Site search cost: what you actually pay for each option
The engine license is free for the core editions of Meilisearch, Typesense and OpenSearch, so the total cost of self-hosted site search is set by infrastructure, implementation, operations and staff time, and the final figure depends on catalog size, tenancy, data location and uptime needs. A team weighing a build, a partner or a SaaS subscription should first price each line from its own environment rather than from the engine's sticker price. Vendors publish SaaS prices, so those can be compared directly: Meilisearch Cloud starts at $20 per month with a 14-day free trial and usage-based or resource-based billing, and Typesense Cloud offers a free cluster to start, with bandwidth billed separately. Self-hosted costs have no equivalent published figure, so they must be built from your own hosting, staffing and support assumptions.
Start with the lines that recur every month. On a self-hosted build, you pay for compute and memory for the search nodes, storage and backups, and the people who run updates, monitoring and scaling. Typesense's system requirements guide states that keyword search typically needs 2 to 3 times the size of the searchable fields in RAM, and strongly recommends a highly available 3-node or 5-node cluster for production, so a multi-node cluster multiplies the memory line rather than adding a small increment. Vector search adds 7 bytes of memory per dimension per vector, and built-in ML models need roughly 2 to 6 GB of extra RAM, so semantic features are a cost driver to price separately. Confirm the record count, the size of the searchable fields and the expected query volume with your engineering team before asking any provider for a number.
Cost components across build, partner and SaaS
This table compares who pays for each cost line when self-hosting site search in-house, with an implementation partner, or on a SaaS search service. Read the one-time rows as project spend and the recurring rows as the cost that continues after go-live. The main takeaway is that the engine license is rarely the largest line, and infrastructure, indexing work and staff time decide the total.
| Cost component | In-house build | Implementation partner | SaaS search |
|---|---|---|---|
| Engine license | Meilisearch Community, Typesense server (GPL-3.0) and OpenSearch (Apache 2.0) carry no license fee. One-time: evaluation. Recurring: none for the engine itself. | Same open-source license terms apply. Partner fees cover the work, not the engine. Confirm that the scope excludes any vendor enterprise edition. | Recurring subscription or usage fee set by the vendor. Meilisearch Cloud and Typesense Cloud publish starting prices and usage-based billing on their pricing pages; confirm current terms before budgeting. |
| Infrastructure (compute, memory, storage, bandwidth) | Recurring. Typesense keyword search typically needs 2 to 3 times the size of searchable fields in RAM, and each vector needs 7 bytes per dimension. Built-in ML models need roughly 2 to 6 GB of extra RAM. Production usually means a 3-node or 5-node cluster. | Recurring, billed to the client's cloud account or data center. The partner sizes it in the written scope, so the client sees the same memory and node drivers. | Recurring, included in or billed on top of the subscription. Typesense Cloud bills bandwidth separately. Ask how replicas and vector memory affect the plan tier. |
| Initial configuration and indexing pipeline | One-time. Internal engineers build the sync from the catalog or content database, index settings, and ranking rules. | One-time, fixed scope. The partner defines the pipeline and delivery dates in writing before work starts. | One-time integration work on the client side, plus vendor onboarding. Costs fall lower when a vendor connector exists for the platform. |
| Migration from an existing search (exports and parameter mapping) | One-time. Meilisearch documents an Algolia export with browseObjects and import in batches of 100,000. Parameters such as aroundPrecision have no direct equivalent and need redesign. | One-time. Partners map parameters, convert query rules and test relevance against the old results. | Not applicable when moving between SaaS vendors. Moving off SaaS requires an export and remapping, which the client should price in advance. |
| Front-end adapter work | One-time. Typesense supports InstantSearch widgets through its adapter with small changes. Meilisearch needs the Instant Meilisearch adapter for InstantSearch front ends. | One-time, included in scope when the front end is listed. Confirm which widgets are covered. | Usually lower if the vendor ships a matching client library. Custom widgets still need developer time. |
