yunchan/d3ro-voice
1
0
Fork 0
Code Issues Pull requests Projects Releases 2 Packages Wiki Activity Actions

feat(V2-5차): pgvector 시맨틱 검색 + chat 스트리밍 + doc 편집 + VoiceAction + Realtime + storage + 메모

Browse source 
묶음 Q — pgvector + 시맨틱 검색:
- migrations/20260410000004_pgvector_knowledge.sql
  - vector extension, embedding vector(1536) 컬럼
  - ivfflat cosine 인덱스
  - match_knowledge_chunks(query_embedding, match_count, similarity_threshold) RPC
  - RLS: user_id 또는 소속 팀 기준
- functions/embed-chunks: 문서 소유권 확인 후 OpenAI text-embedding-3-small 배치 호출 → knowledge_chunks.embedding 업데이트
- functions/search-knowledge: 쿼리 텍스트 → OpenAI 임베딩 → user 권한 RPC 호출 → 상위 청크 반환
- config.toml에 embed-chunks/search-knowledge 등록
- components/knowledge/knowledge-search.tsx: 검색창 + 결과 카드(유사도 %)
- /knowledge 페이지에 검색 UI 추가

묶음 R — /chat SSE 스트리밍:
- functions/llm-proxy: Anthropic Messages API stream 지원
  - ANTHROPIC_API_KEY 없으면 SSE placeholder 스트림
  - stream=true일 때 response.body 그대로 프록시 (text/event-stream)
  - stream=false는 JSON 응답
- components/chat/chat-panel.tsx:
  - stream=true로 요청
  - ReadableStream 파싱 (SSE: data: {type:content_block_delta, delta:{text_delta}})
  - assistantId 메시지를 progressive 업데이트, scrollToBottom
  - 불필요한 LlmResponse 인터페이스 제거

묶음 S — DocumentEditor:
- components/meetings/document-editor.tsx
  - 문서 박스 클릭 → MUI Dialog (fullWidth, maxWidth md)
  - TextField multiline 20~40 rows, monospace
  - 제목 편집 + 저장/삭제 버튼
- meetings/[id] 페이지 Documents 섹션을 DocumentEditor로 교체 (+ typoSx 미사용 import 제거)

묶음 T — /actions (VoiceAction 이식):
- components/actions/action-runner.tsx
  - SYSTEM_PROMPT로 JSON 스키마 강제 (create_meeting/search_knowledge/create_memo/send_team_invite/unknown)
  - LLM 응답에서 JSON 추출 → 파싱 → 확인 후 실행
  - 각 type별 실행 로직 (meetings/memos INSERT, 검색은 안내만)
- app/(app)/actions/page.tsx + Sidebar Actions 메뉴 + AutoAwesomeIcon

묶음 U — Realtime + Storage + 메모 UI:
- CloudSyncService:
  - RealtimeChannel import 추가
  - startRealtime(): meetings/history/dictionary 변경 구독, debounce 후 pullAll 자동 트리거
  - stopRealtime(), signIn 직후/세션 복원 시 자동 시작, signOut 시 종료
- apps/web/components/meetings/memo-form.tsx
  - TextField + 저장 버튼, 회의 시작 기준 경과 ms 자동 계산
  - Realtime 구독자에게 자동 전파
- meetings/[id] MEMOS 섹션에 MemoForm 렌더
- apps/web/components/record/mic-recorder.tsx
  - STT 성공 후 Supabase Storage 'audio' 버킷에 {user_id}/{ts}.webm 업로드
  - meetings 테이블에 INSERT (raw_transcript, audio_storage_key, duration_ms, ended_at)
  - Storage/meetings 실패는 전사 결과는 유지하며 경고

묶음 V — 11개 locale nav 키:
- en/ja/zh/zh-TW/es/fr/de/pt/ru/vi/th 에 nav.chat/nav.knowledge/nav.actions 추가
- ko.json에 nav.actions 추가
- Sidebar에 Actions 메뉴(AutoAwesomeIcon) 등록

