional application of `stripInternalFields` to all tool outputs (not just hosted-key executions) remains a concern previously flagged. Cost/usage tracking is cleanly deduplicated into `applyHostedKeyCostToResult`, which is an improvement over an earlier duplicate-code design.
apps/sim/lib/core/rate-limiter/hosted-key/hosted-key-rate-limiter.ts	New rate limiter with per-billing-actor token-bucket limiting and round-robin key selection. The in-memory `roundRobinCounters` singleton won't distribute load evenly across serverless instances. Both `checkActorRateLimit` and `preCheckDimensions` fail-open on storage errors, fully bypassing rate limits if the DB is unavailable.
apps/sim/lib/core/rate-limiter/hosted-key/types.ts	Well-structured type definitions for rate limit config and results. The `requestsPerMinute` JSDoc says "maximum" but the actual token-bucket allows a 2× burst due to `DEFAULT_BURST_MULTIPLIER = 2`, which is misleading to callers configuring tools.
apps/sim/tools/exa/types.ts	Adds `ExaCostDollars` and `costDollars?` fields to four response interfaces, but the `transformResponse` implementations store cost under `__costDollars`. The public type declares a field that is always `undefined` at runtime, creating a misleading contract for consumers.
apps/sim/blocks/blocks/exa.ts	Splits the single `apiKey` subblock into two conditionally exclusive subblocks: one hidden when hosted (for all operations except `exa_research`) and one always visible (for `exa_research`). Follows the existing canonical-pair pattern correctly; both share the same `id` so they resolve to one underlying value.
apps/sim/lib/core/rate-limiter/hosted-key/hosted-key-rate-limiter.test.ts	Comprehensive test coverage for `acquireKey` (per-request and custom modes) and `reportUsage`, including error paths like storage failures, missing keys, and depleted dimensions. Tests correctly verify round-robin cycling and optimistic concurrency handling.
apps/sim/tools/index.test.ts	Adds substantial new test suites for hosted key injection, rate limiting/retry logic, and cost field handling. Tests properly mock `isHosted`, `getBYOKKey`, `logFixedUsage`, and the rate limiter. Some tests include soft assertions (`if (result.output.cost)`) that reduce coverage confidence on the billing path.
apps/sim/lib/core/rate-limiter/storage/db-token-bucket.ts	Fixes the token bucket SQL: the old `ELSE ${rateLimitBucket.tokens}` kept the bucket at its pre-consumption value when overdrawn (causing `allowed = true` incorrectly). New `ELSE -1` correctly marks the bucket as overdrawn. This is a correctness fix; the intentional precision loss for overdraft tracking was accepted by the team.

Sequence Diagram

sequenceDiagram
    participant UI as Browser UI
    participant Serializer
    participant Executor as executeTool()
    participant BYOK as getBYOKKey()
    participant RateLimiter as HostedKeyRateLimiter
    participant DB as Token Bucket DB
    participant Exa as Exa API

    UI->>Serializer: Serialize workflow (key field hidden when hosted)
    Note over Serializer: isSubBlockHiddenByHostedKey skips field
    Serializer-->>Executor: params without key

    Executor->>Executor: injectHostedKeyIfNeeded()
    Executor->>BYOK: getBYOKKey(workspaceId, exa)
    alt BYOK key found
        BYOK-->>Executor: key result (no billing)
    else No BYOK key
        BYOK-->>Executor: null
        Executor->>RateLimiter: acquireKey(provider, prefix, config, actorId)
        RateLimiter->>DB: consumeTokens(actor bucket, 1)
        alt Actor rate limited
            DB-->>RateLimiter: allowed false
            RateLimiter-->>Executor: billingActorRateLimited true
            Executor-->>UI: Error 429
        else Within limit
            DB-->>RateLimiter: allowed true
            RateLimiter-->>Executor: success with hosted key
        end
    end

    loop executeWithRetry up to 3 retries on 429
        Executor->>Exa: executeToolRequest(params)
        alt 429 from Exa
            Exa-->>Executor: HTTP 429
            Note over Executor: Wait 1s then 2s then 4s
        else Success
            Exa-->>Executor: Response with cost data
        end
    end

    Executor->>Executor: applyHostedKeyCostToResult()
    Executor->>DB: logFixedUsage(userId, cost)
    Executor->>Executor: stripInternalFields removes internal fields
    Executor-->>UI: output with cost total and timing

Comments Outside Diff (3)

apps/sim/lib/core/rate-limiter/hosted-key/hosted-key-rate-limiter.ts, line 1118-1124 (link)

Fail-open on storage errors bypasses rate limiting entirely

Both checkActorRateLimit and preCheckDimensions catch storage errors and return null / silently skip the failing dimension, which causes the rate limiting to be fully bypassed when the database is unavailable:
```
// checkActorRateLimit
} catch (error) {
  logger.error(`Error checking billing actor rate limit for ${provider}`, { error, billingActorId })
  return null  // ← request is allowed through
}

// preCheckDimensions – same pattern
} catch (error) {
  logger.error(`Error pre-checking dimension ${dimension.name} for ${provider}`, { error, billingActorId })
  // ← silently skips this dimension, request proceeds
}
```
If the DB becomes unavailable (even briefly), every workspace's actor-level rate limit and every custom dimension budget are silently bypassed. All requests are served using hosted Exa keys until the storage recovers, with no enforcement. This could cause a large uncontrolled spike in upstream Exa API usage during a storage outage.

Consider failing closed (blocking requests) when storage fails, or at minimum adding a metric/alarm that fires on repeated storage errors so the ops team is notified promptly. If fail-open is the intentional policy, a comment explaining the decision would help future maintainers understand the tradeoff.
apps/sim/lib/core/rate-limiter/hosted-key/types.ts, line 1383-1384 (link)

requestsPerMinute JSDoc misleads — actual effective limit is 2× due to burst

The JSDoc says "Maximum requests per minute per billing actor (enforced - blocks if exceeded)", but toTokenBucketConfig sets maxTokens = requestsPerMinute * burstMultiplier (default 2). Token buckets start full, so a workspace can immediately issue 2× the stated limit as a burst before being throttled.

For example, exa_answer with requestsPerMinute: 5 will allow 10 rapid requests before blocking, then refill at 5/minute.

This is a standard token-bucket tradeoff (allowing short bursts while averaging to the target rate), but it is surprising to anyone reading the config value as a hard cap. Consider updating the JSDoc to clarify burst semantics:
```
/**
 * Target sustained rate per minute per billing actor.
 * Due to burst buffering (maxTokens = requestsPerMinute * burstMultiplier),
 * the short-term effective limit may be up to `requestsPerMinute * burstMultiplier`.
 * @see DEFAULT_BURST_MULTIPLIER
 */
requestsPerMinute: number
```
apps/sim/tools/exa/types.ts, line 1853-1855 (link)

costDollars in output types will always be undefined at runtime

The four response interfaces (ExaSearchResponse, ExaGetContentsResponse, ExaFindSimilarLinksResponse, ExaAnswerResponse) each add costDollars?: ExaCostDollars to their output type. However, all four transformResponse implementations store the Exa cost data under the key __costDollars (double-underscore prefix):
```
// search.ts
output: {
  results: ...,
  __costDollars: data.costDollars,  // stored as __costDollars
}
```
TypeScript consumers that read response.output.costDollars (relying on the declared type) will always get undefined. The real data lives at the untyped __costDollars key.

Since __costDollars is intentionally an internal field (stripped before returning to users), it should not appear in the public output type at all. These costDollars? fields should either be removed from the four response interfaces, or the type should declare __costDollars as an internal field if it needs to be accessible within the pipeline.

_{Last reviewed commit: a0fc749}

greptile-apps

_{22 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

apps/sim/blocks/blocks/exa.ts

apps/sim/providers/utils.ts

icecrasher321 · 2026-03-07T03:02:29Z

apps/sim/providers/anthropic/core.ts


    const toolCalls = []
-    const toolResults = []
+    const toolResults: any[] = []


we're trying to avoid any annoations in the codebase -- can the function signature be updated with right shared type? if not use unknown

icecrasher321 · 2026-03-07T03:03:45Z

apps/sim/tools/exa/find_similar_links.ts

+        if (costDollars?.total != null) {
+          return { cost: costDollars.total, metadata: { costDollars } }
+        }
+        // Fallback: $5/1000 (1-25 results) or $25/1000 (26-100 results)


we tend to avoid fallbacks like this -- is there a reason they might not give us costDollars? We want to fail loudly and fix the root cause if things don't work since hard to know when it flips to the fallback

icecrasher321 · 2026-03-07T03:04:03Z

apps/sim/tools/exa/find_similar_links.ts

          highlights: result.highlights,
          score: result.score || 0,
        })),
+        __costDollars: data.costDollars,


underscore to hide it from display as an internal field?

nvm mentioned in the comment below, thanks

Correct - basically wanted to account for being able to pass costDollars internally without needing it displayed to the customer, which is useful if we want to add a hosted key fee without the user getting confused.

icecrasher321 · 2026-03-07T03:04:32Z

apps/sim/tools/exa/search.ts

+          return { cost: costDollars.total, metadata: { costDollars } }
+        }
+
+        // Fallback: estimate based on search type and result count


apps/sim/providers/index.ts

cursor

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

cursor · 2026-03-07T03:54:25Z

apps/sim/tools/index.ts

+      },
+    }
+  }
+}


Unhandled pricing error crashes successful tool execution

High Severity

applyHostedKeyCostToResult and processHostedKeyCost have no error handling around calculateToolCost. If getCost throws (e.g., Exa API response missing __costDollars), the error propagates unhandled up to executeTool's outer catch block, turning a successful tool execution into a failed result. Since the API call already succeeded and the user already consumed the Exa resource, this silently discards the valid result over a billing calculation failure.