| Testing and relevance tuning | One-time for the first release, then recurring as content and catalog change. Staff own synonyms, ranking and typo settings. | One-time for baseline tuning, with an option for recurring tuning under a retainer. | Recurring internal time for merchandising and content owners. Vendor tools may cover analytics, but confirm which features exist for the chosen plan. |
| Security hardening and identity integration | One-time build, then recurring patching. OpenSearch Security provides SAML, OpenID Connect, LDAP and TLS options that must be configured and tested. | One-time, fixed scope, covering API keys, network controls, and identity integration with the client's identity provider. | Recurring vendor responsibility for the hosted service. The client still owns API key handling and access to its data. Request security evidence before signing. |
| Backups, snapshots and upgrades | Recurring. OpenSearch maintains versions for a set period, and Typesense and Meilisearch publish frequent releases. Staff schedule upgrades and test rollback. | One-time runbook, then recurring if the retainer covers upgrades, testing and restore drills. | Usually included in the service, with restore terms set in the contract. Confirm recovery targets and retention. |
| 24/7 operations and support contract | Recurring. On-call rota, monitoring and incident response are internal costs. | Recurring only if a support retainer is signed. Support terms, response times and scope are set in writing. | Recurring vendor support tier. Uptime commitments differ by plan, and the enterprise tier names an uptime SLA. Ask for the exact figure in writing. |
| Internal staff time | Recurring and often the largest line. Covers engineers, search owners, security reviewers and release managers. | One-time for oversight and acceptance, plus recurring time for the client's product owner. | Recurring for integration, merchandising and vendor management, usually lower than in-house operations. |
Which drivers move each cost line
Six decisions set most of the cost of self-hosted site search, and they interact. Each card names the cost line it raises or lowers, so a team can see which choices to settle before asking for a quote. Several of these drivers also change whether a partner or a SaaS subscription is the cheaper route.
Catalog and document size
Memory is the main infrastructure driver. Typesense documents that keyword search typically needs 2 to 3 times the size of the searchable fields in RAM, and each vector adds 7 bytes per dimension. A large catalog with semantic vectors raises the server tier, the hosting line and the load-testing effort. Ask for the record count, average searchable field size and vector dimensions before comparing the hosted and self-hosted lines.
Tenancy model
The choice between one shared index with scoped keys and separate indexes per tenant changes the resource count. Meilisearch lists multi-tenancy that personalizes results for any number of application tenants, but a per-tenant index design multiplies memory and operational overhead. Hosted plans often bill by index, document or request volume, so tenancy can move the subscription line as much as the infrastructure line. Confirm the isolation model before quoting.
Data location and sovereignty
Self-hosting in a chosen region or inside a defined boundary adds hosting cost in the client's own accounts, but removes the need to send queries and content to vendor regions. This line raises the infrastructure and staff cost for the client and can reduce transfer documentation work. The lowest-cost SaaS plan may not be available in the required region, so confirm regional options with each vendor before comparing totals.
Uptime target
A production search tier with high availability needs more nodes. A Typesense cluster needs a highly available 3-node or 5-node setup to tolerate one or two node failures, which multiplies server cost. OpenSearch needs replicas and node roles, and Meilisearch replication depends on the edition, with the Enterprise terms governed by a commercial agreement. Hosted SLA tiers are priced separately, so ask what uptime each option commits to in writing.
Migration source and front-end adapters
Moving from Algolia adds one-time engineering cost. Meilisearch states that aroundPrecision, restrictHighlightAndSnippetArrays and disablePrefixOnAttributes have no equivalent, and InstantSearch front ends need the Instant Meilisearch adapter. Typesense is about 85% feature parity with Algolia, and server-side A/B testing and built-in personalization are missing. Each gap lowers the savings from self-hosting or raises the rebuild effort, so inventory Algolia features in use before estimating.
