26 KiB
26 KiB
DreamChat Database Schema
Overview
PostgreSQL with pgvector extension for vector storage. All data is stored locally (offline-first).
Extensions Required
-- Enable required extensions
CREATE EXTENSION IF NOT EXISTS "uuid-ossp";
CREATE EXTENSION IF NOT EXISTS "pgvector";
Entity Relationship Diagram
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ users │ │ characters │ │ conversations │
├─────────────────┤ ├─────────────────┤ ├─────────────────┤
│ id (PK) │◄──────│ user_id (FK) │ │ id (PK) │
│ email │ │ id (PK) │◄──────│ character_id(FK)│
│ username │ │ name │ │ user_id (FK) │
│ password_hash │ │ avatar_url │ │ title │
│ keycloak_sub │ │ personality │ │ created_at │
│ role │ │ attributes │ │ updated_at │
│ created_at │ │ created_at │ └────────┬────────┘
│ updated_at │ │ updated_at │ │
└─────────────────┘ └────────┬────────┘ │
│ │
┌────────┴────────┐ │
│character_knowledge│ │
├─────────────────┤ │
│ id (PK) │◄───────────────┤
│ character_id │ │
│ name │ │
│ source_type │ │
│ raw_content │ │
│ status │ │
└────────┬────────┘ │
│ │
┌─────────────────┐ ┌────────┴────────┐ ┌────────┴────────┐
│import_documents │ │ vector_memories│ │ messages │
├─────────────────┤ ├─────────────────┤ ├─────────────────┤
│ id (PK) │ │ id (PK) │ │ id (PK) │
│ user_id (FK) │ │ content │ │ conversation_id │
│ source_type │ │ embedding │ │ role │
│ source_name │ │ memory_type │ │ content │
│ content │ │ conversation_id │ │ tokens_used │
│ status │ │ character_id │ │ model │
└─────────────────┘ │ knowledge_id │ │ metadata │
│ created_at │ │ created_at │
└─────────────────┘ └─────────────────┘
│
┌────────┴────────┐
│ story_branches │ (Phase 2)
├─────────────────┤
│ id (PK) │
│ conversation_id │
│ parent_id (FK) │
│ content │
│ direction │
│ metadata │
│ created_at │
└─────────────────┘
Table Definitions
1. users
Stores user account information. Supports both Keycloak (OIDC) and local password authentication.
CREATE TABLE users (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
email VARCHAR(255) UNIQUE NOT NULL,
username VARCHAR(100) UNIQUE NOT NULL,
password_hash VARCHAR(255), -- NULL if using Keycloak
keycloak_sub VARCHAR(255) UNIQUE, -- NULL if using password auth
role VARCHAR(20) DEFAULT 'USER' CHECK (role IN ('USER', 'ADMIN')),
is_active BOOLEAN DEFAULT true,
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
updated_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
-- At least one auth method must be set
CONSTRAINT auth_method_check CHECK (
(password_hash IS NOT NULL) OR (keycloak_sub IS NOT NULL)
)
);
CREATE INDEX idx_users_email ON users(email);
CREATE INDEX idx_users_keycloak_sub ON users(keycloak_sub);
2. characters
Character definitions with complex attribute system. Character knowledge is stored separately in character_knowledge with embeddings.
CREATE TABLE characters (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
user_id UUID NOT NULL REFERENCES users(id) ON DELETE CASCADE,
name VARCHAR(255) NOT NULL,
avatar_url TEXT,
-- Core personality prompt sent to LLM
personality_prompt TEXT NOT NULL,
-- Complex attribute system (structured JSON)
-- Example: {"traits": ["brave", "witty"], "age": 25, "species": "human"}
attributes JSONB DEFAULT '{}',
-- Character configuration
config JSONB DEFAULT '{
"model": "openai/gpt-4o",
"temperature": 0.7,
"max_tokens": 2048,
"memory_enabled": true
}',
is_public BOOLEAN DEFAULT false,
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
updated_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
);
CREATE INDEX idx_characters_user_id ON characters(user_id);
CREATE INDEX idx_characters_name ON characters(name);
CREATE INDEX idx_characters_attributes ON characters USING GIN(attributes);
3. conversations
Chat sessions between user and character.