Additional Locations (1)

apps/sim/tools/index.ts#L257-L269

cursor

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

apps/sim/tools/index.ts

This reverts commit 158d523.

This reverts commit 158d523. Co-authored-by: Theodore Li <[email protected]>

This reverts commit 635179d.

* feat(hosted keys): Implement serper hosted key * Handle required fields correctly for hosted keys * Add rate limiting (3 tries, exponential backoff) * Add custom pricing, switch to exa as first hosted key * Add telemetry * Consolidate byok type definitions * Add warning comment if default calculation is used * Record usage to user stats table * Fix unit tests, use cost property * Include more metadata in cost output * Fix disabled tests * Fix spacing * Fix lint * Move knowledge cost restructuring away from generic block handler * Migrate knowledge unit tests * Lint * Fix broken tests * Add user based hosted key throttling * Refactor hosted key handling. Add optimistic handling of throttling for custom throttle rules. * Remove research as hosted key. Recommend BYOK if throtttling occurs * Make adding api keys adjustable via env vars * Remove vestigial fields from research * Make billing actor id required for throttling * Switch to round robin for api key distribution * Add helper method for adding hosted key cost * Strip leading double underscores to avoid breaking change * Lint fix * Remove falsy check in favor for explicit null check * Add more detailed metrics for different throttling types * Fix _costDollars field * Handle hosted agent tool calls * Fail loudly if cost field isn't found * Remove any type * Fix type error * Fix lint * Fix usage log double logging data * Fix test * Add browseruse hosted key * Add firecrawl and serper hosted keys * feat(hosted key): Add exa hosted key (#3221) * feat(hosted keys): Implement serper hosted key * Handle required fields correctly for hosted keys * Add rate limiting (3 tries, exponential backoff) * Add custom pricing, switch to exa as first hosted key * Add telemetry * Consolidate byok type definitions * Add warning comment if default calculation is used * Record usage to user stats table * Fix unit tests, use cost property * Include more metadata in cost output * Fix disabled tests * Fix spacing * Fix lint * Move knowledge cost restructuring away from generic block handler * Migrate knowledge unit tests * Lint * Fix broken tests * Add user based hosted key throttling * Refactor hosted key handling. Add optimistic handling of throttling for custom throttle rules. * Remove research as hosted key. Recommend BYOK if throtttling occurs * Make adding api keys adjustable via env vars * Remove vestigial fields from research * Make billing actor id required for throttling * Switch to round robin for api key distribution * Add helper method for adding hosted key cost * Strip leading double underscores to avoid breaking change * Lint fix * Remove falsy check in favor for explicit null check * Add more detailed metrics for different throttling types * Fix _costDollars field * Handle hosted agent tool calls * Fail loudly if cost field isn't found * Remove any type * Fix type error * Fix lint * Fix usage log double logging data * Fix test --------- Co-authored-by: Theodore Li <[email protected]> * Fail fast on cost data not being found * Add hosted key for google services * Add hosting configuration and pricing logic for ElevenLabs TTS tools * Add linkup hosted key * Add jina hosted key * Add hugging face hosted key * Add perplexity hosting * Add broader metrics for throttling * Add skill for adding hosted key * Lint, remove vestigial hosted keys not implemented * Revert agent changes * fail fast * Fix build issue * Fix build issues * Fix type error * Remove byok types that aren't implemented * Address feedback * Use default model when model id isn't provided * Fix cost default issues * Remove firecrawl error suppression * Restore original behavior for hugging face * Add mistral hosted key * Remove hugging face hosted key * Fix pricing mismatch is mistral and perplexity * Add hosted keys for parallel and brand fetch * Add brandfetch hosted key * Update types * Change byok name to parallel_ai * Add telemetry on unknown models --------- Co-authored-by: Theodore Li <[email protected]>