Feature scope and embeddings
Hybrid or semantic search adds cost that keyword search does not carry. Embeddings need a model, either a hosted API with usage fees or a self-hosted model with its own compute, and vector storage grows with dimensions and documents. Typesense built-in ML models need roughly 2 to 6 GB of extra RAM. Each added capability raises the build, testing and run lines, so define the first release scope and treat semantic results as a separate phase with its own measurement.
Where self-hosting costs land over time
Self-hosting site search splits cost into two groups that behave differently. Setup, migration and tuning are one-time costs: standing up the cluster, mapping the existing index settings, moving records from a hosted engine and tuning ranking and filters. Migration guides for both Meilisearch and Typesense show that most of this work is mapping, not copying. Meilisearch's Algolia guide maps attributesForFaceting to filterableAttributes and notes that some parameters have no equivalent, while Typesense's guide converts query rules to the Overrides API and states that its APIs are not wire-compatible with Algolia. Ask any partner to list which parameters in your current configuration have no direct equivalent, because that list drives the one-time effort more than record count does.
Recurring costs are the ones most self-hosting estimates leave out. Infrastructure runs every month whether or not anyone touches the search. Meilisearch's pricing page states that self-hosting is free of license fees but that operators manage updates, backups and scaling themselves, so those labor lines recur. OpenSearch targets a new minor version about every eight weeks, with 3.9.0 released on September 29, 2026, and each upgrade needs testing, a maintenance window and a rollback plan. Typesense recommends a 3-node or 5-node cluster for production, so the cluster is three to five servers before backups, monitoring and on-call coverage are added. Budget these as a standing line with an owner, not as a project.
On the SaaS side, the price is a subscription plus usage. Meilisearch Cloud starts at $20 per month with a 14-day free trial, and its published base plan example is $30 per month for 100K documents and 50K searches, according to Meilisearch's pricing page. Typesense Cloud bills dedicated clusters by memory and bills bandwidth separately, according to its pricing page. The number that matters for a three-year view is how the bill moves as your catalog grows: a higher document count or memory tier raises the subscription, and bandwidth grows with search traffic. Ask any SaaS quote for the tier your projected record count and query volume would land on in year three, and for every line item that bills outside the base plan.
When each option costs less
Building tends to cost less when a team already runs search or Kubernetes, the catalog and content change slowly, and at least one engineer understands relevance tuning and index operations. In that case the main spend is staff time you already pay for, and the infrastructure line is modest for a single-node deployment of Meilisearch or Typesense. The saving holds only while the people who own it stay in place.
- Existing platform skills: Teams that already operate clusters can absorb a search node into their monitoring, patching and backup routines without a new vendor.
- Low change rate: A stable catalog with few schema changes keeps reindexing and relevance work small and predictable.
- Hidden cost, key-person risk: One engineer who knows the tuning and the Algolia-style migration path is a single point of failure that the budget rarely shows.
- Hidden cost, upgrades and on-call: Self-hosted engines need security patches, version upgrades and an on-call path for outages, all recurring.
- Hidden cost, memory sizing: Typesense documents that keyword search typically needs 2 to 3 times the size of the searchable fields in RAM, so sizing errors show up as cost overruns.
Before choosing this path, confirm who owns the search service after go-live and how much of their week it will take.
An implementation partner tends to cost less when the scope is fixed, the go-live date is tight, hosting must stay inside your boundary, or auditors need documented evidence that your team cannot produce quickly on its own. You pay for the work once, against a written scope, and the platform stays in your own infrastructure. The saving comes from avoiding a long internal learning curve and a rushed launch, not from a lower unit price.
- Fixed-scope delivery: Deliverables, timelines and acceptance criteria are agreed in writing before work starts, so cost is tied to outcomes you can review.
- Faster migration: Existing data and front-end widgets can be moved using vendor migration guides, with the mapping work and testing handled by people who have done it before.
- Regulated hosting: Installation in your own cloud account, data center or air-gapped network keeps data where your rules require it.