검증:
- desktop typecheck + build OK
- web typecheck + build OK (15 라우트: 기존 14 + /actions)
- api-client test 19 passed
- 회귀 없음

통계:
- 총 Edge Functions 12개 (embed-chunks/search-knowledge 추가)
- 총 SQL 마이그레이션 8개
- 웹 라우트 15개 (accept-invite/actions/billing/chat/dashboard/knowledge/login/meetings/meetings[id]/record/teams/teams[id]/auth-callback/root/_not-found)
This commit is contained in:
yunchan8804 2026-04-10 09:30:41 +09:00
parent 1fa24ce3c9
commit b386733d1e
29 changed files with 1363 additions and 81 deletions
Download patch file Download diff file Expand all files Collapse all files

6
server/supabase/config.toml
View file

@ -102,5 +102,11 @@ verify_jwt = true
[functions.send-push]
verify_jwt = true
[functions.embed-chunks]
verify_jwt = true
[functions.search-knowledge]
verify_jwt = true
[analytics]
enabled = false

159
server/supabase/functions/embed-chunks/index.ts Normal file
View file

@ -0,0 +1,159 @@
// server/supabase/functions/embed-chunks/index.ts
// 특정 document의 청크들을 OpenAI 임베딩으로 변환하여 knowledge_chunks.embedding 컬럼 업데이트.
//
// 요청:
// POST { document_id: "uuid" }
// 응답:
// { embedded: number, errors: string[] }
//
// 환경변수:
// OPENAI_API_KEY (text-embedding-3-small 사용)
import { corsHeaders, handleCorsPreflightRequest } from '../_shared/cors.ts'
import { requireUser, authErrorResponse, type AuthError } from '../_shared/auth.ts'
import { createServiceRoleClient } from '../_shared/quota.ts'
interface EmbedRequest {
document_id: string
}
interface OpenAIEmbeddingResponse {
data: Array<{ embedding: number[]; index: number }>
model: string
usage: { prompt_tokens: number; total_tokens: number }
}
// @ts-expect-error — Deno 런타임 전역
Deno.serve(async (req: Request) => {
const preflight = handleCorsPreflightRequest(req)
if (preflight) return preflight
if (req.method !== 'POST') {
return new Response(JSON.stringify({ error: 'Method not allowed' }), {
status: 405,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})
}
try {
const user = await requireUser(req)
const body = (await req.json()) as EmbedRequest
if (!body.document_id) {
return new Response(JSON.stringify({ error: 'document_id 필요' }), {
status: 400,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})
}
// @ts-expect-error — Deno.env
const openaiKey = Deno.env.get('OPENAI_API_KEY') ?? ''
if (!openaiKey) {
return new Response(
JSON.stringify({ error: 'openai_not_configured', message: 'OPENAI_API_KEY 미설정' }),
{ status: 503, headers: { ...corsHeaders, 'Content-Type': 'application/json' } }
)
}
const serviceClient = createServiceRoleClient()
// 문서 소유권 확인
const { data: doc } = await serviceClient
.from('knowledge_documents')
.select('id, user_id')
.eq('id', body.document_id)
.maybeSingle()
if (!doc || (doc.user_id as string) !== user.id) {
return new Response(
JSON.stringify({ error: 'forbidden' }),
{ status: 403, headers: { ...corsHeaders, 'Content-Type': 'application/json' } }
)
}
// 임베딩 대상 청크 조회 (embedding IS NULL인 것만)
const { data: chunks, error: chunksErr } = await serviceClient
.from('knowledge_chunks')
.select('id, content')
.eq('document_id', body.document_id)
.is('embedding', null)
if (chunksErr) {
throw new Error(`청크 조회 실패: ${chunksErr.message}`)
}
if (!chunks || chunks.length === 0) {
return new Response(JSON.stringify({ embedded: 0, errors: [] }), {
status: 200,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})
}
// OpenAI 임베딩 API 배치 호출 (한 번에 최대 100개)
const errors: string[] = []
let embedded = 0
const batchSize = 100
for (let i = 0; i < chunks.length; i += batchSize) {
const batch = chunks.slice(i, i + batchSize) as Array<{ id: string; content: string }>
try {
const resp = await fetch('https://api.openai.com/v1/embeddings', {
method: 'POST',
headers: {
Authorization: `Bearer ${openaiKey}`,
'Content-Type': 'application/json'
},
body: JSON.stringify({
model: 'text-embedding-3-small',
input: batch.map((c) => c.content),
dimensions: 1536
})
})
if (!resp.ok) {
const errText = await resp.text()
errors.push(`OpenAI ${resp.status}: ${errText.slice(0, 200)}`)
continue
}
const data = (await resp.json()) as OpenAIEmbeddingResponse
// 각 청크에 embedding 업데이트
for (const item of data.data) {
const chunk = batch[item.index]
const { error: updateErr } = await serviceClient
.from('knowledge_chunks')
.update({ embedding: item.embedding })
.eq('id', chunk.id)
if (updateErr) {
errors.push(`chunk ${chunk.id}: ${updateErr.message}`)
} else {
embedded++
}
}
} catch (e) {
errors.push(e instanceof Error ? e.message : String(e))
}
}
// 문서 indexed 플래그 업데이트
await serviceClient
.from('knowledge_documents')
.update({ indexed: true, indexed_at: new Date().toISOString() })
.eq('id', body.document_id)
return new Response(JSON.stringify({ embedded, errors }), {
status: 200,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})
} catch (err) {
if (err && typeof err === 'object' && 'status' in err && 'message' in err) {
return authErrorResponse(err as AuthError, corsHeaders)
}
const message = err instanceof Error ? err.message : 'Unknown error'
return new Response(JSON.stringify({ error: message }), {
status: 500,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})
}
})

