Merge pull request #15325 from Budibase/budi-8833-allow-ai-column-to-be-used-in-view-calculations

Allow use of AI fields in view calculations.

packages/pro

@@ -1 +1 @@
-Subproject commit 32d84f109d4edc526145472a7446327312151442
+Subproject commit 15b7f0907e66e4144338404bb071bc1ccfc98137

packages/server/src/api/controllers/row/staticFormula.ts

@@ -162,6 +162,7 @@ export async function finaliseRow(
     dynamic: false,
     contextRows: [enrichedRow],
   })
+
   const aiEnabled =
     ((await features.isEnabled(FeatureFlag.BUDIBASE_AI)) &&
       (await pro.features.isBudibaseAIEnabled())) ||
@@ -184,11 +185,6 @@ export async function finaliseRow(
   enrichedRow = await processFormulas(table, enrichedRow, {
     dynamic: false,
   })
-  if (aiEnabled) {
-    enrichedRow = await processAIColumns(table, enrichedRow, {
-      contextRows: [enrichedRow],
-    })
-  }
 
   // this updates the related formulas in other rows based on the relations to this row
   if (updateFormula) {

packages/server/src/api/routes/tests/row.spec.ts

@@ -8,7 +8,13 @@ import {
 import tk from "timekeeper"
 import emitter from "../../../../src/events"
 import { outputProcessing } from "../../../utilities/rowProcessor"
-import { context, InternalTable, tenancy, utils } from "@budibase/backend-core"
+import {
+  context,
+  setEnv,
+  InternalTable,
+  tenancy,
+  utils,
+} from "@budibase/backend-core"
 import { quotas } from "@budibase/pro"
 import {
   AIOperationEnum,
@@ -42,19 +48,8 @@ import { InternalTables } from "../../../db/utils"
 import { withEnv } from "../../../environment"
 import { JsTimeoutError } from "@budibase/string-templates"
 import { isDate } from "../../../utilities"
-
-jest.mock("@budibase/pro", () => ({
-  ...jest.requireActual("@budibase/pro"),
-  ai: {
-    LargeLanguageModel: {
-      forCurrentTenant: async () => ({
-        llm: {},
-        run: jest.fn(() => `Mock LLM Response`),
-        buildPromptFromAIOperation: jest.fn(),
-      }),
-    },
-  },
-}))
+import nock from "nock"
+import { mockChatGPTResponse } from "../../../tests/utilities/mocks/openai"
 
 const timestamp = new Date("2023-01-26T11:48:57.597Z").toISOString()
 tk.freeze(timestamp)
@@ -99,6 +94,8 @@ if (descriptions.length) {
         const ds = await dsProvider()
         datasource = ds.datasource
         client = ds.client
+
+        mocks.licenses.useCloudFree()
       })
 
       afterAll(async () => {
@@ -172,10 +169,6 @@ if (descriptions.length) {
         )
       }
 
-      beforeEach(async () => {
-        mocks.licenses.useCloudFree()
-      })
-
       const getRowUsage = async () => {
         const { total } = await config.doInContext(undefined, () =>
           quotas.getCurrentUsageValues(
@@ -3224,10 +3217,17 @@ if (descriptions.length) {
       isInternal &&
         describe("AI fields", () => {
           let table: Table
+          let envCleanup: () => void
 
           beforeAll(async () => {
             mocks.licenses.useBudibaseAI()
             mocks.licenses.useAICustomConfigs()
+            envCleanup = setEnv({
+              OPENAI_API_KEY: "sk-abcdefghijklmnopqrstuvwxyz1234567890abcd",
+            })
+
+            mockChatGPTResponse("Mock LLM Response")
+
             table = await config.api.table.save(
               saveTableRequest({
                 schema: {
@@ -3251,7 +3251,9 @@ if (descriptions.length) {
           })
 
           afterAll(() => {
-            jest.unmock("@budibase/pro")
+            nock.cleanAll()
+            envCleanup()
+            mocks.licenses.useCloudFree()
           })
 