- Evidence preparation: Configuration records, access controls and backup tests can be prepared for assessors, though DeployOpen does not issue audit reports or certifications.
- Hidden cost, ongoing support: Support after go-live is usually a separate retainer, so the one-time project fee understates the total cost.
Ask how the fixed scope is defined, what is excluded, and what happens to upgrades and reindexing after handover.
Hosted search tends to cost less when the catalog is small, there is no requirement to keep data or queries in a specific location, and the evaluation window is short. Vendors publish plans with a free starting tier, so a small team can have a working index quickly and pay for resources as usage grows. The saving depends on how usage develops, so model it against your expected document count and query volume.
- Small footprint: Meilisearch Cloud and Typesense Cloud both offer entry tiers, which suit a modest site with limited records.
- Low setup effort: No cluster design, memory sizing or patching is needed in the first months.
- Hidden cost, usage billing: Meilisearch Cloud bills by usage or resources, and Typesense Cloud bills bandwidth separately, so costs can rise with traffic.
- Hidden cost, data location: Query logs and indexed content leave your infrastructure, which may conflict with residency rules and require transfer documentation.
- Hidden cost, migration out: Moving later means re-mapping settings and widgets, so lock-in costs grow with every feature you adopt.
Confirm the data location rules and the expected growth in records and queries before accepting a hosted price as the long-term figure.
How to build your own site search cost estimate
This sequence turns a vague sense of search spend into a line-by-line estimate that covers infrastructure, migration, staff time and support, so you can compare a self-hosted build with a partner engagement or a SaaS subscription at the same volume.
Inventory documents, records and their growth
Start by counting what the search index will hold: pages, products, articles, attachments and any per-tenant records, plus the expected growth over the next budget cycle. Ask the content owners and the catalog or database team to provide counts and the fields that will be searchable, since those fields drive memory use. The step is done when you have a written table of record counts, average record size and a growth assumption that a named owner has signed off.
Size memory for the chosen engine
Convert the inventory into a memory estimate using the sizing rules for your engine. For Typesense, the system requirements guide says keyword search typically needs 2 to 3 times the size of the searchable fields in RAM, each vector needs 7 bytes of memory per dimension, and built-in ML models need roughly 2 to 6 GB of extra RAM. For Meilisearch or OpenSearch, ask the vendor or your engineers for the equivalent guidance and test it against a sample of your real data. The step is done when you have a memory figure per node and a note on whether semantic search is in scope, since vectors can change the total substantially.
Define tenancy, data location and uptime targets
Decide how many search indexes or tenants the platform must isolate, where the data may be stored and processed, and how much downtime the business can accept for search. These answers set the node count and the environments you need. For high availability, the Typesense documentation recommends a 3-node or 5-node cluster for production, and OpenSearch and Meilisearch have their own replication options that depend on edition and design. The step is done when the targets are written down and approved by security, privacy and the business owner.
Price the infrastructure in your own environment
Price the compute, memory, storage, backups, load balancing and non-production copies in the cloud account or data center where the search will run. Cost is recurring for most of these lines, and it rises with node count, memory per node and retention. Ask your infrastructure team for current rates from the providers you already use, and include the cost of a staging environment for upgrades and reindexing. The step is done when each line item has a unit price, a quantity and a monthly or annual total sourced from your own contracts.
List migration and integration work
Count the work to move existing search from its current provider or database and to connect the index to your sources. Migration guides for Meilisearch and Typesense describe export and import paths from Algolia, and both list parameters that need remapping or that have no direct equivalent, which means front-end and query changes. Include the connectors for the CMS, commerce platform or application database, identity integration, and monitoring. The step is done when each integration has a named owner, a short description of the work, and an effort estimate that the owner can defend.
Assign internal staff hours by role
List the roles that will touch the search service during build and during operation, such as platform engineering, application developers, search relevance, security and support. For each role, estimate hours for the build, for routine operation and for upgrades. This is usually the largest line that organizations underestimate, because internal time is rarely tracked against search. The step is done when each role has hours per phase, a cost rate from finance and a note on which staff members are already committed to other work.