CREATE TABLE conversations (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
user_id UUID NOT NULL REFERENCES users(id) ON DELETE CASCADE,
character_id UUID NOT NULL REFERENCES characters(id) ON DELETE CASCADE,
title VARCHAR(255), -- Auto-generated or user-defined
-- Context window management
message_count INTEGER DEFAULT 0,
total_tokens INTEGER DEFAULT 0,
-- Conversation settings
settings JSONB DEFAULT '{}',
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
updated_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
);
CREATE INDEX idx_conversations_user_id ON conversations(user_id);
CREATE INDEX idx_conversations_character_id ON conversations(character_id);
CREATE INDEX idx_conversations_created_at ON conversations(created_at);
4. messages
Individual chat messages.
CREATE TYPE message_role AS ENUM ('user', 'assistant', 'system');
CREATE TABLE messages (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
conversation_id UUID NOT NULL REFERENCES conversations(id) ON DELETE CASCADE,
role message_role NOT NULL,
content TEXT NOT NULL,
-- LLM metadata
tokens_used INTEGER,
model VARCHAR(100),
-- Additional metadata (temperature, latency, etc.)
metadata JSONB DEFAULT '{}',
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
);
CREATE INDEX idx_messages_conversation_id ON messages(conversation_id);
CREATE INDEX idx_messages_created_at ON messages(created_at);
CREATE INDEX idx_messages_conversation_created ON messages(conversation_id, created_at);
5. character_knowledge
Multiple knowledge sources for characters. Each source is chunked and stored with embeddings in vector_memories.
CREATE TYPE import_source_type AS ENUM ('file', 'url', 'manual');
CREATE TYPE import_status AS ENUM ('pending', 'processing', 'completed', 'failed');
CREATE TABLE character_knowledge (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
character_id UUID NOT NULL REFERENCES characters(id) ON DELETE CASCADE,
-- Knowledge source info
name VARCHAR(255) NOT NULL, -- Display name
source_type import_source_type NOT NULL,
source_name VARCHAR(255) NOT NULL, -- Original filename or URL
mime_type VARCHAR(100),
file_size BIGINT,
-- Raw content (before chunking)
raw_content TEXT,
-- Processing status
status import_status DEFAULT 'pending',
processing_info JSONB, -- chunks count, errors, etc.
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
updated_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
);
CREATE INDEX idx_character_knowledge_character ON character_knowledge(character_id);
CREATE INDEX idx_character_knowledge_status ON character_knowledge(status);
6. vector_memories
Unified vector embeddings storage for:
- Character knowledge - Background info, imported documents (linked to
character_knowledge) - Conversation history - Chat context (linked to
conversations)
CREATE TYPE memory_type AS ENUM ('conversation', 'character');
CREATE TABLE vector_memories (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
-- The text chunk
content TEXT NOT NULL,
-- Vector embedding (configurable dimension based on model)
-- Common sizes: 384 (all-MiniLM-L6-v2), 768 (all-mpnet-base-v2), 1024 (BGE)
-- Must match the EMBEDDING_DIMENSION env var
embedding VECTOR({{EMBEDDING_DIMENSION}}),
-- Memory type determines the context
memory_type memory_type DEFAULT 'conversation',
-- Metadata (chunk_index, source_info, etc.)
metadata JSONB,
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
-- Polymorphic relations (at least one must be set)
-- For conversation context
conversation_id UUID REFERENCES conversations(id) ON DELETE CASCADE,
-- For character knowledge
character_id UUID REFERENCES characters(id) ON DELETE CASCADE,
knowledge_id UUID REFERENCES character_knowledge(id) ON DELETE CASCADE
);
-- HNSW index for efficient similarity search
-- Note: Index is created after table creation based on actual dimension
-- CREATE INDEX idx_vector_memories_embedding ON vector_memories
-- USING hnsw (embedding vector_cosine_ops);
CREATE INDEX idx_vector_memories_conversation ON vector_memories(conversation_id) WHERE conversation_id IS NOT NULL;
CREATE INDEX idx_vector_memories_character ON vector_memories(character_id) WHERE character_id IS NOT NULL;
CREATE INDEX idx_vector_memories_knowledge ON vector_memories(knowledge_id) WHERE knowledge_id IS NOT NULL;
CREATE INDEX idx_vector_memories_type ON vector_memories(memory_type);
7. import_documents
General-purpose imported documents (not linked to characters). For character knowledge, use character_knowledge.