           it("should be able to save a row with an AI column", async () => {

packages/server/src/api/routes/tests/viewV2.spec.ts

@@ -1,4 +1,5 @@
 import {
+  AIOperationEnum,
   ArrayOperator,
   BasicOperator,
   BBReferenceFieldSubType,
@@ -42,7 +43,9 @@ import {
 } from "../../../integrations/tests/utils"
 import merge from "lodash/merge"
 import { quotas } from "@budibase/pro"
-import { context, db, events, roles } from "@budibase/backend-core"
+import { context, db, events, roles, setEnv } from "@budibase/backend-core"
+import { mockChatGPTResponse } from "../../../tests/utilities/mocks/openai"
+import nock from "nock"
 
 const descriptions = datasourceDescribe({ exclude: [DatabaseName.MONGODB] })
 
@@ -100,6 +103,7 @@ if (descriptions.length) {
 
       beforeAll(async () => {
         await config.init()
+        mocks.licenses.useCloudFree()
 
         const ds = await dsProvider()
         rawDatasource = ds.rawDatasource
@@ -109,7 +113,6 @@ if (descriptions.length) {
 
       beforeEach(() => {
         jest.clearAllMocks()
-        mocks.licenses.useCloudFree()
       })
 
       describe("view crud", () => {
@@ -507,7 +510,6 @@ if (descriptions.length) {
             })
 
             it("readonly fields can be used on free license", async () => {
-              mocks.licenses.useCloudFree()
               const table = await config.api.table.save(
                 saveTableRequest({
                   schema: {
@@ -933,6 +935,95 @@ if (descriptions.length) {
                 }
               )
             })
+
+          isInternal &&
+            describe("AI fields", () => {
+              let envCleanup: () => void
+              beforeAll(() => {
+                mocks.licenses.useBudibaseAI()
+                mocks.licenses.useAICustomConfigs()
+                envCleanup = setEnv({
+                  OPENAI_API_KEY: "sk-abcdefghijklmnopqrstuvwxyz1234567890abcd",
+                })
+
+                mockChatGPTResponse(prompt => {
+                  if (prompt.includes("elephant")) {
+                    return "big"
+                  }
+                  if (prompt.includes("mouse")) {
+                    return "small"
+                  }
+                  if (prompt.includes("whale")) {
+                    return "big"
+                  }
+                  return "unknown"
+                })
+              })
+
+              afterAll(() => {
+                nock.cleanAll()
+                envCleanup()
+                mocks.licenses.useCloudFree()
+              })
+
+              it("can use AI fields in view calculations", async () => {
+                const table = await config.api.table.save(
+                  saveTableRequest({
+                    schema: {
+                      animal: {
+                        name: "animal",
+                        type: FieldType.STRING,
+                      },
+                      bigOrSmall: {
+                        name: "bigOrSmall",
+                        type: FieldType.AI,
+                        operation: AIOperationEnum.CATEGORISE_TEXT,
+                        categories: "big,small",
+                        columns: ["animal"],
+                      },
+                    },
+                  })
+                )
+
+                const view = await config.api.viewV2.create({
+                  tableId: table._id!,
+                  name: generator.guid(),
+                  type: ViewV2Type.CALCULATION,
+                  schema: {
+                    bigOrSmall: {
+                      visible: true,
+                    },
+                    count: {
+                      visible: true,
+                      calculationType: CalculationType.COUNT,
+                      field: "animal",
+                    },
+                  },
+                })
+
+                await config.api.row.save(table._id!, {
+                  animal: "elephant",
+                })
+
+                await config.api.row.save(table._id!, {
+                  animal: "mouse",
+                })
+
+                await config.api.row.save(table._id!, {
+                  animal: "whale",
+                })
+
+                const { rows } = await config.api.row.search(view.id, {
+                  sort: "bigOrSmall",
+                  sortOrder: SortOrder.ASCENDING,
+                })
+                expect(rows).toHaveLength(2)
+                expect(rows[0].bigOrSmall).toEqual("big")
+                expect(rows[1].bigOrSmall).toEqual("small")
+                expect(rows[0].count).toEqual(2)
+                expect(rows[1].count).toEqual(1)
+              })
+            })
         })
 
         describe("update", () => {
@@ -1836,7 +1927,6 @@ if (descriptions.length) {
               },
             })
 
