Replacing Algolia & Elastic Cloud: Self-Hosted Search with Meilisearch, Typesense & OpenSearch (Benchmarks & Docker Compose)

Replacing Algolia & Elastic Cloud: Self-Hosted Search with Meilisearch, Typesense & OpenSearch

Search is a core component of modern web applications, SaaS products, and e-commerce stores. However, managed search services like Algolia charge aggressive metered pricing ($0.50+ per 1,000 search requests and steep index record limits). On the other end, managed Elastic Cloud clusters typically start at $95/month and consume gigabytes of JVM memory before indexing a single document.

Fortunately, modern self-hosted search engines offer sub-50ms search-as-you-type performance, typo tolerance, multi-faceted filtering, and vector hybrid search—all deployable on an inexpensive $6 to $12/month VPS.

In this guide, we compare Meilisearch, Typesense, and OpenSearch, and provide production-ready Docker Compose configurations.

---

Engine Comparison Matrix

Feature	Meilisearch	Typesense	OpenSearch / Elastic
Primary Architecture	Rust (LMDB memory-mapped)	C++ (In-memory + RocksDB)	Java / Lucene (Distributed cluster)
Search Latency (<100k docs)	< 10ms (Blazing)	< 10ms (Blazing)	25-50ms
RAM Consumption	Moderate (Memory-mapped cache)	High (Full dataset kept in RAM)	Very High (JVM heap 2GB - 32GB)
Typo Tolerance & Ranking	⭐⭐⭐⭐⭐ (Out-of-the-box perfect)	⭐⭐⭐⭐⭐ (Configurable typo rules)	⭐⭐⭐ (Requires custom tokenizers)
Vector & Hybrid Search	Native (Embedders & Auto-generation)	Native (Built-in ONNX / OpenAI models)	Native (k-NN plugin / neural search)
Multi-Node Clustering	Cloud-native / Enterprise	Raft Consensus (Built-in High Availability)	Native Sharding & Distributed Cluster
Best Used For	E-commerce, Docs, Apps (<5M docs)	Fast SaaS Search, Geo-search, High QPS	Big Data Logs, SIEM, massive datasets (>10M docs)

---

Architectural Deep Dive

1. Meilisearch — Developer Experience & Instant Relevance

Meilisearch is written in Rust and is hyper-focused on instant search-as-you-type experiences.

• Why Choose Meilisearch: Default ranking rules (words, typo, proximity, attribute, exactness) work flawlessly without manual NLP tuning.
• Vector Search Support: Supports hybrid search combining BM25 keyword matching with dense vector embeddings directly from HuggingFace, OpenAI, or local Ollama endpoints.
• Resource Footprint: Uses LMDB, so indexes reside on disk and cache heavily in OS RAM.

2. Typesense — C++ In-Memory Speed & Native Clustering

Typesense is built in C++ specifically as an open-source alternative to Algolia.

• Why Choose Typesense: Holds the full index in RAM with RocksDB disk persistence, offering ultra-low search latency even under thousands of concurrent queries per second (QPS).
• Clustering: Includes native 3-node or 5-node Raft consensus clustering out-of-the-box without needing external orchestrators.

3. OpenSearch — Distributed Big Data Scale

OpenSearch (the Apache-2.0 fork of Elasticsearch maintained by AWS and the Linux Foundation) remains the industry standard for distributed text search across millions of large documents and log archives.

• Why Choose OpenSearch: Massive scalability, complex aggregations, distributed sharding, and deep plugin ecosystems.
• Caveat: High operational overhead. Requires careful JVM heap tuning and at least 4GB to 8GB of RAM.

---

Production Docker Compose: Meilisearch with Persistent Storage & Master Key Security

Here is a hardened docker-compose.yml for running Meilisearch in production behind Traefik or Caddy:

version: '3.8'

services:
  meilisearch:
    image: getmeili/meilisearch:v1.12
    restart: unless-stopped
    environment:
      - MEILI_ENV=production
      - MEILI_MASTER_KEY=${MEILI_MASTER_KEY:?Required}
      - MEILI_NO_ANALYTICS=true
      - MEILI_MAX_INDEXING_MEMORY=2Gb
      - MEILI_MAX_INDEXING_THREADS=2
      - MEILI_HTTP_PAYLOAD_SIZE_LIMIT=104857600 # 100MB batch upload cap
    volumes:
      - meili_data:/meili_data
    ports:
      - "127.0.0.1:7700:7700"
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:7700/health"]
      interval: 10s
      timeout: 5s
      retries: 5
    networks:
      - search_net

volumes:
  meili_data:

networks:
  search_net:
    driver: bridge

---

Typesense Production Alternative

For teams requiring native in-memory caching and Raft clustering, here is the equivalent Typesense configuration:

version: '3.8'

services:
  typesense:
    image: typesense/typesense:27.1
    restart: unless-stopped
    environment:
      - TYPESENSE_API_KEY=${TYPESENSE_API_KEY:?Required}
      - TYPESENSE_DATA_DIR=/data
      - TYPESENSE_ENABLE_CORS=true
    volumes:
      - typesense_data:/data
    ports:
      - "127.0.0.1:8108:8108"
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8108/health"]
      interval: 10s
      timeout: 5s
      retries: 5
    networks:
      - search_net

volumes:
  typesense_data:

networks:
  search_net:
    driver: bridge

---

Production Security & Deployment Tips

1. Never Expose the Master Key to Frontend Clients: Generate scoped Search-Only API keys with limited index access and expiration timestamps using the engine's Key Management API.
2. Backups & Snapshots: Meilisearch and Typesense support scheduled automatic snapshot exports. Set up a simple cron container that dumps snapshots to an S3-compatible bucket (e.g. MinIO, Cloudflare R2, or Garage).
3. RAM Sizing: For Typesense, ensure your VPS has at least 1.5x the RAM of your total uncompressed dataset. For Meilisearch, ensure fast NVMe storage for optimal memory-mapped file paging.

---

Which One Should You Choose?

• Choose Meilisearch for fast SaaS search, docs, mobile apps, and instant typo-tolerant search bars with minimal configuration.
• Choose Typesense for high-throughput e-commerce catalogs requiring sub-millisecond in-memory response times and built-in Raft clustering.
• Choose OpenSearch for complex distributed enterprise analytics, log indexing, and multi-terabyte search clusters.

Check out more comparisons, migration benchmarks, and production-tested architectures at SelfHostStack.