improvement(mothership): make user_table limit cap internal, not model-facing

The model can now pass any limit — no "cannot exceed 1000" rejection. 1000
becomes an internal threshold: query_rows clamps the page to MAX_QUERY_LIMIT
(totalCount signals truncation; the model pages with offset), and bulk filter
ops above the cap run as background jobs.

update_rows_by_filter loads full row data inline, so an explicit limit above
the cap escalates to the background worker with a new maxRows budget (the worker
stops after maxRows; update has no read mask so the cap is exact). delete only
loads ids inline, so an explicit limit (any size) stays inline — only unbounded
deletes use the masked background path, which would over-hide a bounded delete.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

apps/sim/lib/copilot/generated/tool-catalog-v1.ts

@@ -3959,7 +3959,7 @@ export const UserTable: ToolCatalogEntry = {
           limit: {
             type: 'number',
             description:
-              'Maximum rows to return or affect (default 100, max 1000). For update_rows_by_filter / delete_rows_by_filter, omit to act on every match — large match sets run as a background job.',
+              'Maximum rows to return or affect (optional, default 100). Any value is allowed — large operations run in the background automatically. Omit on update_rows_by_filter / delete_rows_by_filter to act on every match.',
           },
           mapping: {
             type: 'object',

apps/sim/lib/copilot/generated/tool-schemas-v1.ts

@@ -3687,7 +3687,7 @@ export const TOOL_RUNTIME_SCHEMAS: Record<string, ToolRuntimeSchemaEntry> = {
             limit: {
               type: 'number',
               description:
-                'Maximum rows to return or affect (default 100, max 1000). For update_rows_by_filter / delete_rows_by_filter, omit to act on every match — large match sets run as a background job.',
+                'Maximum rows to return or affect (optional, default 100). Any value is allowed — large operations run in the background automatically. Omit on update_rows_by_filter / delete_rows_by_filter to act on every match.',
             },
             mapping: {
               type: 'object',

apps/sim/lib/copilot/tools/server/table/user-table.test.ts

@@ -692,15 +692,15 @@ describe('userTableServerTool.query_rows', () => {
     })
   })
 
-  it('rejects limits above MAX_QUERY_LIMIT', async () => {
+  it('clamps an over-large query limit to MAX_QUERY_LIMIT instead of rejecting', async () => {
     const result = await userTableServerTool.execute(
       { operation: 'query_rows', args: { tableId: 'tbl_1', limit: 100000 } },
       { userId: 'user-1', workspaceId: 'workspace-1' }
     )
 
-    expect(result.success).toBe(false)
-    expect(result.message).toBe('Limit cannot exceed 1000')
-    expect(mockQueryRows).not.toHaveBeenCalled()
+    expect(result.success).toBe(true)
+    const options = mockQueryRows.mock.calls[0][1] as Record<string, unknown>
+    expect(options.limit).toBe(1000)
   })
 
   it('queries without execution metadata and passes limit/offset through', async () => {
@@ -732,7 +732,7 @@ describe('userTableServerTool.delete_rows_by_filter', () => {
     })
   })
 
-  it('rejects limits above MAX_BULK_OPERATION_SIZE', async () => {
+  it('runs an explicit large limit inline without escalating (delete loads only ids)', async () => {
     const result = await userTableServerTool.execute(
       {
         operation: 'delete_rows_by_filter',
@@ -741,9 +741,11 @@ describe('userTableServerTool.delete_rows_by_filter', () => {
       { userId: 'user-1', workspaceId: 'workspace-1' }
     )
 
-    expect(result.success).toBe(false)
-    expect(result.message).toBe('Limit cannot exceed 1000')
-    expect(mockDeleteRowsByFilter).not.toHaveBeenCalled()
+    expect(result.success).toBe(true)
+    // An explicit limit never counts/escalates — it deletes inline, bounded by the limit.
+    expect(mockQueryRows).not.toHaveBeenCalled()
+    expect(mockDeleteRowsByFilter).toHaveBeenCalledTimes(1)
+    expect(mockDeleteRowsByFilter.mock.calls[0][1]).toMatchObject({ limit: 5000 })
   })
 
   it('deletes inline when the unbounded match count is within the cap', async () => {
@@ -853,7 +855,14 @@ describe('userTableServerTool.update_rows_by_filter', () => {
     mockQueryRows.mockResolvedValue({ rows: [], rowCount: 0, totalCount: 5, limit: 1, offset: 0 })
   })
 