* Fix lint * improvement(sidebar): loading * fix(sidebar): use client-generated UUIDs for stable optimistic updates (#3439) * fix(sidebar): use client-generated UUIDs for stable optimistic updates * fix(folders): use zod schema validation for folder create API Replace inline UUID regex with zod schema validation for consistency with other API routes. Update test expectations accordingly. * fix(sidebar): add client UUID to single workflow duplicate hook The useDuplicateWorkflow hook was missing newId: crypto.randomUUID(), causing the same temp-ID-swap issue for single workflow duplication from the context menu. * fix(folders): avoid unnecessary Set re-creation in replaceOptimisticEntry Only create new expandedFolders/selectedFolders Sets when tempId differs from data.id. In the common happy path (client-generated UUIDs), this avoids unnecessary Zustand state reference changes and re-renders. * Mothership block logs * Fix mothership block logs * improvement(knowledge): make connector-synced document chunks readonly (#3440) * improvement(knowledge): make connector-synced document chunks readonly * fix(knowledge): enforce connector chunk readonly on server side * fix(knowledge): disable toggle and delete actions for connector-synced chunks * Job exeuction logs * Job logs * fix(connectors): remove unverifiable requiredScopes for Linear connector * fix(connectors): remove legacy requiredScopes from Jira and Confluence connectors Jira and Confluence OAuth tokens don't return legacy scope names like read:jira-work or read:confluence-content.all, causing the 'Update access' banner to always appear. Set requiredScopes to empty array like Linear. * feat(tasks): add rename to task context menu (#3442) * Revert "fix(connectors): remove legacy requiredScopes from Jira and Confluence connectors" This reverts commit a0be3ff414bebe69e85eabe46334b0b66a2d24b9. * fix(connectors): restore Linear connector requiredScopes Linear OAuth does return scopes in the token response. The previous fix of emptying requiredScopes was based on an incorrect assumption. Restoring requiredScopes: ['read'] as it should work correctly. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(knowledge): pass workspaceId to useOAuthCredentials in connector card The ConnectorCard was calling useOAuthCredentials(providerId) without a workspaceId, causing the credentials API to return an empty array. This meant the credential lookup always failed, getMissingRequiredScopes received undefined, and the "Update access" banner always appeared. Co-Authored-By: Claude Opus 4.6 <[email protected]> * Fix oauth link callback from mothership task * feat(connectors): add Fireflies connector and API key auth support (#3448) * feat(connectors): add Fireflies connector and API key auth support Extend the connector system to support both OAuth and API key authentication via a discriminated union (`ConnectorAuthConfig`). Add Fireflies as the first API key connector, syncing meeting transcripts via the Fireflies GraphQL API. Schema changes: - Make `credentialId` nullable (null for API key connectors) - Add `encryptedApiKey` column (AES-256-GCM encrypted, null for OAuth) This eliminates the `'_apikey_'` sentinel and inline `sourceConfig._encryptedApiKey` patterns, giving each auth mode its own clean column. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(fireflies): allow 0 for maxTranscripts (means unlimited) Co-Authored-By: Claude Opus 4.6 <[email protected]> --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * Add context * fix(fireflies): correct types from live API validation (#3450) * fix(fireflies): correct types from live API validation - speakers.id is number, not string (API returns 0, 1, 2...) - summary.action_items is a single string, not string[] - Update formatTranscriptContent to handle action_items as string Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(fireflies): correct tool types from live API validation - FirefliesSpeaker.id: string -> number - FirefliesSentence.speaker_id: string -> number - FirefliesSpeakerAnalytics.speaker_id: string -> number - FirefliesSummary.action_items: string[] -> string - FirefliesSummary.outline: string[] -> string - FirefliesSummary.shorthand_bullet: string[] -> string - FirefliesSummary.bullet_gist: string[] -> string - FirefliesSummary.topics_discussed: string[] -> string Co-Authored-By: Claude Opus 4.6 <[email protected]> --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * feat(knowledge): add connector tools and expand document metadata (#3452) * feat(knowledge): add connector tools and expand document metadata * fix(knowledge): address PR review feedback on new tools * fix(knowledge): remove unused params from get_document transform * refactor, improvement * fix: correct knowledge block canonical pair pattern and subblock migration - Rename manualDocumentId to documentId (advanced subblock ID should match canonicalParamId, consistent with airtable/gmail patterns) - Fix documentSelector.dependsOn to reference knowledgeBaseSelector (basic depends on basic, not advanced) - Remove unnecessary documentId migration (ID unchanged from main) Co-Authored-By: Claude Opus 4.6 <[email protected]> * lint * fix: resolve post-merge test and lint failures - airtable: sync tableSelector condition with tableId (add getSchema) - backfillCanonicalModes test: add documentId mode to prevent false backfill - schedule PUT test: use invalid action string now that disable is valid - schedule execute tests: add ne mock, sourceType field, use mockReturnValueOnce for two db.update calls - knowledge tools: fix biome formatting (single-line arrow functions) Co-Authored-By: Claude Opus 4.6 <[email protected]> * Fixes * Fixes * Clean vfs * Fix * Fix lint * fix(connectors): add rate limiting, concurrency controls, and bug fixes (#3457) * fix(connectors): add rate limiting, concurrency controls, and bug fixes across knowledge connectors - Add Retry-After header support to fetchWithRetry for all 18 connectors - Batch concurrent API calls (concurrency 5) in Dropbox, Google Docs, Google Drive, OneDrive, SharePoint - Batch concurrent API calls (concurrency 3) in Notion to match 3 req/s limit - Cache GitHub tree in syncContext to avoid re-fetching on every pagination page - Batch GitHub blob fetches with concurrency 5 - Fix GitHub base64 decoding: atob() → Buffer.from() for UTF-8 safety - Fix HubSpot OAuth scope: 'tickets' → 'crm.objects.tickets.read' (v3 API) - Fix HubSpot syncContext key: totalFetched → totalDocsFetched for consistency - Add jitter to nextSyncAt (10% of interval, capped at 5min) to prevent thundering herd - Fix Date consistency in connector DELETE route * fix(connectors): address PR review feedback on retry and SharePoint batching - Remove 120s cap on Retry-After — pass all values through to retry loop - Add maxDelayMs guard: if Retry-After exceeds maxDelayMs, throw immediately instead of hammering with shorter intervals (addresses validate timeout concern) - Add early exit in SharePoint batch loop when maxFiles limit is reached to avoid unnecessary API calls * fix(connectors): cap Retry-After at maxDelayMs instead of aborting Match Google Cloud SDK behavior: when Retry-After exceeds maxDelayMs, cap the wait to maxDelayMs and log a warning, rather than throwing immediately. This ensures retries are bounded in duration while still respecting server guidance within the configured limit. * fix(connectors): add early-exit guard to Dropbox, Google Docs, OneDrive batch loops Match the SharePoint fix — skip remaining batches once maxFiles limit is reached to avoid unnecessary API calls. * improvement(turbo): align turborepo config with best practices (#3458) * improvement(turbo): align turborepo config with best practices * fix(turbo): address PR review feedback * fix(turbo): add lint:check task for read-only lint+format CI checks lint:check previously delegated to format:check which only checked formatting. Now it runs biome check (no --write) which enforces both lint rules and formatting without mutating files. * upgrade turbo * improvement(perf): apply react and js performance optimizations across codebase (#3459) * improvement(perf): apply react and js performance optimizations across codebase - Parallelize independent DB queries with Promise.all in API routes - Defer PostHog and OneDollarStats via dynamic import() to reduce bundle size - Use functional setState in countdown timers to prevent stale closures - Replace O(n*m) .filter().find() with Set-based O(n) lookups in undo-redo - Use .toSorted() instead of .sort() for immutable state operations - Use lazy initializers for useState(new Set()) across 20 components - Remove useMemo wrapping trivially cheap expressions (typeof, ternary, template strings) - Add passive: true to scroll event listener * fix(perf): address PR review feedback - Extract IIFE Set patterns to named consts for readability in use-undo-redo - Hoist Set construction above loops in BATCH_UPDATE_PARENT cases - Add .catch() error handler to PostHog dynamic import - Convert session-provider posthog import to dynamic import() to complete bundle split * fix(analytics): add .catch() to onedollarstats dynamic import * improvement(resource): tables, files * improvement(resources): all outer page structure complete * refactor(queries): comprehensive TanStack Query best practices audit (#3460) * refactor: comprehensive TanStack Query best practices audit and migration - Add AbortSignal forwarding to all 41 queryFn implementations for proper request cancellation - Migrate manual fetch patterns to useMutation hooks (useResetPassword, useRedeemReferralCode, usePurchaseCredits, useImportWorkflow, useOpenBillingPortal, useAllowedMcpDomains) - Migrate standalone hooks to TanStack Query (use-next-available-slot, use-mcp-server-test, use-webhook-management, use-referral-attribution) - Fix query key factories: add missing `all` keys, replace inline keys with factory methods - Fix optimistic mutations: use onSettled instead of onSuccess for cache reconciliation - Replace overly broad cache invalidations with targeted key invalidation - Remove keepPreviousData from static-key queries where it provides no benefit - Add staleTime to queries missing explicit cache duration - Fix `any` type in UpdateSettingParams with proper GeneralSettings typing - Remove dead code: loadingWebhooks/checkedWebhooks from subblock store, unused helper functions - Update settings components (general, debug, referral-code, credit-balance, subscription, mcp) to use mutation state instead of manual useState for loading/error/success Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: remove unstable mutation object from useCallback deps openBillingPortal mutation object is not referentially stable, but .mutate() is stable in TanStack Query v5. Remove from deps to prevent unnecessary handleBadgeClick recreations. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: add missing byWorkflows invalidation to useUpdateTemplate The onSettled handler was missing the byWorkflows() invalidation that was dropped during the onSuccess→onSettled migration. Without this, the deploy modal (useTemplateByWorkflow) would show stale data after a template update. Co-Authored-By: Claude Opus 4.6 <[email protected]> * docs: add TanStack Query best practices to CLAUDE.md and cursor rules Add comprehensive React Query best practices covering: - Hierarchical query key factories with intermediate plural keys - AbortSignal forwarding in all queryFn implementations - Targeted cache invalidation over broad .all invalidation - onSettled for optimistic mutation cache reconciliation - keepPreviousData only on variable-key queries - No manual fetch in components rule - Stable mutation references in useCallback deps Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: address PR review feedback - Fix syncedRef regression in use-webhook-management: only set syncedRef.current=true when webhook is found, so re-sync works after webhook creation (e.g., post-deploy) - Remove redundant detail(id) invalidation from useUpdateTemplate onSettled since onSuccess already populates cache via setQueryData Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: address second round of PR review feedback - Reset syncedRef when blockId changes in use-webhook-management so component reuse with a different block syncs the new webhook - Add response.ok check in postAttribution so non-2xx responses throw and trigger TanStack Query retry logic Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: use lists() prefix invalidation in useCreateWorkspaceCredential Use workspaceCredentialKeys.lists() instead of .list(workspaceId) so filtered list queries are also invalidated on credential creation, matching the pattern used by update and delete mutations. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: address third round of PR review feedback - Add nullish coalescing fallback for bonusAmount in referral-code to prevent rendering "undefined" when server omits the field - Reset syncedRef when queryEnabled becomes false so webhook data re-syncs when the query is re-enabled without component remount Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: address fourth round of PR review feedback - Add AbortSignal to testMcpServerConnection for consistency - Wrap handleTestConnection in try/catch for mutateAsync error handling - Replace broad subscriptionKeys.all with targeted users()/usage() invalidation - Add intermediate users() key to subscription key factory for prefix matching - Add comment documenting syncedRef null-webhook behavior - Fix api-keys.ts silent error swallowing on non-ok responses - Move deployments.ts cache invalidation from onSuccess to onSettled Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: achieve full TanStack Query best practices compliance - Add intermediate plural keys to api-keys, deployments, and schedules key factories for prefix-based invalidation support - Change copilot-keys from refetchQueries to invalidateQueries - Add signal parameter to organization.ts fetch functions (better-auth client does not support AbortSignal, documented accordingly) - Move useCreateMcpServer invalidation from onSuccess to onSettled Co-Authored-By: Claude Opus 4.6 <[email protected]> --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * ran lint * Fix tables row count * Update mothership to match copilot in logs * improvement(resource): layout * fix(knowledge): compute KB tokenCount from documents instead of stale column (#3463) The knowledge_base.token_count column was initialized to 0 and never updated. Replace with COALESCE(SUM(document.token_count), 0) in all read queries, which already JOIN on documents with GROUP BY. * improvement(resources): layout and items * feat(knowledge): add v1 knowledge base API, Obsidian/Evernote connectors, and docs (#3465) * feat(knowledge): add v1 knowledge base API, Obsidian/Evernote connectors, and docs - Add v1 REST API for knowledge bases (CRUD, document management, vector search) - Add Obsidian and Evernote knowledge base connectors - Add file type validation to v1 file and document upload endpoints - Update OpenAPI spec with knowledge base endpoints and schemas - Add connectors documentation page - Apply query hook formatting improvements * fix(knowledge): address PR review feedback - Remove validateFileType from v1/files route (general file upload, not document-only) - Reject tag filters when searching multiple KBs (tag defs are KB-specific) - Cache tag definitions to avoid duplicate getDocumentTagDefinitions call - Fix Obsidian connector silent empty results when syncContext is undefined * improvement(connectors): add syncContext to getDocument, clean up caching - Update docs to say 20+ connectors - Add syncContext param to ConnectorConfig.getDocument interface - Use syncContext in Evernote getDocument to cache tag/notebook maps - Replace index-based cache check with Map keyed by KB ID in search route * fix(knowledge): address second round of PR review feedback - Fix Zod .default('text') overriding tag definition's actual fieldType - Fix encodeURIComponent breaking multi-level folder paths in Obsidian - Use 413 instead of 400 for file-too-large in document upload - Add knowledge-bases to API reference docs navigation Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(knowledge): prevent cross-workspace KB access in search Filter accessible KBs by matching workspaceId from the request, preventing users from querying KBs in other workspaces they have access to but didn't specify. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(knowledge): audit resourceId, SSRF protection, recursion depth limit - Fix recordAudit using knowledgeBaseId instead of newDocument.id - Add SSRF validation to Obsidian connector (reject private/loopback URLs) - Add max recursion depth (20) to listVaultFiles Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(obsidian): remove SSRF check that blocks localhost usage The Obsidian connector is designed to connect to the Local REST API plugin running on localhost (127.0.0.1:27124). The SSRF check was incorrectly blocking this primary use case. Co-Authored-By: Claude Opus 4.6 <[email protected]> --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * improvement(resources): segmented API * fix(execution): ensure background tasks await post-execution DB status updates (#3466) The fire-and-forget IIFE in execution-core.ts for post-execution logging could be abandoned when trigger.dev tasks exit, leaving executions permanently stuck in "running" status. Store the promise on LoggingSession so background tasks can optionally await it before returning. * improvement(resource): sorting and icons * fix(resource): sorting * improvement(settings): fix mcp modal, add option to edit JSON and add Sim as an MCP client (#3467) * improvement(settings): fix mcp modal, add option to edit JSON and add Sim as an MCP client * added docs link in sidebar * ack comments * ack comments * fixed error msg * feat(mothership): billing (#3464) * Billing update * more billing improvements * credits UI * credit purchase safety * progress * ui improvements * fix cancel sub * fix types * fix daily refresh for teams * make max features differentiated * address bugbot comments * address greptile comments * revert isHosted * address more comments * fix org refresh bar * fix ui rounding * fix minor rounding * fix upgrade issue for legacy plans * fix formatPlanName * fix email dispay names * fix legacy team reference bugs * referral bonus in credits * fix org upgrade bug * improve logs * respect toggle for paid users * fix landing page pro features and usage limit checks * fixed query and usage * add unit test * address more comments * enterprise guard * fix limits bug * pass period start/end for overage * fix(sidebar): restore drag-and-drop for workflows and folders (#3470) * fix(sidebar): restore drag-and-drop for workflows and folders Made-with: Cursor * update docs, unrelated * improvement(tables): consolidation * feat(schedules): add schedule creator modal for standalone jobs Add modal to create standalone scheduled jobs from the Schedules page. Includes POST API endpoint, useCreateSchedule mutation hook, and full modal with schedule type selection, timezone, lifecycle, and live preview. Co-Authored-By: Claude Opus 4.6 <[email protected]> * feat(schedules): add edit support with context menu for standalone jobs * style(schedules): apply linter formatting * improvement: tables, favicon * feat(files): inline file viewer with text editing (#3475) * feat(files): add inline file viewer with text editing and create file modal Add file preview/edit functionality to the workspace files page. Text files (md, json, txt, yaml, etc.) open in an editable textarea with Cmd/Ctrl+S save. PDFs render in an iframe. New file button creates empty .md files via a modal. Uses ResourceHeader breadcrumbs and ResourceOptionsBar for save/download/delete. Co-Authored-By: Claude Opus 4.6 <[email protected]> * improvement(files): add UX polish, PR review fixes, and context menu - Add unsaved changes guard modal (matching credentials manager pattern) - Add delete confirmation modal for both viewer and context menu - Add save status feedback (Save → Saving... → Saved) - Add right-click context menu with Open, Download, Delete actions - Add 50MB file size limit on content update API - Add storage quota check before content updates - Add response.ok guard on download to prevent corrupt files - Add skeleton loading for pending file selection (prevents flicker) - Fix updateContent in handleSave dependency array Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): propagate save errors and remove redundant sizeDiff - Remove try/catch in TextEditor.handleSave so errors propagate to parent, which correctly shows save failure status - Remove redundant inner sizeDiff declaration that shadowed outer scope Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): remove unused textareaRef Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): move Cmd+S to parent, add save error feedback, hide save for non-text files - Move Cmd+S keyboard handler from TextEditor to Files so it goes through the parent handleSave with proper status management - Add 'error' save status with red "Save failed" label that auto-resets - Only show Save button for text-editable file types (md, txt, json, etc.) Co-Authored-By: Claude Opus 4.6 <[email protected]> * improvement(files): add save tooltip, deduplicate text-editable extensions - Add Tooltip on Save button showing Cmd+S / Ctrl+S shortcut - Export TEXT_EDITABLE_EXTENSIONS from file-viewer and reuse in files.tsx instead of duplicating the list inline Co-Authored-By: Claude Opus 4.6 <[email protected]> * refactor: extract isMacPlatform to shared utility Move isMacPlatform() from global-commands-provider.tsx to lib/core/utils/platform.ts so it can be reused by files.tsx tooltip without duplication. Co-Authored-By: Claude Opus 4.6 <[email protected]> * refactor(files): deduplicate delete modal, use shared formatFileSize - Extract DeleteConfirmModal component to eliminate duplicate modal markup between viewer and list modes - Replace local formatFileSize with shared utility from file-utils.ts Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): fix a11y label lint error and remove mutation object from useCallback deps Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): add isDirty guard on handleSave, return proper HTTP status codes Prevents "Saving → Saved" flash when pressing Cmd+S with no changes. Returns 404 for file-not-found and 402 for quota-exceeded instead of 500. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): reset isDirty/saveStatus on delete and discard, remove deprecated navigator.platform - Clear isDirty and saveStatus when deleting the currently-viewed file to prevent spurious beforeunload prompts - Reset saveStatus on discard to prevent stale "Save failed" when opening another file - Remove deprecated navigator.platform, userAgent fallback covers all cases Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): prevent concurrent saves on rapid Cmd+S, add YAML MIME types - Add saveStatus === 'saving' guard to handleSave to prevent duplicate concurrent PUT requests from rapid keyboard shortcuts - Add yaml/yml MIME type mappings to getMimeTypeFromExtension Co-Authored-By: Claude Opus 4.6 <[email protected]> * refactor(files): reuse shared extension constants, parallelize cancelQueries - Replace hand-rolled SUPPORTED_EXTENSIONS with composition from existing SUPPORTED_DOCUMENT/AUDIO/VIDEO_EXTENSIONS in validation.ts - Parallelize sequential cancelQueries calls in delete mutation onMutate Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): guard handleCreate against duplicate calls while pending Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): show upload progress on the Upload button, not New file Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(files): use ref-based guard for create pending state to avoid stale closure The uploadFile.isPending check was stale because the mutation object is excluded from useCallback deps (per codebase convention). Using a ref ensures the guard works correctly across rapid Enter key presses. Co-Authored-By: Claude Opus 4.6 <[email protected]> * cleanup(files): use shared icon import, remove no-op props, wrap handler in useCallback Co-Authored-By: Claude Opus 4.6 <[email protected]> --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * improvement: tables, dropdown * improvement(docs): align sidebar method badges and polish API reference styling (#3484) * improvement(docs): align sidebar method badges and polish API reference styling * fix(docs): revert className prop on DocsPage for CI compatibility * fix(docs): restore oneOf schema for delete rows and use rem units in CSS * fix(docs): replace :has() selectors with direct className for reliable prod layout The API docs layout was intermittently narrow in production because CSS :has(.api-page-header) selectors are unreliable in Tailwind v4 production builds. Apply className="openapi-page" directly to DocsPage and replace all 64 :has() selectors with .openapi-page class targeting. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(docs): bypass TypeScript check for className prop on DocsPage Use spread with type assertion to pass className to DocsPage, working around a CI type resolution issue where the prop exists at runtime but is not recognized by TypeScript in the Vercel build environment. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(docs): use inline style tag for grid layout, revert CSS to :has() selectors The className prop on DocsPage doesn't exist in the fumadocs-ui version resolved on Vercel, so .openapi-page was never applied and all 64 CSS rules broke. Revert to :has(.api-page-header) selectors for styling and use an inline <style> tag for the critical grid-column layout override, which is SSR'd and doesn't depend on any CSS selector matching. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(docs): add pill styling to footer navigation method badges The footer nav badges (POST, GET, etc.) had color from data-method rules but lacked the structural pill styling (padding, border-radius, font-size). Co-Authored-By: Claude Opus 4.6 <[email protected]> --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * fix(docs): use named grid lines instead of numeric column indices (#3487) Root cause: the fumadocs grid template has 3 columns in production but 5 columns in local dev. Our CSS used `grid-column: 3 / span 2` which targeted the wrong column in the 3-column grid, placing content in the near-zero-width TOC column instead of the main content column. Fix: use `grid-column: main-start / toc-end` which uses CSS named grid lines from grid-template-areas, working regardless of column count. Co-authored-by: Claude Opus 4.6 <[email protected]> * improvement(resource): layout * improvement: icon, resource header options * improvement: icons * fix(files): icon * feat(tables): column operations, row ordering, V1 API (#3488) * feat(tables): add column operations, row ordering, V1 columns API, and OpenAPI spec Adds column rename/delete/type change/constraint updates to the tables module, row ordering via position column, UI metadata schema, V1 public API for column operations with rate limiting and audit logging, and OpenAPI documentation. Key changes: - Service-layer column operations with validation (name pattern, type compatibility, unique/required constraints) - Position column on user_table_rows with composite index for efficient ordering - V1 /api/v1/tables/{tableId}/columns endpoint (POST/PATCH/DELETE) with rate limiting and audit - Shared Zod schemas extracted to table/utils.ts using COLUMN_TYPES constant - Targeted React Query invalidation (row vs schema mutations) with consistent onSettled usage - OpenAPI 3.1.0 spec for columns endpoint with code samples - Position field added to all row response mappings for consistency - Sort fallback to position ordering when buildSortClause returns null Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(tables): use specific error prefixes instead of broad "Cannot" match Prevents internal TypeErrors (e.g. "Cannot read properties of undefined") from leaking as 400 responses. Now matches only domain-specific errors: "Cannot delete the last column" and "Cannot set column". Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(tables): reject Infinity and NaN in number type compatibility check Number.isFinite rejects Infinity, -Infinity, and NaN, preventing non-finite values from passing column type validation. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(tables): invalidate table list on row create/delete for stale rowCount Row create and delete mutations now invalidate the table list cache since it includes a computed rowCount. Row updates (which don't change count) continue to only invalidate row queries. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(tables): add column name length check, deduplicate name gen, reset pagination on clear - Add MAX_COLUMN_NAME_LENGTH validation to addTableColumn (was missing, renameColumn already had it) - Extract generateColumnName helper to eliminate triplicated logic across handleAddColumn, handleInsertColumnLeft, handleInsertColumnRight - Reset pagination to page 0 when clearing sort/filter to prevent showing empty pages after narrowing filters are removed Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: hoist tableId above try block in V1 columns route, add detail invalidation to invalidateRowCount - V1 columns route: `tableId` was declared inside `try` but referenced in `catch` logger.error, causing undefined in error logs. Hoisted `await params` above try in all three handlers (POST, PATCH, DELETE). - invalidateRowCount: added `tableKeys.detail(tableId)` invalidation since the single-table GET response includes `rowCount`, which becomes stale after row create/delete without this. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: add position to all row mutation responses, remove dead filter code - Add `position` field to POST (single + batch) and PATCH row responses across both internal and V1 routes, matching GET responses and OpenAPI spec. - Remove unused `filterConfig`, `handleFilterToggle`, `handleFilterClear`, and `activeFilters` — dead code left over from merge conflict resolution. `handleFilterApply` (the one actually wired to JSX) is preserved. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: invalidateTableSchema now also invalidates table list cache Column add/rename/delete/update mutations now invalidate tableKeys.list() since the list endpoint returns schema.columns for each table. Without this, the sidebar table list would show stale column schemas until staleTime expires. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: replace window.prompt/confirm with emcn Modal dialogs Replace non-standard browser dialogs with proper emcn Modal components to match the existing codebase pattern (e.g. delete table confirmation). - Column rename: Modal with Input field + Enter key support - Column delete: Modal with destructive confirmation Co-Authored-By: Claude Opus 4.6 <[email protected]> --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * update schedule creation ui and run lint * improvement: logs * improvement(tables): multi-select and efficiencies * Table tools * improvement(folder-selection): folder deselection + selection order should match visual * fix(selections): more nested folder inaccuracies * Tool updates * Store tool call results * fix(landing): wire agent input to mothership * feat(mothership): resource viewer * fix tests * fix(streaming): smoother streaming with throttled rendering, ResizeObserver scroll, and batched updates (#3471) * fix(streaming): smoother streaming with throttled rendering, ResizeObserver scroll, and batched updates - Add useThrottledValue hook (100ms trailing-edge throttle) to gate DOM re-renders during streaming across all chat surfaces - Replace 100ms setInterval scroll polling with ResizeObserver-based auto-scroll, programmatic scroll timestamp tracking, and nested [data-scrollable] region handling - Extract processContentBuffer from inline content handler for cleaner code organization in copilot SSE handlers - Add RAF-based update batching (50ms max interval) to floating chat and home chat streaming paths - Add useProgressiveList hook for progressive rendering of long conversation histories via requestAnimationFrame Made-with: Cursor * ack PR comments * fix search modal * more comments * ack comments * count * ack comments * ack comment * improvement(mothership): worklfow resource * Fix tool call persistence in chat * Tool results * Fix error status * File uploads to mothership * feat(templates): landing page templates workflow states * improvement(mothership): chat stability * improvement(mothership): chat history and stability * improvement(tables): click-to-select navigation, inline rename, column resize (#3496) * improvement(tables): click-to-select navigation, inline rename, column resize * fix(tables): address PR review comments - Add doneRef guard to useInlineRename preventing Enter+blur double-fire - Fix PATCH error handler: return 500 for non-validation errors, fix unreachable logger.error - Stop click propagation on breadcrumb rename input * fix(tables): add rows-affected check in renameTable service Prevents silent no-op when tableId doesn't match any record. * fix(tables): useMemo deps + placeholder memo initialCharacter check - Use primitive editingId/editValue in useMemo deps instead of whole useInlineRename object (which creates a new ref every render) - Add initialCharacter comparison to placeholderPropsAreEqual, matching the existing pattern in dataRowPropsAreEqual * fix(tables): address round 2 review comments - Mirror name validation (regex + max length) in PatchTableSchema so validateTableName failures return 400 instead of 500 - Add .returning() + rows-affected check to renameWorkspaceFile, matching the renameTable pattern - Check response.ok before parsing JSON in useRenameWorkspaceFile, matching the useRenameTable pattern * refactor(tables): reuse InlineRenameInput in BreadcrumbSegment Replace duplicated inline input markup with the shared component. Eliminates redundant useRef, useEffect, and input boilerplate. * fix(tables): set doneRef in cancelRename to prevent blur-triggered save Escape → cancelRename → input unmounts → blur → submitRename would save instead of canceling. Now cancelRename sets doneRef like submitRename does, blocking the subsequent blur handler. * fix(tables): pointercancel cleanup + typed FileConflictError - Add pointercancel handler to column resize to prevent listener leaks when system interrupts the pointer (touch-action override, etc.) - Replace stringly-typed error.message.includes('already exists') with FileConflictError class for refactor-safe 409 status detection * fix(tables): stable useCallback dep + rename shadowed variable - Use listRename.startRename (stable ref) instead of whole listRename object in handleContextMenuRename deps - Rename inner 'target' to 'origin' in arrow-key handler to avoid shadowing the outer HTMLElement 'target' * fix(tables): move class below imports, stable submitRename, clear editingCell - Move FileConflictError below import statements (import-first convention) - Make submitRename a stable useCallback([]) by reading editingId and editValue through refs (matches existing onSaveRef pattern) - Add setEditingCell(null) to handleEmptyRowClick for symmetry with handleCellClick * feat(tables): persist column widths in table metadata Column widths now survive navigation and page reloads. On resize-end, widths are debounced (500ms) and saved to the table's metadata field via a new PUT /api/table/[tableId]/metadata endpoint. On load, widths are seeded from the server once via React Query. * fix type checking for file viewer * fix(tables): address review feedback — 4 fixes 1. headerRename.onSave now uses the fileId parameter directly instead of the selectedFile closure, preventing rename-wrong-file race 2. updateMetadataMutation uses ref pattern matching mutateRef/createRef 3. Type-to-enter filters non-numeric chars for number columns, non-date chars for date columns 4. renameValue only passed to actively-renaming ColumnHeaderMenu, preserving React.memo for other columns * fix(tables): position-based gap rows, insert above/below, consistency fixes - Fix gap row insert shifting: only shift rows when target position is occupied, preventing unnecessary displacement of rows below - Switch to position-based indexing throughout (positionMap, maxPosition) instead of array-index for correct sparse position handling - Add insert row above/below to context menu - Use CellContent for pending values in PositionGapRows (matching PlaceholderRows) - Add belowHeader selection overlay logic to PositionGapRows - Remove unnecessary 500ms debounce on column width persistence Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix cells nav w keyboard * added preview panel for html, markdown rendering, completed table --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * fix(tables): one small tables ting (#3497) * feat(exa-hosted-key): Restore exa hosted key (#3499) Co-authored-by: Theodore Li <[email protected]> * improvement(ui): consistent styling * styling alignment * improvements(tables): styling improvements * improve resizer for file preview for html files * updated document icon * fix(credentials): exclude regular login methods from credential sync * update docs * upgrade turbo * improvement: tables, chat * Fix table column delete * small table rename bug, files updates not persisting * Table batch ops * fix(credentials): block usage at execution layer without perms + fix invites * feat(hosted-key-services) Add hosted key for multiple services (#3461) * feat(hosted keys): Implement serper hosted key * Handle required fields correctly for hosted keys * Add rate limiting (3 tries, exponential backoff) * Add custom pricing, switch to exa as first hosted key * Add telemetry * Consolidate byok type definitions * Add warning comment if default