Add support and upgrade cadence
Add the recurring work of keeping the platform current and secure: applying maintenance releases, testing upgrades in staging, monitoring capacity, handling backups and restores, and responding to incidents. Check the support and maintenance policy for your chosen engine and version, including which release lines receive fixes. Decide whether support comes from your own team, an implementation partner under a retainer, or a combination. The step is done when you have a calendar of expected upgrade and test work and a support model with a named escalation path.
Compare with SaaS tiers at the same volume
Take the same record count, query volume, tenancy model and uptime target and price them against published SaaS tiers. Meilisearch Cloud and Typesense Cloud publish usage or resource-based pricing, and their pages state which items are billed separately, such as bandwidth. Request written quotes from any partner using the same scope, and confirm current vendor prices and terms directly with each provider, since published pages change. The step is done when the build, partner and SaaS totals sit side by side with the assumptions for each, so the decision rests on comparable numbers.
When an implementation partner costs less than building alone
The partner saving rarely comes from a cheaper license, because the engines in this comparison are open source and the license is zero in every option. The saving comes from avoiding rework: a migration that has to be run twice because the index settings were wrong, a security hardening pass after the search endpoint was already exposed, a backup routine that was never restore-tested, and an upgrade that breaks the front end because the adapter version was not pinned. For Meilisearch, that means checking how Algolia parameters map to index settings and search parameters, and which ones have no equivalent. For Typesense, it means planning the Overrides conversion and the InstantSearch adapter. For OpenSearch, it means setting roles, TLS and audit logging before any team member queries production data. An internal team can learn each of these, but usually learns them on the live system.
Some costs never leave the internal team. After go-live, your staff still own the query and relevance tuning, the reindexing schedule, on-call response, capacity decisions and the day-to-day decision on when to apply a maintenance release. For a self-hosted engine, this is recurring work that scales with catalog growth and traffic, not a one-time cost. Ask a partner to state which of these items it hands over, which it operates under a retainer, and what documentation each handover includes. The comparison only holds if you ask for the same things from both sides.
The comparison is fair only when both options are priced against the same written scope. Ask each bidder for a fixed scope with named deliverables: the migration plan and its acceptance tests, the hardening baseline, the tested backup and restore runbook, the upgrade procedure, the tuning review at agreed intervals and the handover checklist. Then compare internal staff hours and infrastructure against the same list. Hidden costs usually sit in the items nobody wrote down, so require each quote to say what is excluded. DeployOpen plans and delivers fixed-scope engagements of this kind, and the quote follows from the written scope.
Questions buyers ask before choosing build, partner or SaaS
How much does self-hosted site search cost compared with SaaS?
A self-hosted engine has no license fee for the core products, so the cost is infrastructure, setup, operations and staff time, while SaaS is a recurring subscription that scales with usage. Meilisearch's pricing page says self-hosting is free with full infrastructure control and no usage limits, and lists Meilisearch Cloud from $20 per month with usage-based or resource-based billing. Typesense Cloud bills dedicated clusters by memory and bandwidth. Compare the full three-year cost, not the first invoice. Ask each vendor for a quote based on your record count, search volume and region, and add your own infrastructure and staffing estimate to the self-hosted side.
What work does our team need to do itself?
Your team owns the parts that a SaaS vendor would otherwise run. The main duties are:
- Provisioning and sizing the cluster, including memory for the index
- Index updates, reindexing and backups
- Upgrades, patching and monitoring of the search service
- Relevance tuning, synonyms and the front-end integration
Meilisearch's pricing page states that operators manage updates, backups and scaling themselves. Confirm how many of these tasks your current engineers already perform for other services, because that determines whether the work is an addition or a new skill.
How long does a migration from Algolia take, and what drives it?