-  it('rejects limits above MAX_BULK_OPERATION_SIZE', async () => {
+  it('escalates an explicit limit above the cap to a background update with maxRows', async () => {
+    mockQueryRows.mockResolvedValueOnce({
+      rows: [],
+      rowCount: 0,
+      totalCount: 20000,
+      limit: 1,
+      offset: 0,
+    })
     const result = await userTableServerTool.execute(
       {
         operation: 'update_rows_by_filter',
@@ -861,9 +870,16 @@ describe('userTableServerTool.update_rows_by_filter', () => {
       },
       { userId: 'user-1', workspaceId: 'workspace-1' }
     )
-    expect(result.success).toBe(false)
-    expect(result.message).toBe('Limit cannot exceed 1000')
+    await flushDetached()
+
+    expect(result.success).toBe(true)
+    // target = min(limit 5000, matchCount 20000) = 5000, above the inline cap → background.
+    expect(result.data?.affectedCount).toBe(5000)
     expect(mockUpdateRowsByFilter).not.toHaveBeenCalled()
+    const [, , type, payload] = mockMarkTableJobRunning.mock.calls[0]
+    expect(type).toBe('update')
+    expect(payload).toMatchObject({ affectedCount: 5000, maxRows: 5000 })
+    expect(mockRunTableUpdate.mock.calls[0][0]).toMatchObject({ maxRows: 5000 })
   })
 
   it('updates inline when the unbounded match count is within the cap', async () => {
@@ -909,6 +925,8 @@ describe('userTableServerTool.update_rows_by_filter', () => {
       cutoff: expect.any(String),
       data: { age: 1 },
     })
+    // Unbounded match (no explicit limit) → the worker patches every match, no cap.
+    expect((payload as { maxRows?: number }).maxRows).toBeUndefined()
     expect(mockRunTableUpdate).toHaveBeenCalledTimes(1)
     expect(mockRunTableUpdate.mock.calls[0][0]).toMatchObject({
       jobId,

apps/sim/lib/copilot/tools/server/table/user-table.ts

@@ -200,8 +200,9 @@ async function dispatchUpdateJob(params: {
   filter: Filter
   data: RowData
   cutoff: Date
+  maxRows?: number
 }): Promise<void> {
-  const { jobId, tableId, workspaceId, filter, data, cutoff } = params
+  const { jobId, tableId, workspaceId, filter, data, cutoff, maxRows } = params
   if (isTriggerDevEnabled) {
     try {
       const [{ tableUpdateTask }, { tasks }] = await Promise.all([
@@ -210,7 +211,7 @@ async function dispatchUpdateJob(params: {
       ])
       await tasks.trigger<typeof tableUpdateTask>(
         'table-update',
-        { jobId, tableId, workspaceId, filter, data, cutoff: cutoff.toISOString() },
+        { jobId, tableId, workspaceId, filter, data, cutoff: cutoff.toISOString(), maxRows },
         { tags: [`tableId:${tableId}`, `jobId:${jobId}`] }
       )
     } catch (error) {
@@ -219,10 +220,12 @@ async function dispatchUpdateJob(params: {
     }
   } else {
     runDetached('table-update', () =>
-      runTableUpdate({ jobId, tableId, workspaceId, filter, data, cutoff }).catch(async (error) => {
-        await markTableUpdateFailed(tableId, jobId, error)
-        throw error
-      })
+      runTableUpdate({ jobId, tableId, workspaceId, filter, data, cutoff, maxRows }).catch(
+        async (error) => {
+          await markTableUpdateFailed(tableId, jobId, error)
+          throw error
+        }
+      )
     )
   }
 }
@@ -280,17 +283,16 @@ function parseDeploymentMode(value: unknown): WorkflowGroupDeploymentMode | unde
 }
 
 /**
- * Validates an optional row limit against the same bounds the HTTP contracts
- * enforce. Returns an error message, or `null` when the limit is acceptable.
+ * Validates an optional row limit. There's no upper bound the caller must respect — the model may
+ * ask for any number. `MAX_QUERY_LIMIT` / `MAX_BULK_OPERATION_SIZE` are applied internally instead
+ * (query_rows clamps the page; bulk ops above the bound run as a background job). Returns an error
+ * message, or `null` when the limit is acceptable.
  */
-function limitError(limit: unknown, max: number): string | null {
+function limitError(limit: unknown): string | null {
   if (limit === undefined) return null
   if (typeof limit !== 'number' || !Number.isInteger(limit) || limit < 1) {
     return 'Limit must be an integer of at least 1'
   }
-  if (limit > max) {
-    return `Limit cannot exceed ${max}`
-  }
   return null
 }
 
@@ -579,7 +581,7 @@ export const userTableServerTool: BaseServerTool<UserTableArgs, UserTableResult>
             return { success: false, message: 'Workspace ID is required' }
           }
 