-            mocks.licenses.useCloudFree()
             const view = await getDelegate(res)
             expect(view.schema?.one).toEqual(
               expect.objectContaining({ visible: true, readonly: true })

packages/server/src/sdk/app/views/index.ts

@@ -432,6 +432,21 @@ export async function enrichSchema(
       ...tableSchema[key],
       ...ui,
       order: anyViewOrder ? ui?.order ?? undefined : tableSchema[key]?.order,
+      // When this was written, the only column types in FieldSchema to have columns
+      // field were the relationship columns. We blank this out here to make sure it's
+      // not set on non-relationship columns, then below we populate it by calling
+      // populateRelSchema.
+      //
+      // For Budibase 3.0 we introduced the FieldType.AI fields. Some of these fields
+      // have `columns: string[]` and it flew under the radar here because the
+      // AIFieldMetadata type isn't a union on its subtypes, it has a collection of
+      // optional fields. So columns is `columns?: string[]` which allows undefined,
+      // and doesn't fail this type check.
+      //
+      // What this means in practice is when FieldType.AI fields get enriched, we
+      // delete their `columns`. At the time of writing, I don't believe anything in
+      // the frontend depends on this, but it is odd and will probably bite us at
+      // some point.
       columns: undefined,
     }
 

packages/server/src/tests/utilities/mocks/openai.ts

@@ -0,0 +1,46 @@
+import nock from "nock"
+
+let chatID = 1
+
+export function mockChatGPTResponse(
+  response: string | ((prompt: string) => string)
+) {
+  return nock("https://api.openai.com")
+    .post("/v1/chat/completions")
+    .reply(200, (uri, requestBody) => {
+      let content = response
+      if (typeof response === "function") {
+        const messages = (requestBody as any).messages
+        content = response(messages[0].content)
+      }
+
+      chatID++
+
+      return {
+        id: `chatcmpl-${chatID}`,
+        object: "chat.completion",
+        created: Math.floor(Date.now() / 1000),
+        model: "gpt-4o-mini",
+        system_fingerprint: `fp_${chatID}`,
+        choices: [
+          {
+            index: 0,
+            message: { role: "assistant", content },
+            logprobs: null,
+            finish_reason: "stop",
+          },
+        ],
+        usage: {
+          prompt_tokens: 0,
+          completion_tokens: 0,
+          total_tokens: 0,
+          completion_tokens_details: {
+            reasoning_tokens: 0,
+            accepted_prediction_tokens: 0,
+            rejected_prediction_tokens: 0,
+          },
+        },
+      }
+    })
+    .persist()
+}

packages/server/src/utilities/rowProcessor/utils.ts

@@ -160,7 +160,7 @@ export async function processAIColumns<T extends Row | Row[]>(
 
           return tracer.trace("processAIColumn", {}, async span => {
             span?.addTags({ table_id: table._id, column })
-            const llmResponse = await llmWrapper.run(prompt!)
+            const llmResponse = await llmWrapper.run(prompt)
             return {
               ...row,
               [column]: llmResponse,

packages/types/src/documents/app/row.ts

@@ -154,6 +154,7 @@ export const GroupByTypes = [
   FieldType.BOOLEAN,
   FieldType.DATETIME,
   FieldType.BIGINT,
+  FieldType.AI,
 ]
 
 export function canGroupBy(type: FieldType) {

packages/types/src/documents/app/table/schema.ts

@@ -123,7 +123,7 @@ export interface AIFieldMetadata extends BaseFieldSchema {
   operation: AIOperationEnum
   columns?: string[]
   column?: string
-  categories?: string[]
+  categories?: string
   prompt?: string
   language?: string
 }