CREATE TABLE import_documents (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
user_id UUID NOT NULL REFERENCES users(id) ON DELETE CASCADE,
source_type import_source_type NOT NULL, -- file, url, manual
source_name VARCHAR(255) NOT NULL, -- filename or URL
-- Mime type for files
mime_type VARCHAR(100),
-- File size in bytes
file_size BIGINT,
-- Raw content (preprocessed)
content TEXT,
-- Processing status
status import_status DEFAULT 'pending',
error_message TEXT,
-- Metadata (source info, extraction method, etc.)
metadata JSONB DEFAULT '{}',
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
updated_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
);
CREATE INDEX idx_import_documents_user_id ON import_documents(user_id);
CREATE INDEX idx_import_documents_status ON import_documents(status);
8. story_branches (Phase 2)
Tree structure for branching narratives.
CREATE TABLE story_branches (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
conversation_id UUID NOT NULL REFERENCES conversations(id) ON DELETE CASCADE,
-- Self-referential for tree structure
parent_id UUID REFERENCES story_branches(id) ON DELETE CASCADE,
-- Branch content
title VARCHAR(255),
content TEXT NOT NULL, -- The generated story content
-- User direction that led to this branch
user_direction TEXT,
-- Branch metadata
generation_params JSONB DEFAULT '{}',
-- Tree position
depth INTEGER DEFAULT 0,
branch_order INTEGER DEFAULT 0,
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW()
);
CREATE INDEX idx_story_branches_conversation ON story_branches(conversation_id);
CREATE INDEX idx_story_branches_parent ON story_branches(parent_id);
9. conversation_participants (Phase 3 - Multi-Character)
Supports multiple characters in a single conversation.
CREATE TABLE conversation_participants (
id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
conversation_id UUID NOT NULL REFERENCES conversations(id) ON DELETE CASCADE,
character_id UUID NOT NULL REFERENCES characters(id) ON DELETE CASCADE,
-- Participant settings
is_active BOOLEAN DEFAULT true,
auto_respond BOOLEAN DEFAULT true, -- Auto-generate responses
created_at TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
UNIQUE(conversation_id, character_id)
);
CREATE INDEX idx_participants_conversation ON conversation_participants(conversation_id);
Enums
-- User roles
CREATE TYPE user_role AS ENUM ('USER', 'ADMIN');
-- Message roles
CREATE TYPE message_role AS ENUM ('user', 'assistant', 'system');
-- Import/knowledge source types
CREATE TYPE import_source_type AS ENUM ('file', 'url', 'manual');
-- Processing status
CREATE TYPE import_status AS ENUM ('pending', 'processing', 'completed', 'failed');
-- Vector memory types
CREATE TYPE memory_type AS ENUM ('conversation', 'character');
Prisma Schema (Reference)
Prisma schema uses the multi-file schema feature. Models are organized in prisma/models/ folder and imported into schema.prisma.
Schema Structure
prisma/
├── schema.prisma # Main schema file with imports
├── seed.ts # Database seeding
└── models/
├── user.prisma # User model + UserRole enum
├── character.prisma # Character + CharacterKnowledge models
├── conversation.prisma # Conversation + ConversationParticipant
├── message.prisma # Message model + MessageRole enum
├── vectorMemory.prisma # VectorMemory + MemoryType enum
├── importDocument.prisma # ImportDocument model
└── storyBranch.prisma # StoryBranch model
Main Schema (schema.prisma)
// schema.prisma
generator client {
provider = "prisma-client-js"
}
datasource db {
provider = "postgresql"
url = env("DATABASE_URL")
}
// Import all models from the models folder
import { User } from "./models/user"
import { Character, CharacterKnowledge } from "./models/character"
import { Conversation, ConversationParticipant } from "./models/conversation"
import { Message } from "./models/message"
import { VectorMemory } from "./models/vectorMemory"
import { ImportDocument } from "./models/importDocument"
import { StoryBranch } from "./models/storyBranch"
Full Schema Definition
// models/user.prisma
generator client {
provider = "prisma-client-js"
}
datasource db {
provider = "postgresql"
url = env("DATABASE_URL")
}
// Enums
enum UserRole {
USER
ADMIN
}
enum MessageRole {
user
assistant
system
}
enum ImportSourceType {
file
url
manual
}
enum ImportStatus {
pending
processing
completed
failed
}
enum MemoryType {
conversation
character
}
// Models
model User {
id String @id @default(uuid())
email String @unique
username String @unique
passwordHash String?
