The retrieval control plane

Every retrieval job,
its best engine.

Your app doesn't have one search box — it has many retrieval jobs: public search, faceted filtering, autocomplete, RAG for on-site chat, and tool-calls for your agents. Headband routes each one to the engine that wins it, under a single API and account.

Typesense for instant search, Elasticsearch for faceted filtering and analytics, Meilisearch for typo-tolerant autocomplete, vector-tuned indexes for RAG — mixed freely across all three. Bring your own engine and your documents stay in your cluster, encrypted.

Get started freeRead the docs
One engine per retrieval job
# Public product search → Meilisearch (typo-tolerant, instant)
curl -X POST https://headband.dev/v1/indexes \
  -H "Authorization: Bearer hb_adm_..." \
  -d '{ "uid": "products", "engine": "meilisearch" }'

# Faceted filtering + analytics → Elasticsearch
curl -X POST https://headband.dev/v1/indexes \
  -H "Authorization: Bearer hb_adm_..." \
  -d '{ "uid": "catalog-facets", "engine": "elasticsearch" }'

# Same API key. Same SDK. Each job, its best engine.
Search
curl -X POST https://headband.dev/v1/search \
  -H "Authorization: Bearer hb_src_..." \
  -H "Content-Type: application/json" \
  -d '{ "index": "products", "q": "macbook" }'

// Engine-reported processing time — routed automatically
{
  "hits": [
    { "id": 1, "title": "MacBook Pro", "price": 1999 }
  ],
  "engine": "meilisearch",
  "processingTimeMs": 2
}

React SDK

Build search UIs in minutes with @headband/react. Hooks, components, and InstantSearch-inspired API.

@headband/react
import { headband, HeadbandProvider, SearchBox, Hits, RefinementList } from '@headband/react'

const client = headband('https://headband.dev', 'hb_src_your_key')

function App() {
  return (
    <HeadbandProvider client={client} index="movies">
      <SearchBox placeholder="Search movies..." />
      <RefinementList attribute="genres" />
      <Hits />
    </HeadbandProvider>
  )
}

Integrations

Source → sync → route → experiment → ship.

Use Headband with the app framework, database, commerce stack, and analytics loop you already trust. Live SDK support, recipe-grade connector patterns, and roadmap runtime features are labeled separately.

Explore integrations

Frameworks

5 recipes

Example: Next.js + React

Sync connectors

6 patterns

Example: Postgres / Neon / Supabase

A/B/C tests

3 playbooks

Example: Engine A/B/C test

One engine per retrieval job

Public search on Typesense, faceted filtering on Elasticsearch, autocomplete on Meilisearch, RAG and agent tool-calls on vector-tuned indexes — every workload on the engine that wins it, all under one account. Nobody else lets you mix.

Bring your own engine

Point indexes at your own Meilisearch, Typesense, or Elasticsearch cluster when you need data residency. Headband keeps routing metadata, keys, usage, and analytics in the control plane; BYOE indexes live in your engine.

Built for AI agents

llms.txt, RAG-ready endpoints, and an MCP server. Headband is the retrieval substrate for your agents.

Three engines, one API

Meilisearch, Typesense, Elasticsearch. Same SDK, same dashboard, same filter syntax. Swap engines without touching client code.

Scoped API keys

Admin keys for writes, search keys for reads. Project-scoped, prefix-enforced, tenant-isolated by default.

Analytics included

Top queries, zero-result rates, response times. Per-engine, per-index, in the dashboard. No extra service to wire up.

Retrieval for agents

Headband is where agents get their context.

llms.txt and RAG-ready endpoints ship today. MCP server and /v1/chat are live. Same API, any engine, any agent.

llms.txt generation

Auto-generate a standards-compliant llms.txt so AI crawlers know what your indexes contain.

RAG-ready search

Clean text chunks, no HTML, structured metadata. One call feeds your RAG pipeline.

MCP server + /v1/chat

Authenticated HTTP JSON-RPC MCP and a chat completions endpoint grounded in your indexes.

See the engines compete in real time

Try the Shootout

Pricing

Simple, predictable pricing.

Free to start. Scale when you need to. Bring your own engine on any plan — your documents stay in your cluster.

Free

$0forever

For side projects and experiments.

  • 1,000 searches/month
  • 10,000 documents
  • 1 project
  • Community support
  • All 3 engines
Get Started Free
Most popular

Pro

$29/month

For teams shipping real products.

  • 100,000 searches/month
  • 1,000,000 documents
  • 10 projects
  • All engines + BYOE
  • Email support
  • Search analytics
Start Pro Trial

Scale

$99/month

For organizations with serious retrieval needs.

  • Unlimited searches
  • Unlimited documents
  • Unlimited projects
  • All engines + BYOE
  • Priority support
  • Advanced analytics
  • MCP over authenticated HTTP JSON-RPC
Contact Sales

All plans include all 3 engines and Bring Your Own Engine. See full pricing details

Start searching in 5 minutes.

Free to start. No credit card required.

Create your account