Skip to main content

​
Storage Types

Storage types handle the persistence of messages and metadata.

​
In-Memory

Default storage type when no storage is specified. 
# No configuration needed - this is the default
memory:
  id: my_memory
  vector:
    use: compozy/vector:pg-vector
    dimensions: 1536
Best Used For:
  • Development and testing
  • Quick prototyping
  • Temporary storage needs
  • Single-instance applications

​
PostgreSQL

Production-grade storage built on PostgreSQL. 
storage:
  use: compozy/storage:postgres
  config:
    url: "{{ env.POSTGRES_URL }}"
Best Used For:
  • Production deployments
  • High-concurrency workloads
  • Complex querying needs
  • Data persistence requirements

​
Redis

In-memory data store with optional persistence. 
storage:
  use: compozy/storage:redis
  config:
    url: "{{ env.REDIS_URL }}"
Best Used For:
  • High-performance needs
  • Caching requirements
  • Real-time applications
  • Temporary storage with persistence

​
Upstash

Serverless Redis-compatible storage. 
storage:
  use: compozy/storage:upstash
  config:
    url: "{{ env.UPSTASH_URL }}"
    token: "{{ env.UPSTASH_TOKEN }}"
Best Used For:
  • Serverless deployments
  • Edge computing
  • Pay-per-use scenarios
  • Low-latency requirements

​
SQLite

Lightweight file-based storage. 
storage:
  use: compozy/storage:sqlite
  config:
    path: ./data/memory.db
Best Used For:
  • Local development
  • Single-file deployments
  • Simple applications
  • Embedded systems

​
LibSQL

Distributed SQLite-compatible storage. 
storage:
  use: compozy/storage:libsql
  config:
    url: "{{ env.LIBSQL_URL }}"
Best Used For:
  • Edge deployments
  • Distributed systems
  • SQLite compatibility
  • Serverless applications

​
Vector Types

Vector types handle the storage and retrieval of embeddings.

​
PgVector

PostgreSQL-based vector storage. 
vector:
  use: compozy/vector:pg-vector
  dimensions: 1536
  config:
    url: "{{ env.POSTGRES_URL }}"
Best Used For:
  • Production deployments
  • Integrated PostgreSQL setups
  • Cost-effective vector storage
  • Simple vector search needs

​
Pinecone

Managed vector database service. 
vector:
  use: compozy/vector:pinecone
  dimensions: 1536
  config:
    apiKey: "{{ env.PINECONE_API_KEY }}"
Best Used For:
  • Large-scale deployments
  • High-performance vector search
  • Managed infrastructure
  • Production ML applications

​
Qdrant

Open-source vector database. 
vector:
  use: compozy/vector:qdrant
  dimensions: 1536
  config:
    url: "{{ env.QDRANT_URL }}"
Best Used For:
  • Complex filtering needs
  • Self-hosted deployments
  • High-performance search
  • Flexible deployment options

​
Chroma

Open-source embedding database. 
vector:
  use: compozy/vector:chroma
  dimensions: 1536
  config:
    path: ./data/embeddings
Best Used For:
  • Local development
  • Simple vector search
  • Quick prototyping
  • Small to medium datasets

​
Milvus

Distributed vector database. 
vector:
  use: compozy/vector:milvus
  dimensions: 1536
  config:
    url: "{{ env.MILVUS_URL }}"
Best Used For:
  • Large-scale deployments
  • Distributed search
  • High availability needs
  • Complex vector operations

​
Embedder Types

Embedder types handle the conversion of text to vector embeddings.

​
OpenAI

OpenAI’s text embedding models. 
embedder:
  use: compozy/embedder:openai
  config:
    model: text-embedding-3-small
Best Used For:
  • High-quality embeddings
  • Production use cases
  • Multi-lingual support
  • Latest embedding models

​
FastEmbed

Lightweight local embedding models. 
embedder:
  use: compozy/embedder:fastembed
  config:
    model: bge-small-en
Best Used For:
  • Local deployment
  • Privacy requirements
  • Cost-effective solutions
  • Quick prototyping

​
Cohere

Cohere’s embedding service. 
embedder:
  use: compozy/embedder:cohere
  config:
    model: embed-english-v3.0
Best Used For:
  • Enterprise use cases
  • Specialized embeddings
  • Multi-lingual needs
  • Alternative to OpenAI

​
HuggingFace

Open-source models from HuggingFace. 
embedder:
  use: compozy/embedder:huggingface
  config:
    model: sentence-transformers/all-mpnet-base-v2
Best Used For:
  • Custom models
  • Self-hosted solutions
  • Research applications
  • Specialized domains

​
Vertex AI

Google Cloud’s embedding service. 
embedder:
  use: compozy/embedder:vertex
  config:
    model: textembedding-gecko
    project: "{{ env.GOOGLE_CLOUD_PROJECT }}"
Best Used For:
  • Google Cloud integration
  • Enterprise deployments
  • Scalable solutions
  • Managed infrastructure

​
Best Practices

Choose the Right Storage

Select storage based on your persistence and performance needs:
  • Use in-memory for development
  • Use PostgreSQL for traditional deployments
  • Use Redis for high-performance needs
  • Use Upstash for serverless deployments

Vector Configuration

Choose vector storage based on scale and requirements:
  • Use PgVector for integrated PostgreSQL setups
  • Use Pinecone for managed vector search
  • Use Qdrant for self-hosted deployments
  • Use Chroma for local development

Embedder Selection

Select embedder based on your use case:
  • Use OpenAI for production quality
  • Use FastEmbed for local deployment
  • Use Cohere for enterprise needs
  • Use HuggingFace for custom models

Security

Implement security best practices:
  • Use environment variables for credentials
  • Enable SSL/TLS in production
  • Implement proper access controls