The duration depends mainly on the number of records, how many index settings and query rules you use, and how much your front end depends on Algolia-specific behavior. Both Meilisearch and Typesense publish export and import paths: Meilisearch imports in batches of 100,000 records, and Typesense exports with the Algolia CLI and imports with an upsert action. Some Algolia features have no direct equivalent, including aroundPrecision and server-side A/B testing for Typesense. Ask a partner to inventory settings, rules and widgets before quoting, and agree the migration scope in writing.
Can we run site search without an implementation partner?
Yes, if your team has the operations skills and time to run a search cluster after go-live. Meilisearch Community and Typesense are open source, and OpenSearch is released under Apache License 2.0 with installation options that include Docker, Helm and the OpenSearch Kubernetes Operator. The harder parts are sizing, high availability, identity integration and evidence for audits. A partner is most useful when the scope spans several of these, when the deadline is fixed, or when the team has no one to take on-call duties. Ask any partner for a fixed-scope written proposal, a named deliverables list and the support terms after handover.
Which costs recur every year?
Recurring costs fall into three groups: infrastructure, people and support. Infrastructure includes compute, memory, storage, backups and network traffic, which grow with the index. People include the staff time spent on upgrades, monitoring and tuning. Support covers any retainer for patches, security testing or upgrades. Upgrades matter here: OpenSearch maintains each version with bug and security fixes for a limited period, and Typesense recommends a highly available three-node or five-node cluster for production. Build these recurring lines into the model before comparing it with a subscription.
How do catalog size and tenancy change the bill?
Catalog size changes memory, and tenancy changes how many indexes or clusters you run. Typesense's guide says keyword search typically needs 2 to 3 times the size of the searchable fields in RAM, and vector data adds 7 bytes per dimension. Meilisearch's README describes multi-tenancy that personalizes results for many application tenants, so isolation design affects the number of indexes. Subscription bills often rise with records and searches, so a large catalog with many tenants can shift the balance. Ask vendors to price your largest tenant and your peak search volume, not an average.
What should we confirm with vendors before signing?
Confirm the licence terms for the exact edition you plan to run, the migration path for your current data and the uptime commitments for any hosted tier. The Meilisearch README says the Enterprise Edition is governed by a commercial license or the Business Source License 1.1 and is not permitted in production without a commercial agreement, so check which features you need. Ask about SSO, replication and support terms in writing. The Typesense GPL-3.0 server allows internal use of modified code without publishing changes if you do not distribute it, which your legal team should review against your deployment model.
How DeployOpen scopes and delivers a fixed-scope search implementation
DeployOpen plans, implements, integrates, hardens, backs up and documents self-hosted search inside your own cloud accounts, data centers or air-gapped networks. The cost of the build is set by a written scope, so the cost structure in this guide becomes concrete only once the workload, hosting and support terms are agreed.
How we scope
- We confirm catalog or content size, number of indexes and tenants, and expected query volume before estimating anything.
- The written scope names the engine, edition, hosting location, identity integration and every migration step for your search.
- Engagements are fixed-scope, so deliverables, timelines and the support tier are agreed in writing before work starts.
How we build
- We size memory, nodes and replicas from your data and the engine's documented requirements, then show the resulting infrastructure line.
- We build indexing pipelines, relevance settings and integrations with your identity provider inside your own environment.
- We test relevance, failover and reindexing, and record the results so you can compare the self-hosted run cost with a SaaS quote.
How we hand over
- You receive runbooks covering updates, backups, scaling and reindexing, so your staff can run the search without us.
- We document licence obligations for the chosen edition, including what changes if you add a commercial feature later.
- We walk your team through the handover and list the recurring costs you will own, such as infrastructure, staff time and upgrades.
How we support
- Ongoing support, upgrades and continuous testing are available as retainers, priced separately from the build.
- A retainer can cover version upgrades and security patching on the engine's supported release lines, scoped in writing.
- Ask any partner to state in the quote which costs are one-time, which recur monthly and which depend on your traffic.