calculation is used * Record usage to user stats table * Fix unit tests, use cost property * Include more metadata in cost output * Fix disabled tests * Fix spacing * Fix lint * Move knowledge cost restructuring away from generic block handler * Migrate knowledge unit tests * Lint * Fix broken tests * Add user based hosted key throttling * Refactor hosted key handling. Add optimistic handling of throttling for custom throttle rules. * Remove research as hosted key. Recommend BYOK if throtttling occurs * Make adding api keys adjustable via env vars * Remove vestigial fields from research * Make billing actor id required for throttling * Switch to round robin for api key distribution * Add helper method for adding hosted key cost * Strip leading double underscores to avoid breaking change * Lint fix * Remove falsy check in favor for explicit null check * Add more detailed metrics for different throttling types * Fix _costDollars field * Handle hosted agent tool calls * Fail loudly if cost field isn't found * Remove any type * Fix type error * Fix lint * Fix usage log double logging data * Fix test * Add browseruse hosted key * Add firecrawl and serper hosted keys * feat(hosted key): Add exa hosted key (#3221) * feat(hosted keys): Implement serper hosted key * Handle required fields correctly for hosted keys * Add rate limiting (3 tries, exponential backoff) * Add custom pricing, switch to exa as first hosted key * Add telemetry * Consolidate byok type definitions * Add warning comment if default calculation is used * Record usage to user stats table * Fix unit tests, use cost property * Include more metadata in cost output * Fix disabled tests * Fix spacing * Fix lint * Move knowledge cost restructuring away from generic block handler * Migrate knowledge unit tests * Lint * Fix broken tests * Add user based hosted key throttling * Refactor hosted key handling. Add optimistic handling of throttling for custom throttle rules. * Remove research as hosted key. Recommend BYOK if throtttling occurs * Make adding api keys adjustable via env vars * Remove vestigial fields from research * Make billing actor id required for throttling * Switch to round robin for api key distribution * Add helper method for adding hosted key cost * Strip leading double underscores to avoid breaking change * Lint fix * Remove falsy check in favor for explicit null check * Add more detailed metrics for different throttling types * Fix _costDollars field * Handle hosted agent tool calls * Fail loudly if cost field isn't found * Remove any type * Fix type error * Fix lint * Fix usage log double logging data * Fix test --------- Co-authored-by: Theodore Li <[email protected]> * Fail fast on cost data not being found * Add hosted key for google services * Add hosting configuration and pricing logic for ElevenLabs TTS tools * Add linkup hosted key * Add jina hosted key * Add hugging face hosted key * Add perplexity hosting * Add broader metrics for throttling * Add skill for adding hosted key * Lint, remove vestigial hosted keys not implemented * Revert agent changes * fail fast * Fix build issue * Fix build issues * Fix type error * Remove byok types that aren't implemented * Address feedback * Use default model when model id isn't provided * Fix cost default issues * Remove firecrawl error suppression * Restore original behavior for hugging face * Add mistral hosted key * Remove hugging face hosted key * Fix pricing mismatch is mistral and perplexity * Add hosted keys for parallel and brand fetch * Add brandfetch hosted key * Update types * Change byok name to parallel_ai * Add telemetry on unknown models --------- Co-authored-by: Theodore Li <[email protected]> * improvement(settings): SSR prefetch, code splitting, dedicated skeletons * fix: bust browser cache for workspace file downloads The downloadFile function was using a plain fetch() that honored the aggressive cache headers, causing newly created files to download empty. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(settings): use emcn Skeleton in extracted skeleton files * fix(settings): extract shared response mappers to prevent server/client shape drift Addresses PR review feedback — prefetch.ts duplicated response mapping logic from client hooks. Extracted mapGeneralSettingsResponse and mapUserProfileResponse as shared functions used by both client fetch and server prefetch. * update byok page * fix(settings): include theme sync in client-side prefetch queryFn Hover-based prefetchGeneralSettings now calls syncThemeToNextThemes, matching the useGeneralSettings hook behavior so theme updates aren't missed when prefetch refreshes stale cache. * fix(byok): use EMCN Input for search field instead of ui Input Replace @/components/ui Input with the already-imported EmcnInput for design-system consistency. * fix(byok): use ui Input for search bar to match other settings pages * fix(settings): use emcn Input for file input in general settings * improvement(settings): add search bar to skeleton loading states Skeletons now include the search bar (and action button where applicable) so the layout matches the final component 1:1. Eliminates layout shift when the dynamic chunk loads — search bar area is already reserved by the skeleton. * fix(settings): align skeleton layouts with actual component structures - Fix list item gap from 12px to 8px across all skeletons (API keys, custom tools, credentials, MCP) - Add OAuth icon placeholder to credential skeleton - Fix credential button group gap from 8px to 4px - Remove incorrect gap-[4px] from credential-sets text column - Rebuild debug skeleton to match real layout (description + input/button row) - Add scrollable wrapper to BYOK skeleton with more representative item count * chore: lint fixes * improvement(sidebar): match workspace switcher popover width to sidebar Use Radix UI's built-in --radix-popover-trigger-width CSS variable instead of hardcoded 160px so the popover matches the trigger width and responds to sidebar resizing. * revert hardcoded ff * fix: copilot, improvement: tables, mothership * feat: inline chunk editor and table batch ops with undo/redo (#3504) * feat: inline chunk editor and table batch operations with undo/redo Replace modal-based chunk editing/creation with inline editor following the files tab pattern (state-based view toggle with ResourceHeader). Add batch update API endpoint, undo/redo support, and Popover-based context menus for tables. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: remove icons from table context menu PopoverItems Icons were incorrectly carried over from the DropdownMenu migration. PopoverItems in this codebase use text-only labels. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: restore DropdownMenu for table context menu The table-level context menu was incorrectly migrated to Popover during conflict resolution. Only the row-level context menu uses Popover; the table context menu should remain DropdownMenu with icons, matching the base branch. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: bound cross-page chunk navigation polling to max 50 retries Prevent indefinite polling if page data never loads during chunk navigation across page boundaries. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: navigate to last page after chunk creation for multi-page documents After creating a chunk, navigate to the last page (where new chunks append) before selecting it. This prevents the editor from showing "Loading chunk..." when the new chunk is not on the current page. The loading state breadcrumb remains as an escape hatch for edge cases. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: add duplicate rowId validation to BatchUpdateByIdsSchema Adds a .refine() check to reject duplicate rowIds in batch update requests, consistent with the positions uniqueness check on batch insert. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: address PR review comments - Fix disableEdit logic: use || instead of && so connector doc chunks cannot be edited from context menu (row click still opens viewer) - Add uniqueness validation for rowIds in BatchUpdateByIdsSchema - Fix inconsistent bg token: bg-background → bg-[var(--bg)] in Pagination Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: remove duplicate rowId uniqueness refine on BatchUpdateByIdsSchema The refine was applied both on the inner updates array and the outer object. Keep only the inner array refine which is cleaner. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: address additional PR review comments - Fix stale rowId after create-row redo: patch undo stack with new row ID using patchUndoRowId so subsequent undo targets the correct row - Fix text color tokens in Pagination: use CSS variable references (text-[var(--text-body)], text-[var(--text-secondary)]) instead of Tailwind semantic tokens for consistency with the rest of the file Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: remove dead code and fix type errors in table context menu Remove unused `onAddData` prop and `isEmptyCell` variable from row context menu (introduced in PR but never wired to JSX). Fix type errors in optimistic update spreads by removing unnecessary `as Record<string, unknown>` casts that lost the RowData type. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: prevent false "Saved" status on invalid content and mark fire-and-forget goToPage calls ChunkEditor.handleSave now throws on empty/oversized content instead of silently returning, so the parent's catch block correctly sets saveStatus to 'error'. Also added explicit `void` to unawaited goToPage(1) calls in filter handlers to signal intentional fire-and-forget. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: handle stale totalPages in handleChunkCreated for new-page edge case When creating a chunk that spills onto a new page, totalPages in the closure is stale. Now polls displayChunksRef for the new chunk, and if not found, checks totalPagesRef for an updated page count and navigates to the new last page before continuing to poll. Co-Authored-By: Claude Opus 4.6 <[email protected]> --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * Streaming fix -- need to test more * Make mothership block use long input instead of prompt input * improvement(billing): isAnnual metadata + docs updates (#3506) * improvement(billing): on demand toggling and infinite limits * store stripe metadata to distinguish annual vs monthly * udpate docs * address bugbot * Add piping * feat(clean-hosted-keys) Remove eleven labs, browseruse. Tweak firecrawl and mistral key impl (#3503) * Remove eleven labs, browseruse, and firecrawl * Remove creditsUsed output * Add back mistral hosting for mistral blocks * Add back firecrawl since they queue up concurrent requests * Fix price calculation, remove agent since its super long running and will clog up queue * Define hosting per tool * Remove redundant token finding --------- Co-authored-by: Theodore Li <[email protected]> * Update vfs to handle hosted keys * improvement(tables): fix cell editing flash, batch API docs, and UI polish (#3507) * fix: show text cursor in chunk editor and ensure textarea fills container Add cursor-text to the editor wrapper so the whole area shows a text cursor. Click on empty space focuses the textarea. Changed textarea from h-full/w-full to flex-1/min-h-0 so it properly fills the flex container. Co-Authored-By: Claude Opus 4.6 <[email protected]> * improvement(tables): fix cell editing flash, add batch API docs, and UI polish Fix stale-data flash when saving inline cell edits by using TanStack Query's isPending+variables pattern instead of manual cache writes. Also adds OpenAPI docs for batch table endpoints, DatePicker support in row modal, duplicate row in context menu, and styling improvements. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: remove dead resolveColumnFromEvent callback Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: unify paste undo into single create-rows action Batch-created rows from paste now push one `create-rows` undo entry instead of N individual `create-row` entries, so a single Ctrl+Z reverses the entire paste operation. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: validate dates in inline editor and displayToStorage InlineDateEditor now validates computed values via Date.parse before saving, preventing invalid strings like "hello" from being sent to the server. displayToStorage now rejects out-of-range month/day values (e.g. 13/32) instead of producing invalid YYYY-MM-DD strings. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: accept ISO date format in inline date editor Fall back to raw draft input when displayToStorage returns null, so valid ISO dates like "2024-03-15" pasted or typed directly are accepted instead of silently discarded. Date.parse still validates the final value. