wacrm/src/lib/ai/knowledge.test.ts

import { describe, it, expect, vi, beforeEach } from 'vitest'
import type { SupabaseClient } from '@supabase/supabase-js'

const h = vi.hoisted(() => ({ embedTexts: vi.fn() }))
vi.mock('./embeddings', () => ({
  embedTexts: h.embedTexts,
  toVectorLiteral: (v: number[]) => `[${v.join(',')}]`,
}))

import { retrieveKnowledge, ingestDocument } from './knowledge'

interface FakeState {
  semantic: { id: string; content: string }[]
  fts: { id: string; content: string }[]
  rpcCalls: string[]
  inserted: Record<string, unknown>[] | null
  deletedFor: string | null
}

function makeDb() {
  const state: FakeState = {
    semantic: [],
    fts: [],
    rpcCalls: [],
    inserted: null,
    deletedFor: null,
  }
  const db = {
    rpc: (name: string) => {
      state.rpcCalls.push(name)
      if (name === 'match_ai_knowledge_semantic')
        return Promise.resolve({ data: state.semantic, error: null })
      if (name === 'match_ai_knowledge_fts')
        return Promise.resolve({ data: state.fts, error: null })
      return Promise.resolve({ data: null, error: null })
    },
    from: () => ({
      delete: () => ({
        eq: (_col: string, val: string) => {
          state.deletedFor = val
          return Promise.resolve({ error: null })
        },
      }),
      insert: (rows: Record<string, unknown>[]) => {
        state.inserted = rows
        return Promise.resolve({ error: null })
      },
    }),
  }
  return { db: db as unknown as SupabaseClient, state }
}

beforeEach(() => {
  h.embedTexts.mockReset()
  h.embedTexts.mockImplementation(async (_key: string, inputs: string[]) =>
    inputs.map((_, i) => [i, i]),
  )
})

describe('retrieveKnowledge', () => {
  it('returns [] for an empty query without touching the DB', async () => {
    const { db, state } = makeDb()
    expect(await retrieveKnowledge(db, 'acct', { embeddingsApiKey: null }, '  ')).toEqual([])
    expect(state.rpcCalls).toEqual([])
  })

  it('uses lexical FTS only when there is no embeddings key', async () => {
    const { db, state } = makeDb()
    state.fts = [{ id: 'f1', content: 'F1' }]
    const out = await retrieveKnowledge(db, 'acct', { embeddingsApiKey: null }, 'q')
    expect(out).toEqual(['F1'])
    expect(state.rpcCalls).toEqual(['match_ai_knowledge_fts'])
    expect(h.embedTexts).not.toHaveBeenCalled()
  })

  it('uses semantic search when an embeddings key is present', async () => {
    const { db, state } = makeDb()
    state.semantic = [
      { id: 's1', content: 'S1' },
      { id: 's2', content: 'S2' },
      { id: 's3', content: 'S3' },
    ]
    const out = await retrieveKnowledge(db, 'acct', { embeddingsApiKey: 'sk-x' }, 'q', 3)
    expect(out).toEqual(['S1', 'S2', 'S3'])
    expect(h.embedTexts).toHaveBeenCalledTimes(1)
    // Enough semantic hits → no FTS top-up.
    expect(state.rpcCalls).toEqual(['match_ai_knowledge_semantic'])
  })

  it('tops up with FTS and dedupes when semantic is short', async () => {
    const { db, state } = makeDb()
    state.semantic = [
      { id: 's1', content: 'S1' },
      { id: 's2', content: 'S2' },
    ]
    state.fts = [
      { id: 's2', content: 'S2-dup' }, // dedup by id
      { id: 'f1', content: 'F1' },
    ]
    const out = await retrieveKnowledge(db, 'acct', { embeddingsApiKey: 'sk-x' }, 'q', 3)
    expect(out).toEqual(['S1', 'S2', 'F1'])
    expect(state.rpcCalls).toEqual([
      'match_ai_knowledge_semantic',
      'match_ai_knowledge_fts',
    ])
  })
})

describe('ingestDocument', () => {
  it('embeds chunks when a key is present', async () => {
    const { db, state } = makeDb()
    await ingestDocument(db, 'acct', { embeddingsApiKey: 'sk-x' }, 'doc-1', 'hello world')
    expect(h.embedTexts).toHaveBeenCalledTimes(1)
    expect(state.deletedFor).toBe('doc-1')
    expect(state.inserted).toHaveLength(1)
    expect(state.inserted![0].embedding).toBe('[0,0]') // literal from mocked embed
    expect(state.inserted![0].account_id).toBe('acct')
  })

  it('stores chunks without embeddings when there is no key', async () => {
    const { db, state } = makeDb()
    await ingestDocument(db, 'acct', { embeddingsApiKey: null }, 'doc-1', 'hello world')
    expect(h.embedTexts).not.toHaveBeenCalled()
    expect(state.inserted![0].embedding).toBeNull()
  })

  it('deletes existing chunks and inserts nothing for empty content', async () => {
    const { db, state } = makeDb()
    await ingestDocument(db, 'acct', { embeddingsApiKey: 'sk-x' }, 'doc-1', '   ')
    expect(state.deletedFor).toBe('doc-1')
    expect(state.inserted).toBeNull()
    expect(h.embedTexts).not.toHaveBeenCalled()
  })
})