--- title: "Deploy Typesense Multilingual" description: "Locale-aware multilingual search with Typesense" category: "Analytics" url: https://railway.com/deploy/deploy-typesense-multilingual-search --- # Deploy Typesense Multilingual Locale-aware multilingual search with Typesense **[Deploy Typesense Multilingual on Railway](https://railway.com/template/deploy-typesense-multilingual-search)** Machine-readable deploy manifest (JSON, validated by TemplateCI): https://railway.com/deploy/deploy-typesense-multilingual-search/manifest.json - **Creator:** onepush - **Category:** Analytics ## Template content ### typesense-railway - **Source:** Shinyduo/typesense-railway - **Public domain:** Yes ## Documentation # Deploy and Host self hosted Typesense Multilingual (Open-Source Instant Search) on Railway One search box, several languages. This template runs Typesense 30.2 on Railway with a persistent volume, so you can give German, Japanese, and English fields their own locale and query them together. ## About Hosting Typesense Multilingual open-source software on Railway (self hosted Typesense template) Run a Japanese catalog through an English tokenizer and words that never had spaces get treated as one giant token, so partial matches just vanish. Typesense fixes that at the tokenizer level, not with a query-time bandage. Set a `locale` parameter on a schema field and Typesense tokenizes that field the way the language expects. On Railway, this is a single C++ server process with an in-memory index and a snapshot on disk. ## Why Deploy Typesense Multilingual, the Algolia alternative on Railway (Railway Free Trial) Algolia is good at instant search, but it's SaaS-only. No self-hosting, and every request and stored record shows up on the invoice. For a nine-language catalog, those costs multiply. The $5 Railway GitHub trial is the practical entry point. Deploy, load a multilingual sample, and test locale-aware search in a browser with CORS already enabled before spending more. Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it. By deploying Typesense Multilingual on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway. ### Railway vs Other Hosting Providers and VPS for Typesense Multilingual self hosting | Provider | Multilingual Typesense fit | Ops overhead | Cost signal | |---|---|---|---| | DigitalOcean | Droplet works, but you manage Docker, volumes, firewall, TLS | High | Predictable monthly droplet price | | AWS | EC2 or ECS works, but IAM, security groups, EBS add complexity | Very high | Hard to forecast without tagging | | Hetzner | Cheap raw compute, zero managed platform features | High | Very low for hardware | | Railway | Volume, env vars, health checks first-class | Low | Compute + volume, no per-request fee | ## Common Use Cases for hosted Typesense Multilingual A marketplace selling one product in five countries needs one search box that understands five languages. Documentation search is another. Open-source projects often have docs in English plus one or two other languages. Locale-aware fields mean a search for "Fehler" returns the German error page, not an unrelated English page. E-commerce with Chinese, Japanese, or Korean is where typo tolerance gets tricky. CJK scripts don't tolerate fuzzy matching like English. ## Dependencies for Typesense Multilingual Docker hosted on Railway Just one service: the Typesense container with a volume at `/data` and a `TYPESENSE_API_KEY` variable. No database, queue, or cache sits next to it. ### Deployment Dependencies for Managed Typesense Multilingual Service (Instant Search) Pin the image to `typesense/typesense:30.2`, not `latest`, so an upgrade never surprises a live index. You need a Railway volume at `/data` so the index survives restarts. ### Implementation Details for Typesense Multilingual (Using Typesense official docker image) Start with `--data-dir /data --api-key=$TYPESENSE_API_KEY --enable-cors`. The CORS flag matters: browser-based InstantSearch clients need it or every request fails. A gotcha: the API key authenticates requests but doesn't encrypt volume data. If you lose it, the index on `/data` is still intact. Set a new `TYPESENSE_API_KEY` and redeploy - Typesense serves existing data with the new key. No reindex needed. Locale config happens in the collection schema, not the Docker command. Each field can carry a `locale` parameter. Documented locales include `th`, `ja`, `zh`, `ko`, `ru`, `el`, `ar`, and `he`. A field without a locale uses the English-style tokenizer. ## How does Typesense Multilingual compare against other Instant Search platforms ### Typesense Multilingual vs Algolia (Algolia Alternative) Algolia's multilingual story is solid, but pricing is the issue. It bills on search requests and records stored, and nine language fields per record raise both. Typesense keeps the same RAM-based architecture whether you have one locale or twelve. ### Typesense Multilingual vs Elasticsearch (Elasticsearch Alternative) Elasticsearch has language analyzers for dozens of languages, but you're managing JVM heap, shard counts, and index settings before tuning search. Typesense gives locale-aware tokenization with a schema field. ### Typesense Multilingual vs Meilisearch (Meilisearch Alternative) Meilisearch is close in spirit: small, fast, open source, clean API. It supports multiple languages, but locale handling is more global than per-field. ### Typesense