keycloakSub String? @unique
role UserRole @default(USER)
isActive Boolean @default(true)
createdAt DateTime @default(now())
updatedAt DateTime @updatedAt
characters Character[]
conversations Conversation[]
importDocs ImportDocument[]
@@index([email])
@@index([keycloakSub])
}
model Character {
id String @id @default(uuid())
name String
avatarUrl String?
personalityPrompt String
attributes Json @default("{}")
config Json @default("{}")
isPublic Boolean @default(false)
createdAt DateTime @default(now())
updatedAt DateTime @updatedAt
userId String
user User @relation(fields: [userId], references: [id], onDelete: Cascade)
conversations Conversation[]
knowledgeSources CharacterKnowledge[]
vectorMemories VectorMemory[]
@@index([userId])
@@index([name])
}
model CharacterKnowledge {
id String @id @default(uuid())
name String
sourceType ImportSourceType
sourceName String
mimeType String?
fileSize BigInt?
rawContent String?
status ImportStatus @default(pending)
processingInfo Json?
createdAt DateTime @default(now())
updatedAt DateTime @updatedAt
characterId String
character Character @relation(fields: [characterId], references: [id], onDelete: Cascade)
vectorMemories VectorMemory[]
@@index([characterId])
@@index([status])
}
model Conversation {
id String @id @default(uuid())
title String?
messageCount Int @default(0)
totalTokens Int @default(0)
settings Json @default("{}")
createdAt DateTime @default(now())
updatedAt DateTime @updatedAt
userId String
user User @relation(fields: [userId], references: [id], onDelete: Cascade)
characterId String
character Character @relation(fields: [characterId], references: [id], onDelete: Cascade)
messages Message[]
vectorMemories VectorMemory[]
storyBranches StoryBranch[]
participants ConversationParticipant[]
@@index([userId])
@@index([characterId])
@@index([createdAt])
}
model Message {
id String @id @default(uuid())
role MessageRole
content String
tokensUsed Int?
model String?
metadata Json?
createdAt DateTime @default(now())
conversationId String
conversation Conversation @relation(fields: [conversationId], references: [id], onDelete: Cascade)
@@index([conversationId])
@@index([createdAt])
@@index([conversationId, createdAt])
}
model VectorMemory {
id String @id @default(uuid())
content String
embedding Unsupported("vector")?
memoryType MemoryType @default(conversation)
metadata Json?
createdAt DateTime @default(now())
conversationId String?
conversation Conversation? @relation(fields: [conversationId], references: [id], onDelete: Cascade)
characterId String?
character Character? @relation(fields: [characterId], references: [id], onDelete: Cascade)
knowledgeId String?
knowledge CharacterKnowledge? @relation(fields: [knowledgeId], references: [id], onDelete: Cascade)
@@index([conversationId])
@@index([characterId])
@@index([knowledgeId])
@@index([memoryType])
}
model ImportDocument {
id String @id @default(uuid())
sourceType ImportSourceType
sourceName String
mimeType String?
fileSize BigInt?
content String?
status ImportStatus @default(pending)
errorMessage String?
metadata Json?
createdAt DateTime @default(now())
updatedAt DateTime @updatedAt
userId String
user User @relation(fields: [userId], references: [id], onDelete: Cascade)
@@index([userId])
@@index([status])
}
model StoryBranch {
id String @id @default(uuid())
title String?
content String
userDirection String
generationParams Json?
depth Int @default(0)
branchOrder Int @default(0)
createdAt DateTime @default(now())
conversationId String
conversation Conversation @relation(fields: [conversationId], references: [id], onDelete: Cascade)
parentId String?