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: add ISO date support to displayToStorage and fix picker Escape displayToStorage now recognizes YYYY-MM-DD input directly, so ISO dates typed or pasted work correctly for both saving and picker sync. DatePicker Escape now refocuses the input instead of saving, so the user can press Escape again to cancel or Enter to confirm — matching the expected cancel behavior. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: remove dead paste boundary check The totalR guard in handlePaste could never trigger since totalR included pasteRows.length, making targetRow always < totalR. Remove the unused variable and simplify the selection focus calc. Co-Authored-By: Claude Opus 4.6 <[email protected]> * update openapi --------- Co-authored-by: Claude Opus 4.6 <[email protected]> * fix dysfunctional unique operation in tables * feat(autosave): files and chunk editor autosave with debounce + refetch (#3508) * feat(files): debounced autosave while editing * address review comments * more comments * fix: unique constraint check crash and copilot table initial rows - Fix TypeError in updateColumnConstraints: db.execute() returns a plain array with postgres-js, not { rows: [...] }. The .rows.length access always crashed, making "Set unique" completely broken. - Add initialRowCount: 20 to copilot table creation so tables created via chat have the same empty rows as tables created from the UI. Co-Authored-By: Claude Opus 4.6 <[email protected]> * Fix signaling * revert: remove initialRowCount from copilot table creation Copilot populates its own data after creating a table, so pre-creating 20 empty rows causes data to start at position 21 with empty rows above. initialRowCount only makes sense for the manual UI creation flow. Co-Authored-By: Claude Opus 4.6 <[email protected]> * improvement: chat, workspace header * chat metadata * Fix schema mismatch (#3510) Co-authored-by: Theodore Li <[email protected]> * Fixes * fix: manual table creation starts with 1 row, 1 column Manual tables now create with a single 'name' column and 1 row instead of 2 columns and 20 rows. Copilot tables remain at 0 rows, 0 columns. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: horizontal scroll in embedded table by replacing overflow-hidden with overflow-clip Cell content spans used Tailwind's `truncate` (overflow: hidden), creating scroll containers that consumed trackpad wheel events on macOS without propagating to the actual scroll ancestor. Replaced with overflow-clip which clips identically but doesn't create a scroll container. Also moved focus target from outer container to the scroll div for correctness. Co-Authored-By: Claude Opus 4.6 <[email protected]> * Fix tool call ordering * Fix tests * feat: add task multi-select, context menu, and subscription UI updates Add shift-click range selection, cmd/ctrl-click toggle, and right-click context menu for tasks in sidebar matching workflow/folder patterns. Update subscription settings tab UI. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(credentials): autosync behaviour cross workspace (#3511) * fix(credentials): autosync behaviour cross workspace * address comments * fix(api-key-reminder) Add reminder on hosted keys that api key isnt needed (#3512) * Add reminder on hosted keys that api key isnt needed * Fix test case --------- Co-authored-by: Theodore Li <[email protected]> * improvement: sidebar, chat * Usage limit * Plan prompt * fix(sidebar): workspace header collapse * fix(sidebar): task navigation * Subagent tool call persistence * Don't drop suabgent text * improvement(ux): streaming * improvement: thinking * fix(random): optimized kb connector sync engine, rerenders in tables, files, editors, chat (#3513) * optimized kb connector sync engine, rerenders in tables, files, editors, chat * refactor(sidebar): rename onTaskClick to onMultiSelectClick for clarity Made-with: Cursor * ack comments, add docsFailed * feat(email-footer) Add "sent with sim ai" for free users (#3515) * Add "sent with sim ai" for free users * Only add prompt injection on free tier * Add try catch around billing info fetch --------- Co-authored-by: Theodore Li <[email protected]> * improvement: modals * ran migrations * fix(mothership): fix hardcoded workflow color, tables drag line overflowing * feat(mothership): file attachment indicators, persistence, and chat input improvements - Show image thumbnails and file-icon cards above user messages in mothership chat - Persist file attachment metadata (key, filename, media_type, size) in DB with user messages - Restore attachments from history via /api/files/serve/ URLs so they survive refresh/navigation - Unify all chat file inputs to use shared CHAT_ACCEPT_ATTRIBUTE constant - Fix file thumbnail overflow: use flex-wrap instead of hidden horizontal scroll - Compact attachment cards in floating workflow chat messages Co-Authored-By: Claude Opus 4.6 <[email protected]> * improvement: search modal * improvement(usage): free plan to 1000 credits (#3516) * improvement(billing): free plan to five dollars * fix comment * remove per month terminology from marketing * generate migration * remove migration * add migration back * feat(workspace): add workspace color changing, consolidate update hooks, fix popover dismiss - Add workspace color change via context menu, reusing workflow ColorGrid UI - Consolidate useUpdateWorkspaceName + useUpdateWorkspaceColor into useUpdateWorkspace - Fix popover hover submenu dismiss by using DismissableLayerBranch with pointerEvents - Remove passthrough wrapper for export, reuse Workspace type for capturedWorkspaceRef - Reorder log columns: workflow first, merge date+time into single column Co-Authored-By: Claude Opus 4.6 <[email protected]> * Update oauth cred tool * fix(diff-controls): fixed positioning for copilot diff controls * fix(font): added back old font for emcn code editor * improvement: panel, special tags * improvement: chat * improvement: loading and file dropping * feat(templates): create home templates * fix(uploads): resolve .md file upload rejection and deduplicate file type utilities Browsers report empty or application/octet-stream MIME types for .md files, causing copilot uploads to be rejected. Added resolveFileType() utility that falls back to extension-based MIME resolution at both client and server boundaries. Consolidated duplicate MIME mappings into module-level constants, removed duplicate isImageFileType from copilot module, and replaced hardcoded ALLOWED_EXTENSIONS with composition from shared validation constants. Also switched file attachment previews to use shared getDocumentIcon utility. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(home): prevent initial view from being scrollable Co-Authored-By: Claude Opus 4.6 <[email protected]> * autofill fixes * added back integrations page, reverted secrets page back to old UI * Fix workspace dropdown getting cut off when sidebar is collapsed * fix(mothership): lint (#3517) * fix(mothership): lint * fix typing * fix tests * fix stale query * fix plan display name * Feat/add mothership manual workflow runs (#3520) * Add run and open workflow buttons in workflow preview * Send log request message after manual workflow run * Make edges in embedded workflow non-editable * Change chat to pass in log as additional context * Revert "Change chat to pass in log as additional context" This reverts commit e957dffb2ffe29c5ca8d5c2ee43a2e58022bb99d. * Revert "Send log request message after manual workflow run" This reverts commit 0fb92751f087f6da30ce2f937b00206f60fd3fd3. * Move run and workflow icons to tab bar * Simplify boolean condition --------- Co-authored-by: Theodore Li <[email protected]> * feat(resource-tab-scroll): Allow vertical scrolling to scroll resource tab * fix(remove-speed-hosted-key) Remove maps speed limit hosted key, it's deprecated (#3521) Co-authored-by: Theodore Li <[email protected]> * improvement: home, sidebar * fix(download-file): render correct file download link for mothership (#3522) * fix(download-file): render correct file download link for mothership * Fix uunecessary call * Use simple strip instead of db lookup and moving behavior * Make regex strip more strict --------- Co-authored-by: Theodore Li <[email protected]> * improvement: schedules, auto-scroll * fix(settings): navigate back to origin page instead of always going home Use sessionStorage to store the return URL when entering settings, and use router.replace for tab switches so history doesn't accumulate. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(schedules): release lastQueuedAt lock on all exit paths to prevent stuck schedules Multiple error/early-return paths in executeScheduleJob and executeJobInline were exiting without clearing lastQueuedAt, causing the dueFilter to permanently skip those schedules — resulting in stale "X hours ago" display for nextRunAt. Co-Authored-By: Claude Opus 4.6 <[email protected]> * feat(mothership): inline rename for resource tabs + workspace_file rename tool - Add double-click inline rename on file and table resource tabs - Wire useInlineRename + useRenameWorkspaceFile/useRenameTable mutations - Add rename operation to workspace_file copilot tool (schema, server, router) - Add knowledge base resource support (type, extraction, rendering, actions) - Accept optional className on InlineRenameInput for context-specific sizing Co-Authored-By: Claude Opus 4.6 <[email protected]> * revert: remove inline rename UI from resource tabs Keep the workspace_file rename tool for the mothership agent. Only the UI-side inline rename (double-click tabs) is removed. Co-Authored-By: Claude Opus 4.6 <[email protected]> * feat(mothership): knowledge base resource extraction + Resource/ResourceTable refactor - Extract KB resources from knowledge subagent respond format (knowledge_bases array) - Add knowledge_base tool to RESOURCE_TOOL_NAMES and TOOL_UI_METADATA - Extract ResourceTable as independently composable memoized component - Move contentOverride/overlay to Resource shell level (not table primitive) - Remove redundant disableHeaderSort and loadingRows props - Rename internal sort state for clarity (sort → internalSort, sortOverride → externalSort) - Export ResourceTable and ResourceTableProps from barrel Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(logs) Run workflows client side in mothership to transmit logs (#3529) * Run workflows client side in mothership to transmit logs * Initialize set as constant, prevent duplicate execution * Fix lint --------- Co-authored-by: Theodore Li <[email protected]> * fix(import) fix missing file * fix(resource): Hide resources that have been deleted (#3528) * Hide resources that have been deleted * Handle table, workflow not found * Add animation to prevent flash when previous resource was deleted * Fix animation playing on every switch * Run workflows client side in mothership to transmit logs * Fix race condition for animation * Use shared workflow tool util file --------- Co-authored-by: Theodore Li <[email protected]> * fix: chat scrollbar on sidebar collapse/open * edit existing workflow should bring up artifact * fix(agent) subagent and main agent text being merged without spacing * feat(mothership): remove resource-level delete tools from copilot Remove delete operations for workflows, folders, tables, and files from the mothership copilot to prevent destructive actions via AI. Row-level and column-level deletes are preserved. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix: stop sidebar from auto-collapsing when resource panel appears (#3540) The sidebar was forcibly collapsed whenever a resource (e.g. workflow) first appeared in the resource panel during a task. This was disruptive on larger screens where users want to keep both the sidebar and resource panel visible simultaneously. Co-authored-by: Claude Opus 4.6 <[email protected]> * fix(mothership): insert copilot-created workflows at top of list (#3537) * feat(mothership): remove resource-level delete tools from copilot Remove delete operations for workflows, folders, tables, and files from the mothership copilot to prevent destructive actions via AI. Row-level and column-level deletes are preserved. Co-Authored-By: Claude Opus 4.6 <[email protected]> * fix(mothership): insert copilot-created workflows at top of list * fix(mothership): server-side top-insertion sort order and deduplicate registry logic * fix(mothership): include folder sort orders when computing top-insertion position * fix(mothership): use getNextWorkflowColor instead of hardcoded color --------- C…