-          const queryLimitError = limitError(args.limit, TABLE_LIMITS.MAX_QUERY_LIMIT)
+          const queryLimitError = limitError(args.limit)
           if (queryLimitError) {
             return { success: false, message: queryLimitError }
           }
@@ -592,12 +594,18 @@ export const userTableServerTool: BaseServerTool<UserTableArgs, UserTableResult>
           const requestId = generateId().slice(0, 8)
           const idByName = buildIdByName(table.schema)
           const nameById = buildNameById(table.schema)
+          // The model may request any number; we serve at most MAX_QUERY_LIMIT per page so a single
+          // tool result can't drain a whole table. `totalCount` in the response signals truncation,
+          // and the model pages with `offset`.
           const result = await queryRows(
             table,
             {
               filter: args.filter ? filterNamesToIds(args.filter, idByName) : undefined,
               sort: args.sort ? sortNamesToIds(args.sort, idByName) : undefined,
-              limit: args.limit,
+              limit:
+                args.limit !== undefined
+                  ? Math.min(args.limit, TABLE_LIMITS.MAX_QUERY_LIMIT)
+                  : undefined,
               offset: args.offset,
               withExecutions: false,
             },
@@ -700,7 +708,7 @@ export const userTableServerTool: BaseServerTool<UserTableArgs, UserTableResult>
           if (!workspaceId) {
             return { success: false, message: 'Workspace ID is required' }
           }
-          const updateLimitError = limitError(args.limit, TABLE_LIMITS.MAX_BULK_OPERATION_SIZE)
+          const updateLimitError = limitError(args.limit)
           if (updateLimitError) {
             return { success: false, message: updateLimitError }
           }
@@ -715,29 +723,34 @@ export const userTableServerTool: BaseServerTool<UserTableArgs, UserTableResult>
           const idFilter = filterNamesToIds(args.filter, idByName)
           const idData = rowDataNameToId(args.data, idByName)
 
-          // Unbounded "update everything matching": measure the blast radius first and hand
-          // anything past the inline cap to the background update worker — same escalation as
-          // delete_rows_by_filter, so a broad update on a huge table doesn't load every matching
-          // row into this request. A patch touching a unique column stays inline (the service
-          // rejects bulk-setting a unique value across multiple rows).
+          // Inline handles up to MAX_BULK_OPERATION_SIZE rows in one request; a larger operation
+          // (an explicit limit above the cap, or unbounded "update everything matching") runs in the
+          // background worker so a broad update on a huge table doesn't load every matching row into
+          // this request. A small explicit limit is the fast path — no count needed. A patch
+          // touching a unique column always stays inline (the service rejects bulk-setting a unique
+          // value across multiple rows).
           const patchTouchesUnique = table.schema.columns.some(
             (c) => c.unique === true && (c.id ?? c.name) in idData
           )
-          if (args.limit === undefined && !patchTouchesUnique) {
+          const updateInlineEligible =
+            args.limit !== undefined && args.limit <= TABLE_LIMITS.MAX_BULK_OPERATION_SIZE
+          if (!updateInlineEligible && !patchTouchesUnique) {
             const { totalCount } = await queryRows(
               table,
               { filter: idFilter, limit: 1, withExecutions: false },
               requestId
             )
             const matchCount = totalCount ?? 0
-            if (matchCount > TABLE_LIMITS.MAX_BULK_OPERATION_SIZE) {
+            const target = args.limit !== undefined ? Math.min(args.limit, matchCount) : matchCount
+            if (target > TABLE_LIMITS.MAX_BULK_OPERATION_SIZE) {
               const cutoff = new Date()
               const jobId = generateId()
               const payload: TableUpdateJobPayload = {
                 filter: idFilter,
                 data: idData,
                 cutoff: cutoff.toISOString(),
-                affectedCount: matchCount,
+                affectedCount: target,
+                maxRows: args.limit,
               }
               assertNotAborted()
               const claimed = await markTableJobRunning(table.id, jobId, 'update', payload)
@@ -751,11 +764,12 @@ export const userTableServerTool: BaseServerTool<UserTableArgs, UserTableResult>
                 filter: idFilter,
                 data: idData,
                 cutoff,
+                maxRows: args.limit,
               })
               return {
                 success: true,
-                message: `Started background update of ${matchCount} matching rows (job ${jobId}). Rows update in the background — query_rows to check progress. Note: background updates don't auto-recompute workflow/enrichment columns; use run_column afterward if needed.`,
-                data: { jobId, affectedCount: matchCount },
+                message: `Started background update of ${target} matching rows (job ${jobId}). Rows update in the background — query_rows to check progress. Note: background updates don't auto-recompute workflow/enrichment columns; use run_column afterward if needed.`,
+                data: { jobId, affectedCount: target },
               }
             }
           }
@@ -789,7 +803,7 @@ export const userTableServerTool: BaseServerTool<UserTableArgs, UserTableResult>
           if (!workspaceId) {
             return { success: false, message: 'Workspace ID is required' }
           }
-          const deleteLimitError = limitError(args.limit, TABLE_LIMITS.MAX_BULK_OPERATION_SIZE)
+          const deleteLimitError = limitError(args.limit)
           if (deleteLimitError) {
             return { success: false, message: deleteLimitError }
           }
@@ -803,10 +817,11 @@ export const userTableServerTool: BaseServerTool<UserTableArgs, UserTableResult>
           const idByName = buildIdByName(table.schema)
           const idFilter = filterNamesToIds(args.filter, idByName)
 