124
server/supabase/functions/llm-proxy/index.ts
View file

@ -78,50 +78,100 @@ Deno.serve(async (req: Request) => {
)
}
// Anthropic API 호출 — placeholder
//
// 실제 구현 시:
// const anthropicKey = Deno.env.get('ANTHROPIC_API_KEY')!
// const resp = await fetch('https://api.anthropic.com/v1/messages', {
// method: 'POST',
// headers: {
// 'Content-Type': 'application/json',
// 'x-api-key': anthropicKey,
// 'anthropic-version': '2023-06-01'
// },
// body: JSON.stringify({
// model: requestedModel,
// max_tokens: body.max_tokens ?? 2048,
// system: body.system,
// messages: body.messages,
// stream: body.stream ?? false
// })
// })
// if (body.stream) {
// return new Response(resp.body, {
// headers: { ...corsHeaders, 'Content-Type': 'text/event-stream' }
// })
// }
// const data = await resp.json()
// ...
// Anthropic API 호출
// @ts-expect-error — Deno.env
const anthropicKey = Deno.env.get('ANTHROPIC_API_KEY') ?? ''
await consumeQuota(user.id, 'llm_process', serviceClient, 1)
const placeholder = {
id: `msg_placeholder_${Date.now()}`,
model: requestedModel,
role: 'assistant',
content: [
if (!anthropicKey) {
// Placeholder 응답 (키 미설정 시)
if (body.stream) {
// 스트리밍 placeholder — SSE로 "설정되지 않음" 메시지 전송
const encoder = new TextEncoder()
const stream = new ReadableStream({
start(controller) {
const msg = '[llm-proxy placeholder — ANTHROPIC_API_KEY 미설정]'
for (const ch of msg) {
controller.enqueue(
encoder.encode(
`data: ${JSON.stringify({ type: 'content_block_delta', delta: { type: 'text_delta', text: ch } })}\n\n`
)
)
}
controller.enqueue(encoder.encode('data: [DONE]\n\n'))
controller.close()
}
})
return new Response(stream, {
status: 200,
headers: {
...corsHeaders,
'Content-Type': 'text/event-stream',
'Cache-Control': 'no-cache',
Connection: 'keep-alive'
}
})
}
return new Response(
JSON.stringify({
id: `msg_placeholder_${Date.now()}`,
model: requestedModel,
role: 'assistant',
content: [
{
type: 'text',
text: '[llm-proxy placeholder — ANTHROPIC_API_KEY 미설정]'
}
],
stop_reason: 'end_turn',
usage: { input_tokens: 0, output_tokens: 0 }
}),
{
type: 'text',
text: '[llm-proxy placeholder — Anthropic API not yet wired]'
status: 200,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
}
],
stop_reason: 'end_turn',
usage: { input_tokens: 0, output_tokens: 0 }
)
}
return new Response(JSON.stringify(placeholder), {
// 실제 Anthropic API 호출
const anthropicResp = await fetch('https://api.anthropic.com/v1/messages', {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'x-api-key': anthropicKey,
'anthropic-version': '2023-06-01'
},
body: JSON.stringify({
model: requestedModel,
max_tokens: body.max_tokens ?? 2048,
system: body.system,
messages: body.messages,
stream: body.stream ?? false
})
})
if (!anthropicResp.ok) {
const errText = await anthropicResp.text()
throw new Error(`Anthropic ${anthropicResp.status}: ${errText.slice(0, 500)}`)
}
if (body.stream && anthropicResp.body) {
// SSE 스트림을 그대로 전달
return new Response(anthropicResp.body, {
status: 200,
headers: {
...corsHeaders,
'Content-Type': 'text/event-stream',
'Cache-Control': 'no-cache',
Connection: 'keep-alive'
}
})
}
const data = await anthropicResp.json()
return new Response(JSON.stringify(data), {
status: 200,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})

