Skip to main content
Follow these guidelines to get the most out of Meilisearch, whether you’re indexing a few hundred documents or scaling to millions.

​
Document formatting

​
Use a simple, consistent structure

Meilisearch can handle complex nested JSON, so you don’t need to flatten everything. However, simpler structures are easier to configure and search through. The key rule is: never use dynamic field names. Every document in an index should share the same field names. If field names change from one document to the next, Meilisearch cannot index them properly. Good: consistent field names, nested objects are fine: 
{
  "id": 1,
  "title": "Running shoes",
  "brand": "Nike",
  "price": 120,
  "details": {
    "color": "black",
    "size": 42
  }
}
Bad: dynamic field names (keys change per document): 
[
  {
    "id": 1,
    "Nike": { "price": 120 }
  },
  {
    "id": 2,
    "Adidas": { "price": 95 }
  }
]

​
Use meaningful primary keys

Choose a primary key that uniquely identifies each document and is stable over time. If your documents come from a database, use the database’s primary key. If Meilisearch doesn’t find a primary key candidate, it will ask you to set one explicitly.

​
Remove unnecessary fields

If a field is not searchable, filterable, sortable, or displayed, remove it from your documents before indexing. Smaller documents are indexed faster and consume less disk and memory.

​
Chunking content

When indexing long-form content (articles, documentation pages, books), break it into smaller, meaningful chunks rather than indexing entire pages as single documents.

​
Why chunk

  • Better relevancy: a search for “installation guide” is more likely to match a specific section than an entire 5,000-word page.
  • Better displayed results: users see targeted excerpts instead of massive blocks of text.
  • Faster indexing: smaller documents are processed more quickly.

​
How to chunk

Split content at logical boundaries like headings, sections, or paragraphs. Each chunk should be self-contained enough to make sense on its own. 
[
  {
    "id": "getting-started-install",
    "page": "Getting started",
    "section": "Installation",
    "content": "Install Meilisearch using the following command...",
    "url": "/docs/getting-started#installation"
  },
  {
    "id": "getting-started-config",
    "page": "Getting started",
    "section": "Configuration",
    "content": "Configure your instance by setting environment variables...",
    "url": "/docs/getting-started#configuration"
  }
]
Include metadata fields (page title, section, URL) so users can navigate back to the original content.

​
Batching requests

​
Send documents in large batches

A single large HTTP payload is processed more quickly than multiple smaller payloads. For example, indexing 100,000 documents in two batches of 50,000 is faster than four batches of 25,000. By default, Meilisearch accepts payloads up to 100MB. Enterprise accounts can request a higher limit.

​
Use compression

Compress your HTTP payloads using gzip, deflate, or br encoding to reduce transfer time and bandwidth. Meilisearch accepts all standard encoding formats.

​
Monitor task completion during large imports

When sending multiple batches, you don’t need to wait for each task to complete before sending the next one. Meilisearch queues all tasks and processes them in order. However, monitoring task status helps you detect errors early.

​
Indexing performance

​
Configure settings before adding documents

Always configure your index settings (ranking rules, filterable attributes, searchable attributes) before adding documents. Changing settings after indexing triggers a full reindex of all documents. 
# 1. Create the index and configure settings
curl -X PATCH 'MEILISEARCH_URL/indexes/movies/settings' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer API_KEY' \
  --data-binary '{
    "searchableAttributes": ["title", "overview"],
    "filterableAttributes": ["genre", "release_year"],
    "sortableAttributes": ["release_year", "rating"]
  }'

# 2. Then add documents
curl -X POST 'MEILISEARCH_URL/indexes/movies/documents' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer API_KEY' \
  --data-binary @movies.json

​
Restrict searchable attributes

By default, all fields are searchable. Limit searchable attributes to only the fields users should be able to search through. This improves both relevancy and indexing speed.

​
Use separate indexes for separate languages

If your dataset contains content in multiple languages, create a separate index for each language. Meilisearch’s tokenizer, stop words, and ranking algorithms are language-specific, so mixing languages in a single index degrades relevancy for all of them.

​
Avoid creating too many indexes

Each index consumes resources. If you notice performance degradation with multi-index searches, consider consolidating indexes where possible.

​
Enable binary quantization for large AI search datasets

If you use AI-powered search with more than 1 million documents and high-dimensional embeddings (1400+ dimensions), consider enabling binary quantization. This reduces semantic search precision slightly but greatly improves indexing and search performance.
Binary quantization is irreversible. Once enabled, the only way to recover original vector precision is to re-vectorize the entire index.

​
Keep Meilisearch up to date

Each release includes indexing and search performance improvements. Check the changelog and GitHub releases regularly.

​
Next steps

Configure index settings

Learn how to set up searchable, filterable, and sortable attributes

Relevancy and ranking

Understand how Meilisearch ranks search results

Working with tasks

Monitor indexing progress with the task API

Security

Secure your Meilisearch instance with API keys