parent StoryBranch? @relation("BranchTree", fields: [parentId], references: [id], onDelete: Cascade)
children StoryBranch[] @relation("BranchTree")
@@index([conversationId])
@@index([parentId])
}
model ConversationParticipant {
id String @id @default(uuid())
isActive Boolean @default(true)
autoRespond Boolean @default(true)
createdAt DateTime @default(now())
conversationId String
conversation Conversation @relation(fields: [conversationId], references: [id], onDelete: Cascade)
characterId String
@@unique([conversationId, characterId])
@@index([conversationId])
}
Prisma Client Usage Examples
// src/shared/prisma/prisma.service.ts
import { Injectable, OnModuleInit, OnModuleDestroy } from '@nestjs/common';
import { PrismaClient } from '@prisma/client';
@Injectable()
export class PrismaService extends PrismaClient implements OnModuleInit, OnModuleDestroy {
async onModuleInit() {
await this.$connect();
}
async onModuleDestroy() {
await this.$disconnect();
}
}
// Repository pattern with Prisma
@Injectable()
export class CharacterRepository {
constructor(private prisma: PrismaService) {}
async findByUser(userId: string) {
return this.prisma.character.findMany({
where: { userId },
orderBy: { updatedAt: 'desc' },
});
}
async create(data: CreateCharacterDto, userId: string) {
return this.prisma.character.create({
data: { ...data, userId },
});
}
}
Vector Memory Query with Prisma
// Similarity search using pgvector with Prisma
async similaritySearch(
targetId: string,
queryEmbedding: number[],
memoryType: MemoryType,
k: number = 5
) {
// Build the where clause based on memory type
const whereClause = memoryType === 'conversation'
? { conversationId: targetId, memoryType }
: { characterId: targetId, memoryType };
const results = await this.prisma.$queryRaw`
SELECT
id,
content,
metadata,
embedding <=> ${queryEmbedding}::vector as distance
FROM "VectorMemory"
WHERE ${whereClause}
ORDER BY embedding <=> ${queryEmbedding}::vector
LIMIT ${k}
`;
return results;
}
// Search character knowledge
async searchCharacterKnowledge(
characterId: string,
queryEmbedding: number[],
k: number = 5
) {
const results = await this.prisma.$queryRaw`
SELECT
vm.id,
vm.content,
vm.metadata,
ck.name as source_name,
1 - (vm.embedding <=> ${queryEmbedding}::vector) as similarity
FROM "VectorMemory" vm
JOIN "CharacterKnowledge" ck ON vm."knowledgeId" = ck.id
WHERE vm."characterId" = ${characterId}
AND vm."memoryType" = 'character'
ORDER BY similarity DESC
LIMIT ${k}
`;
return results;
}
Embedding Configuration
// Configuration for embedding providers
interface EmbeddingConfig {
provider: 'local' | 'huggingface-api';
model: string;
dimension: number;
// For local provider
localModelPath?: string;
// For HuggingFace API
apiKey?: string;
apiEndpoint?: string;
}
// Example configurations:
// Local: { provider: 'local', model: 'Xenova/all-MiniLM-L6-v2', dimension: 384 }
// HF API: { provider: 'huggingface-api', model: 'sentence-transformers/all-mpnet-base-v2', dimension: 768 }
Prisma Migration Strategy
Initial Migration
# 1. Initialize Prisma
npx prisma init
# 2. Define schema in prisma/schema.prisma
# 3. Create first migration
npx prisma migrate dev --name init
# 4. Generate Prisma Client
npx prisma generate
Migration Workflow
# After schema changes
npx prisma migrate dev --name descriptive_name
# Production deployment
npx prisma migrate deploy
# Generate client (in CI/CD)
npx prisma generate
Migration File Example
-- migrations/20240223120000_init/migration.sql
-- Enable extensions
CREATE EXTENSION IF NOT EXISTS "uuid-ossp";
CREATE EXTENSION IF NOT EXISTS "vector";
-- CreateEnum
CREATE TYPE "UserRole" AS ENUM ('USER', 'ADMIN');
-- CreateEnum
CREATE TYPE "MessageRole" AS ENUM ('user', 'assistant', 'system');
-- CreateTable
CREATE TABLE "User" (
"id" UUID NOT NULL DEFAULT uuid_generate_v4(),
"email" TEXT NOT NULL,
"username" TEXT NOT NULL,
-- ... etc
);
-- CreateIndex
CREATE UNIQUE INDEX "User_email_key" ON "User"("email");
-- Vector index (HNSW) - created manually after migration
CREATE INDEX idx_vector_memories_embedding ON "VectorMemory"
USING hnsw (embedding vector_cosine_ops);
Seeding
// prisma/seed.ts
import { PrismaClient } from '@prisma/client';
const prisma = new PrismaClient();
async function main() {
// Seed default admin or test data
await prisma.user.create({
data: {
email: 'admin@dreamchat.local',
username: 'admin',
role: 'ADMIN',
},
});
}
main()
.catch(console.error)
.finally(() => prisma.$disconnect());
# Run seed
npx prisma db seed
Backup Strategy
# pg_dump with custom format
docker exec dreamchat-postgres pg_dump -U postgres -Fc dreamchat > backup.dump
# Restore
pg_restore -U postgres -d dreamchat backup.dump