114
server/supabase/functions/search-knowledge/index.ts Normal file
View file

@ -0,0 +1,114 @@
// server/supabase/functions/search-knowledge/index.ts
// 쿼리 텍스트 → OpenAI 임베딩 → match_knowledge_chunks RPC → 상위 청크 반환.
//
// 요청:
// POST { query: "...", count?: number }
// 응답:
// { results: Array<{ id, document_id, chunk_index, content, similarity }> }
import { corsHeaders, handleCorsPreflightRequest } from '../_shared/cors.ts'
import { requireUser, authErrorResponse, type AuthError } from '../_shared/auth.ts'
interface SearchRequest {
query: string
count?: number
}
// @ts-expect-error — Deno
import { createClient } from 'https://esm.sh/@supabase/supabase-js@2.103.0'
// @ts-expect-error — Deno 런타임 전역
Deno.serve(async (req: Request) => {
const preflight = handleCorsPreflightRequest(req)
if (preflight) return preflight
if (req.method !== 'POST') {
return new Response(JSON.stringify({ error: 'Method not allowed' }), {
status: 405,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})
}
try {
await requireUser(req) // 인증만 확인. RPC는 auth.uid() 기반 RLS
const body = (await req.json()) as SearchRequest
if (!body.query || body.query.trim().length === 0) {
return new Response(JSON.stringify({ error: 'query 필요' }), {
status: 400,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})
}
// @ts-expect-error — Deno.env
const openaiKey = Deno.env.get('OPENAI_API_KEY') ?? ''
if (!openaiKey) {
return new Response(
JSON.stringify({ error: 'openai_not_configured' }),
{ status: 503, headers: { ...corsHeaders, 'Content-Type': 'application/json' } }
)
}
// 1) 쿼리 임베딩
const embedResp = await fetch('https://api.openai.com/v1/embeddings', {
method: 'POST',
headers: {
Authorization: `Bearer ${openaiKey}`,
'Content-Type': 'application/json'
},
body: JSON.stringify({
model: 'text-embedding-3-small',
input: body.query,
dimensions: 1536
})
})
if (!embedResp.ok) {
const errText = await embedResp.text()
throw new Error(`임베딩 실패: ${errText.slice(0, 200)}`)
}
const embedData = (await embedResp.json()) as {
data: Array<{ embedding: number[] }>
}
const queryEmbedding = embedData.data[0]?.embedding
if (!queryEmbedding) {
throw new Error('임베딩 응답이 비어있음')
}
// 2) RLS 컨텍스트에서 RPC 호출 (authenticated 유저 토큰 사용)
const authHeader = req.headers.get('Authorization') ?? ''
// @ts-expect-error — Deno.env
const supabaseUrl = Deno.env.get('SUPABASE_URL') ?? ''
// @ts-expect-error — Deno.env
const anonKey = Deno.env.get('SUPABASE_ANON_KEY') ?? ''
const userClient = createClient(supabaseUrl, anonKey, {
global: { headers: { Authorization: authHeader } }
})
const { data: matches, error: rpcErr } = await userClient.rpc('match_knowledge_chunks', {
query_embedding: queryEmbedding,
match_count: body.count ?? 5,
similarity_threshold: 0.5
})
if (rpcErr) {
throw new Error(`RPC 실패: ${rpcErr.message}`)
}
return new Response(JSON.stringify({ results: matches ?? [] }), {
status: 200,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})
} catch (err) {
if (err && typeof err === 'object' && 'status' in err && 'message' in err) {
return authErrorResponse(err as AuthError, corsHeaders)
}
const message = err instanceof Error ? err.message : 'Unknown error'
return new Response(JSON.stringify({ error: message }), {
status: 500,
headers: { ...corsHeaders, 'Content-Type': 'application/json' }
})
}
})

