Torque vs Typesense

Torque serves the Typesense v30.1 API, so switching is a config change - and in our benchmarks it runs 9.5x faster. Typesense is the open-source original, with built-in high availability and a managed cloud.

Typesense is a popular open-source (GPL-3.0) search engine written in C++, with typo tolerance, faceting, vector search, built-in Raft high-availability clustering, and a managed Typesense Cloud. Torque is built to be API-compatible with it: it implements the Typesense v30.1 API - and because Typesense's current release, v30.2, is a bug-fix update on that same API, your existing schemas, queries, and client libraries keep working.

That makes this less an "alternative" and more a same-API choice. Torque is written in Rust, adds GPU-accelerated search and a built-in MCP server, and in our own benchmarks it runs 9.5x faster on queries and 7.8x on ingestion, on the same hardware. Typesense, in turn, is fully open source and offers built-in high-availability clustering, which Torque does not. The right pick depends on which of those matters most to you.

How they compare

TorqueTypesense
Foundations
LanguageRust, in-memoryC++, in-memory
Storage & persistenceIn-memory index, persisted as memory-mapped segment files (roaring bitmaps, columnar)In-memory index; raw documents stored on disk in RocksDB
Bulk ingestionBinary TCP streaming protocol and binary file upload, plus HTTP importHTTP JSONL import
LicenseCommercial (Truespar EULA)Open source (GPL-3.0)
DeploymentSelf-hostedSelf-hosted or Typesense Cloud
Client SDKsFive (Python, Node.js, Go, Java, .NET); Typesense clients also workOfficial Typesense clients
Search features
Typo toleranceYes (SymSpell)Yes (Levenshtein)
Faceting & filteringYes (roaring bitmaps)Yes
Vector & hybrid searchYes (RaBitQ quantization)Yes (float32)
Conversational & NL searchYes (any OpenAI-compatible LLM)Yes
GPU-accelerated searchYes (CUDA)No (Cloud offers GPU for embedding generation only)
Built-in MCP server for AI agentsYesNo
Where Typesense leads
Open sourceCommercial licenseGPL-3.0, fully open source
High-availability clusteringSingle-nodeBuilt in (Raft-based)
Managed cloudSelf-hosted onlyTypesense Cloud
Community & maturityFocused, newerLarge, established community

Based on Typesense's published documentation, license, and Typesense Cloud pricing as of July 2026. Torque figures reflect the current 0.6 release. Benchmark figures below are our own measurements - see the methodology on the performance page.

Performance

Because both engines implement the same API, we can benchmark them head to head on identical data and queries. On 309,000 real-world records (67 fields with nested objects, geo, and facets), measured over HTTP on the same hardware, CPU-only:

  • 9.5x faster on the geometric mean of 15 query patterns, winning every pattern.
  • 7.8x faster ingestion - 69,000 documents per second over Torque's binary TCP protocol, versus HTTP JSONL import into Typesense. Bulk loads over the binary file format are faster still.
  • Biggest gaps on sorting, grouping, and faceting; smallest on plain text search, where both are already fast.

See the full benchmarks and methodology.

A search store, not a database

Torque and Typesense sit in the same place in your stack. Both keep their index in memory for speed, and both are secondary data stores: you keep the source of truth in your primary database and sync a searchable copy into the engine. Neither is meant to hold the only copy of your data.

They share the same limits, too. Because the index lives in RAM, neither is a good fit for high-volume application logs, where keeping the full dataset in memory gets expensive and instant search-as-you-type is not the point. For that, a system built for logs - such as Elasticsearch - is the better tool.

Where each one fits

Choose Typesense when a fully open-source (GPL-3.0) engine matters, you want built-in high-availability clustering, or you would rather run on managed Typesense Cloud with a large, established community behind you.

Choose Torque when you want the same Typesense API with more query throughput on our benchmarks, faster bulk loads over the binary ingest protocol, GPU-accelerated search, and a built-in MCP server for AI agents - running on your own hardware, with commercial support.

Migrating from Typesense

This is the easy one. Because Torque implements the Typesense v30.1 API, moving is a configuration change: point your existing Typesense clients at Torque and your schemas, queries, and integrations work unchanged. Re-index your data over the HTTP import or the faster binary protocol, and you are running.