Multilingual vs Pinecone (Pinecone Alternative) Pinecone is a vector database, not an instant search engine. Use it with multilingual embeddings for semantic similarity across languages. Use Typesense for keyword search with typo tolerance, locale-aware tokenization, and instant dropdowns. ## How to use Typesense Multilingual (the OSS Instant Search)? Decide your collection layout first. Single collection with one field per language - `title_en`, `title_de`, `title_ja` - queried together via `query_by` is simpler for a unified search box. Set `locale` on each field in your schema. In a single collection, `title_en` gets no locale, `title_de` gets `locale: de`, `title_ja` gets `locale: ja`. Tokenizer behavior varies by locale, so test with your own data. Tune tokenization with `token_separators` and `symbols_to_index`. A German hyphenated compound and a French apostrophe need different handling. For CJK fields, set `num_typos` to 0 or 1 instead of the default 2. Generous typo tolerance produces noisy results for Japanese and Chinese. Add locale-aware synonyms and stopwords. Typesense lets you attach a locale to synonym and stopword sets so a German "Handy" (mobile phone) doesn't affect English queries for "handy." ## How to self host Typesense Multilingual on other VPS Services (Typesense Multilingual self hosting guide) ### Clone the Repository Typesense is distributed as a Docker image, not a source repo you need to clone. Pull `typesense/typesense:30.2` from Docker Hub. The full GPL-3.0 source is on GitHub, but the image is the canonical deployment path. ### Install Dependencies Install Docker and Docker Compose. That's the only dependency beyond the OS. Typesense doesn't need a database, message queue, or separate cache. The mounted `/data` directory holds the on-disk snapshot; the in-memory index serves queries. ### Configure Environment Variables Set `TYPESENSE_API_KEY` to a long random string. It's the only required variable. Store it in a password manager or a secrets file, not a public repo. If you lose it, the data on `/data` is still there; set a new key and restart. No reindexing. ### Start the Typesense Multilingual Application Run the container with `--data-dir /data --api-key=$TYPESENSE_API_KEY --enable-cors`. Map port 8108 to the host and point a health check at `/health`. Then create your collection schema with per-field locale settings and index. ## Official Pricing of Typesense Multilingual (Typesense Multilingual pricing) Typesense Multilingual is GPL-3.0 open source, so self-hosting the software is free. You pay for infrastructure. On Railway, a small single-node deployment typically lands in single-digit to low-teens USD per month with no per-request or per-record fee. Typesense Cloud bills dedicated RAM and vCPU by the hour plus bandwidth, with no per-search fee. Check the Typesense Cloud pricing page for current numbers. ## Typesense Multilingual cloud vs self hosted comparison (Pricing, features, costs, and more) ### Monthly cost of self hosting Typesense Multilingual on Railway On Railway, a small Typesense node runs on modest compute with a volume for `/data`. Total cost depends on RAM, since the index lives in memory. ### System Requirements for Hosting Typesense Multilingual on a VPS RAM is the hard requirement. Typesense holds the entire index in memory. Allocate enough RAM for dataset size plus 20-30% for indexing buffers and query overhead. A 500 MB dataset fits in 1 GB; a 5 GB dataset needs at least 6-7 GB. CPU matters less. ## Frequently Asked Questions (FAQs) ### What happens if I lose my TYPESENSE_API_KEY? Don't panic. The key authenticates API requests but doesn't encrypt the volume data. Set a new `TYPESENSE_API_KEY` in Railway, redeploy, and Typesense serves the existing index with the new key. No reindexing. ### Can I mix languages in one collection or use separate collections? Both work. One collection with `title_en`, `title_de`, `title_ja` and a combined `query_by` is simpler and keeps one search box. Separate collections help when ranking signals differ or you need to reindex one language without downtime for others. ### Why do my Japanese queries return irrelevant results? Almost certainly typo tolerance. The default `num_typos` is tuned for English, where small character changes often don't change meaning. Japanese doesn't work that way. Set `num_typos` to 0 or 1 on CJK fields in your schema and test with real user queries. ### Do I need to configure anything special for CORS? Yes, if you use browser-based InstantSearch clients. The `--enable-cors` flag lets browsers make requests to your node. Without it, curl and server-side code work but every browser request fails. ### How do I handle diacritics and accents across languages? Test with your own data. Tokenizer behavior varies by locale and the specific characters in your dataset. Some locales fold accents more aggressively than others. ## Similar templates - [Formbricks | Open-Source Typeform & Qualtrics Alternative](https://railway.com/deploy/formbricks-surveys) — Surveys, NPS and in-app feedback on your own Postgres, Hub and Cube - [Typesense PHP](https://railway.com/deploy/typesense-php) — official PHP client against Railway - [Typesense vs Meilisearch](https://railway.com/deploy/typesense-vs-meilisearch) — self-hosted Typesense vs Meilisearch Open this page in a browser: https://railway.com/deploy/deploy-typesense-multilingual-search