Migration Guide
Migrate from any search provider to Skirre. Whether you're coming from Algolia, Meilisearch, Typesense, Elasticsearch, or a platform's built-in search, the process is straightforward.
Overview
Skirre is designed to be easy to migrate to. The API follows familiar patterns from other search engines: you create an index, push documents, configure settings, and search. Most concepts from other providers map directly to Skirre equivalents.
AI-Assisted Migration
The fastest way to migrate is with an AI assistant. Skirre provides a comprehensive API reference document designed specifically for AI consumption. Feed it to your AI assistant alongside your migration request, and it can generate the correct migration code for your specific setup.
- In the SDK package: After installing
@skirre/sdk, the file is atnode_modules/@skirre/sdk/docs/migration.md - Via URL: Fetch
https://www.skirre.com/llms.txt— this is a machine-readable reference following thellms.txtconvention
Example prompt for your AI assistant:
I'm migrating from Algolia to Skirre. Here's my current setup: - 3 indices: products, blog_posts, help_articles - Using instantsearch.js on the frontend - Filters on category, price, brand - Synonyms configured for product terms Please read the Skirre migration reference at node_modules/@skirre/sdk/docs/migration.md and generate the migration script.
Your AI already knows the source platform. It only needs accurate knowledge of Skirre's API surface, which the reference document provides in full.
Concept Mapping
Core concepts map across all search providers:
| Concept | Skirre | Notes |
|---|---|---|
| Index / Collection | skirre.index('uid') | Created with field configuration |
| Documents | index.addDocuments() | Schemaless JSON, upsert by primary key |
| Search | index.search(query, params) | Filter string, facets, sort, highlights |
| Searchable fields | searchable_attributes | With per-field weights and language |
| Facet fields | filterable_attributes | With field type (string/number/boolean) |
| Sort fields | sortable_attributes | Array of field names |
| Synonyms | settings.synonyms | One-directional key → values map |
| Scoped keys | Tenant tokens | HMAC-signed with per-index filter restrictions |
| Async writes | Task system | All writes return TaskInfo, poll for completion |
Step-by-Step Process
- Get an API key — create an admin key from the dashboard or via API
- Map your schema — identify which fields are searchable, filterable, and sortable in your current provider and translate them to Skirre's field configuration
- Create the index — with your field config and primary key
- Configure settings — synonyms, typo tolerance, semantic search, language
- Transform documents — map your source format to flat JSON objects with the configured field names
- Batch upload — chunk documents into batches of 500–1000, use
addDocumentsAndWait() - Verify — test searches, check facets, confirm document count
- Switch search calls — update your frontend/backend to use the Skirre SDK
- Add event tracking — integrate click, purchase, and page view events
- Set up aliases — for zero-downtime reindexing in the future
Platform-Specific Notes
From Algolia
- Filter syntax differs: Algolia uses
facetFiltersarrays, Skirre uses a string expression (category = "Electronics" AND price > 100) - Pagination is offset-based, not page-based. Convert:
offset = page * hitsPerPage - All writes are async — use
*AndWait()methods for migration scripts - Replicas → create separate indices with different sort configurations + aliases
From Meilisearch
- Very similar API shape — task system, filter syntax, and SDK patterns are close
- Field config is set at index creation time (or via settings update) rather than separate per-setting API calls
- Searchable field weights replace Meilisearch's ordered list priority
From Typesense
- Typesense requires a strict schema; Skirre configures which fields to index but documents are schemaless JSON
- Filter syntax differs: Typesense uses
field:value, Skirre usesfield = "value"
From Elasticsearch / OpenSearch
- No Query DSL — Skirre uses simple string filters, not nested JSON query objects
- Facets are count-based only (no aggregation pipeline)
- Aliases work the same conceptually — zero-downtime index swaps
node_modules/@skirre/sdk/docs/migration.md.