-          // Unbounded "delete everything matching": measure the blast radius
-          // first, and hand anything past the inline cap to the background
-          // delete worker (same path as the UI's select-all delete) instead of
-          // loading every matching row id into this request.
+          // An explicit limit runs inline (delete loads only row ids, so even a large bounded
+          // delete is light). Only an unbounded "delete everything matching" measures the blast
+          // radius and hands off to the background delete worker (same path as the UI's select-all
+          // delete) — the read-path mask hides exactly the all-matching set, which a bounded delete
+          // would over-hide.
           if (args.limit === undefined) {
             const { totalCount } = await queryRows(
               table,

apps/sim/lib/table/types.ts

@@ -231,6 +231,8 @@ export interface TableUpdateJobPayload {
   /** ISO timestamp; rows created after it are not patched. */
   cutoff: string
   affectedCount?: number
+  /** Stop after updating this many rows (an explicit caller-supplied limit). Omitted = every match. */
+  maxRows?: number
 }
 
 /**

apps/sim/lib/table/update-runner.test.ts

@@ -163,6 +163,23 @@ describe('runTableUpdate', () => {
     expect(mockUpdatePageByIds).not.toHaveBeenCalled()
   })
 
+  it('stops once maxRows is reached and never over-fetches a page', async () => {
+    // budget 3 with page size 2: first page fills 2, second page is capped to the remaining 1.
+    mockSelectRowDataPage
+      .mockResolvedValueOnce([row('a'), row('b')])
+      .mockResolvedValueOnce([row('c')])
+
+    await runTableUpdate(basePayload({ maxRows: 3 }))
+
+    expect(mockSelectRowDataPage).toHaveBeenCalledTimes(2)
+    expect(mockSelectRowDataPage.mock.calls[0][0]).toMatchObject({ limit: 2 })
+    expect(mockSelectRowDataPage.mock.calls[1][0]).toMatchObject({ limit: 1 })
+    expect(mockUpdatePageByIds).toHaveBeenCalledTimes(2)
+    expect(mockAppendTableEvent).toHaveBeenCalledWith(
+      expect.objectContaining({ status: 'ready', progress: 3 })
+    )
+  })
+
   it('passes the cutoff and filter clause through to the page query', async () => {
     mockSelectRowDataPage.mockResolvedValueOnce([])
 

apps/sim/lib/table/update-runner.ts

@@ -37,6 +37,8 @@ export interface TableUpdatePayload {
   data: RowData
   /** Only rows created at/before this instant are patched, so mid-job inserts are spared. */
   cutoff: Date
+  /** Stop after updating this many rows (an explicit caller-supplied limit). Omitted = every match. */
+  maxRows?: number
 }
 
 /**
@@ -57,8 +59,9 @@ export interface TableUpdatePayload {
  * failed via `markTableUpdateFailed`. A superseded run returns quietly.
  */
 export async function runTableUpdate(payload: TableUpdatePayload): Promise<void> {
-  const { jobId, tableId, workspaceId, filter, data, cutoff } = payload
+  const { jobId, tableId, workspaceId, filter, data, cutoff, maxRows } = payload
   const requestId = generateId().slice(0, 8)
+  const budget = maxRows ?? Number.POSITIVE_INFINITY
 
   try {
     const table = await getTableById(tableId, { includeArchived: true })
@@ -81,7 +84,7 @@ export async function runTableUpdate(payload: TableUpdatePayload): Promise<void>
     let lastReported = resumed
     let afterId: string | undefined
 
-    while (true) {
+    while (processed < budget) {
       const owns = await updateJobProgress(tableId, processed, jobId)
       if (!owns) throw new JobSupersededError()
 
@@ -91,7 +94,7 @@ export async function runTableUpdate(payload: TableUpdatePayload): Promise<void>
         cutoff,
         filterClause,
         afterId,
-        limit: TABLE_LIMITS.DELETE_PAGE_SIZE,
+        limit: Math.min(TABLE_LIMITS.DELETE_PAGE_SIZE, budget - processed),
         // Skip rows already carrying the patch so a retried run resumes without re-walking /
         // double-counting the rows an earlier attempt updated (updated rows still exist and may
         // still match the filter, unlike deletes).