Theodore Li added 12 commits February 12, 2026 18:32

feat(hosted keys): Implement serper hosted key

2cdb896

Handle required fields correctly for hosted keys

3e6527a

Add rate limiting (3 tries, exponential backoff)

e5c8aec

Add custom pricing, switch to exa as first hosted key

8a78f80

Add telemetry

d174a6a

Consolidate byok type definitions

c12e92c

Add warning comment if default calculation is used

2a36143

Record usage to user stats table

36e6464

Fix unit tests, use cost property

f237d6f

Include more metadata in cost output

0a002fd

Fix disabled tests

36d49ef

Fix spacing

fbd1cdf

TheodoreSpeaks requested a review from icecrasher321 February 14, 2026 07:03

greptile-apps bot reviewed Feb 14, 2026

View reviewed changes

Fix lint

dc4c611

vercel bot temporarily deployed to Preview February 14, 2026 16:56 Inactive

Move knowledge cost restructuring away from generic block handler

68da290

vercel bot temporarily deployed to Preview February 16, 2026 05:47 Inactive

TheodoreSpeaks changed the title ~~feat(hosted key): Add hosted key, exa as shared key4.~~ feat(hosted key): Add hosted key, exa as shared key Feb 16, 2026

Migrate knowledge unit tests

ce02a30

vercel bot temporarily deployed to Preview February 16, 2026 06:04 Inactive

Lint

e6d98c6

vercel bot temporarily deployed to Preview February 16, 2026 06:25 Inactive

TheodoreSpeaks marked this pull request as draft March 5, 2026 19:30

Theodore Li added 2 commits March 5, 2026 11:34

Fix broken tests

ecdbe29

Merge branch 'staging' into feat/sim-provided-key

2325535

vercel bot deployed to Preview March 5, 2026 19:51 View deployment

Add user based hosted key throttling

693a3d3

vercel bot temporarily deployed to Preview March 7, 2026 01:59 Inactive

cursor bot reviewed Mar 7, 2026

View reviewed changes

apps/sim/blocks/blocks/exa.ts Show resolved Hide resolved

apps/sim/providers/utils.ts Show resolved Hide resolved

icecrasher321 reviewed Mar 7, 2026

View reviewed changes

Theodore Li added 2 commits March 6, 2026 19:10

Fail loudly if cost field isn't found

ee2e123

Remove any type

09a1b5c

vercel bot temporarily deployed to Preview March 7, 2026 03:18 Inactive

Fix type error

0836131

vercel bot temporarily deployed to Preview March 7, 2026 03:22 Inactive

Fix lint

427627a

vercel bot temporarily deployed to Preview March 7, 2026 03:25 Inactive

cursor bot reviewed Mar 7, 2026

View reviewed changes

apps/sim/providers/index.ts Show resolved Hide resolved

Fix usage log double logging data

d29d613

vercel bot temporarily deployed to Preview March 7, 2026 03:49 Inactive

cursor bot reviewed Mar 7, 2026

View reviewed changes

Fix test

3e94ce3

vercel bot temporarily deployed to Preview March 7, 2026 03:56 Inactive

cursor bot reviewed Mar 7, 2026

View reviewed changes

apps/sim/tools/index.ts Show resolved Hide resolved

TheodoreSpeaks merged commit 158d523 into staging Mar 7, 2026
12 checks passed

TheodoreSpeaks deleted the feat/sim-provided-key branch March 7, 2026 18:07

waleedlatif1 mentioned this pull request Mar 9, 2026

v0.5.109: obsidian and evernote integrations, slack fixes, remove memory instrumentation #3493

Merged

TheodoreSpeaks pushed a commit that referenced this pull request Mar 9, 2026

Revert "feat(hosted key): Add exa hosted key (#3221)"

fe577b0

This reverts commit 158d523.

waleedlatif1 pushed a commit that referenced this pull request Mar 9, 2026

Revert "feat(hosted key): Add exa hosted key (#3221)" (#3495)

635179d

This reverts commit 158d523. Co-authored-by: Theodore Li <[email protected]>

greptile-apps bot mentioned this pull request Mar 9, 2026

Revert "feat(hosted key): Add exa hosted key (#3221)" #3495

Merged

10 tasks

TheodoreSpeaks pushed a commit that referenced this pull request Mar 10, 2026

Reapply "feat(hosted key): Add exa hosted key (#3221)" (#3495)

4161a20

This reverts commit 635179d.

TheodoreSpeaks mentioned this pull request Mar 10, 2026

feat(exa-hosted-key): Restore exa hosted key #3499

Merged

10 tasks

Conversation

TheodoreSpeaks commented Feb 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Important: API keys need to be stored in prod and staging before merging:

Type of Change

Testing

Checklist

Screenshots/Videos

Uh oh!

vercel bot commented Feb 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

greptile-apps bot commented Feb 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 3/5

Important Files Changed

Sequence Diagram

Comments Outside Diff (3)

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

icecrasher321 Mar 7, 2026

Choose a reason for hiding this comment

Uh oh!

icecrasher321 Mar 7, 2026

Choose a reason for hiding this comment

Uh oh!

icecrasher321 Mar 7, 2026

Choose a reason for hiding this comment

Uh oh!

icecrasher321 Mar 7, 2026

Choose a reason for hiding this comment

Uh oh!

TheodoreSpeaks Mar 7, 2026

Choose a reason for hiding this comment

Uh oh!

icecrasher321 Mar 7, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

cursor bot Mar 7, 2026

Choose a reason for hiding this comment

Unhandled pricing error crashes successful tool execution

Uh oh!

cursor bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

TheodoreSpeaks commented Feb 14, 2026 •

edited

Loading

vercel bot commented Feb 14, 2026 •

edited

Loading

greptile-apps bot commented Feb 14, 2026 •

edited

Loading