64
server/supabase/migrations/20260410000004_pgvector_knowledge.sql Normal file
View file

@ -0,0 +1,64 @@
-- ============================================================================
-- Phase V2-Q: pgvector 활성화 + knowledge 시맨틱 검색
-- embedding 컬럼 + cosine 유사도 검색 RPC
-- ============================================================================
CREATE EXTENSION IF NOT EXISTS vector;
-- 기존 knowledge_chunks에 embedding 컬럼 추가 (1536 차원 — OpenAI text-embedding-3-small)
ALTER TABLE public.knowledge_chunks
ADD COLUMN embedding vector(1536);
-- IVFFlat 인덱스 (빠른 근사 최근접)
-- 데이터 삽입 후 `REINDEX TABLE knowledge_chunks;` 권장
CREATE INDEX idx_knowledge_chunks_embedding
ON public.knowledge_chunks
USING ivfflat (embedding vector_cosine_ops)
WITH (lists = 100);
-- ============================================================================
-- match_knowledge_chunks: 쿼리 임베딩과 가장 유사한 청크 반환
-- ============================================================================
CREATE OR REPLACE FUNCTION public.match_knowledge_chunks(
query_embedding vector(1536),
match_count integer DEFAULT 5,
similarity_threshold double precision DEFAULT 0.5
)
RETURNS TABLE (
id uuid,
document_id uuid,
chunk_index integer,
content text,
similarity double precision
)
LANGUAGE plpgsql
STABLE
AS $$
BEGIN
RETURN QUERY
SELECT
kc.id,
kc.document_id,
kc.chunk_index,
kc.content,
(1 - (kc.embedding <=> query_embedding))::double precision AS similarity
FROM public.knowledge_chunks kc
INNER JOIN public.knowledge_documents kd ON kd.id = kc.document_id
WHERE
kc.embedding IS NOT NULL
AND (
kd.user_id = auth.uid()
OR (
kd.team_id IS NOT NULL
AND kd.team_id IN (
SELECT team_id FROM public.team_members WHERE user_id = auth.uid()
)
)
)
AND (1 - (kc.embedding <=> query_embedding)) > similarity_threshold
ORDER BY kc.embedding <=> query_embedding
LIMIT match_count;
END;
$$;
GRANT EXECUTE ON FUNCTION public.match_knowledge_chunks TO authenticated;