Apache Search and Knowledge Tools for Small Business

Apache Answer is the practical owner-run option for a question-and-answer site. Solr and Nutch need a developer or provider. The remaining projects connect repositories, operate search infrastructure or supply code inside other products.

October 30, 2026 update: added setup time and costs when an AI agent does the typing.

There are several different problems hiding behind “we need better search.” Staff may need one approved answer to a recurring question. Customers may need to find a part by an alternative name. A company may need documents from several systems to appear together without exposing private files. Buying one search engine does not settle all three.

Before choosing software, collect ten questions people actually ask. Include a question whose answer has changed and a document that only some staff may see. Those examples will tell you more about the fit than an attractive demonstration with public sample data.

We read official pages on October 29, 2026 and have not installed these projects for this series. The business examples are illustrations. Our cost comparisons cover modest, defined jobs and spell out the integration work they exclude.

Every project in this group

The Solr downloads and news distinguish the current 10.0.0 major release from the later-dated 9.11.0 release on the previous major branch. “Latest” needs that qualification. Lucene.Net's displayed release is explicitly a beta. Nutch's date comes from its tagged release log; its news announcement appeared in September and its download page carries an inconsistent older date. ManifoldCF and PyLucene use the current official pages to correct older catalog details.

Answer: keep the useful answer after the conversation ends

Answer gives a team a Q&A site with questions, answers, tags and community participation. It can make repeated explanations findable. It does not decide which answer is correct or which procedure has expired. Assign someone to maintain the content before adding another place where old instructions can accumulate.

For example, a nine person HVAC company wants technicians to stop calling the senior technician about the same controller reset. It would create tagged questions with an approved answer, model references and a review date. On a normal day the technician searches before calling; a supervisor still reviews safety-sensitive instructions. Our one-host budget is $92 to $229 monthly, including routine administration, plus setup. This does not estimate time spent writing or approving answers.

For example, a two-location coffee shop with fourteen employees needs consistent grinder-cleaning and opening instructions. A shift lead would maintain a small collection of questions, with a separate tag for each machine. Staff can flag an unclear answer during a shift instead of editing the procedure informally. The same budget applies under our assumption of one small installation; fourteen staff is a business example, not a measured capacity claim.

The installation guide recommends Docker Compose and also documents Docker and binary routes. It supports several databases. The FAQ recommends MySQL for actual use and says database migration is not currently supported. Decide the database before building a valuable archive; the easiest demonstration database is not automatically the right long-term choice.

Those pages publish no universal numerical CPU, RAM or disk minimum. Our one-server, 2 vCPU, 4 GiB choice is a pilot budget for Answer and a small database, not an upstream sizing promise. You need Docker, DNS and TLS skills, persistent storage, a database restore procedure and someone to manage accounts. Registration email needs SMTP configuration. We assume an existing suitable mail allowance; obtaining a new mail service adds cost.

The download page confirms 2.0.3 dated October 9, 2026. Start with that release and the official installation instructions. Back up the database and uploaded files together. A copy of a container image does not preserve the questions people wrote.

The practical managed alternative here is Discourse Pro, $100 monthly, with five staff seats, unlimited members, a custom domain, private categories and the Solved plugin. It is a broader forum product rather than hosted Answer. Discourse also offers a free plan; compare that first if its limits fit. Our worked case selects Pro because the example wants its domain and community controls, not because payment is compulsory for every forum.

Our low formula is 4 × $68.09 + 36 × ($24 + $0.278 + $68.09) = $3,598. Discourse is 36 × $100 = $3,600. That leaves only about $2 of modeled savings over three years and a break even in month 36. At the high labor assumption Answer costs $5,457 more and never breaks even. One unexpected support task removes the low-case saving. Moderation and content writing are excluded on both sides, not assumed to take no time.

Skip Answer if your existing shared handbook already answers the questions and staff use it. A six person nonprofit may need an edited volunteer guide more than a community site. If Q&A is appropriate, start with ten real questions, test access from an ordinary staff account and agree how answers get approved, corrected and retired.

Solr: make a product or document collection searchable

Solr is a search service. It indexes documents and supports queries, filters and relevance tuning, but your storefront or staff application still needs an interface and a reliable way to load records. Installing the engine does not automatically connect a stock system or understand the vocabulary customers use.

For example, a six person auto-parts retailer has one shop and an online catalog. Customers use a familiar part nickname while the supplier uses a formal description. A developer would index approved product fields, add suitable synonyms and keep stock updates separate from descriptive text. The normal-day change is that customers can find a part using either term. Our single-server operating budget is $116 to $253 monthly, plus setup and a separate integration quote.

For example, a twelve person machine shop wants staff to find approved public specifications by material and tolerance. It would index a controlled collection and provide filters for document type and revision. Staff would see the current approved file rather than an old download someone kept. The same server estimate applies only to a modest collection that passes a load test. It does not include scanning paper drawings or cleaning poor source documents.

Solr 10 requirements call for Java 21 on the server. The production guide documents a single Linux instance using the supplied service installer. Binary archives and Docker images are available through the official downloads. A cluster is not compulsory for a small installation, but one host has no failover.

The guide does not provide a universal production RAM minimum. Its small default heap is for getting started, and it says production heaps of 8 to 16 GB are not uncommon. Our 8 GiB whole-machine budget is therefore a constrained pilot assumption, not a promise that a busy production index fits. Leave memory for the operating system and filesystem cache; an 8 GiB server cannot devote all of that to the Java heap. Increase the quote if the representative test needs more.

Your provider must own schemas, indexing, authentication, backups and query performance. An index contains another copy of your data. Deleting a file from the source does not prove its indexed text, snippets and backup copies are gone. Test deletion and permission changes as carefully as a successful search.

Algolia Grow is a practical managed search alternative. The verified pay-as-you-go allowance includes 10,000 search requests and 100,000 records monthly, then $0.50 per 1,000 extra requests and $0.40 per 1,000 extra records. Our scenario assumes 100,000 requests and 50,000 records at both low and high ends: (100,000 minus 10,000) / 1,000 × $0.50 = $45 monthly, with no record overage or other billable features. This is an alternative search platform, not managed Solr.

The self hosted low calculation is 4 × $68.09 + 36 × ($48 + $0.278 + $68.09) = $4,462. The managed search calculation is 36 × $45 = $1,620. Self hosting costs $2,842 to $8,301 more and never breaks even in this scenario. Custom storefront integration is excluded on both sides. Search-as-you-type may generate several requests while a person types one phrase; measure requests instead of budgeting by customer visits.

If you specifically need managed Solr, SearchStax lists Core production from $2,499 monthly equivalent billed annually. Its $53 Development node is for development or noncritical work and carries no SLA; it is not a like-for-like production offer. The difference in scope is a reason to get a real quote, not to compare the smallest number on each page.

Skip Solr if the existing shop platform has good enough search and no one can name a missing capability. Start with an approved sample collection and the ten real queries. Include spelling variants, empty results and a permission-restricted document. Agree on acceptable results before asking a developer to tune the engine.

Nutch: collect pages for a search system

Nutch crawls websites and prepares content for indexing. It is a collector, not a finished search site. It still needs somewhere to send the collected documents and a search interface for readers. That separation makes its cost comparison different from Solr's.

For example, a garden retailer with two locations and eleven employees has seasonal care pages scattered across its own site. A provider would restrict Nutch to those public pages, schedule collection and remove content that disappears. Staff could keep the underlying collection current without copying every page manually. Our collector-only server budget is $116 to $253 monthly plus setup; search serving is outside both sides of the comparison below.

For example, a four person print shop maintains public guides to paper stock and finishing options. It would collect only approved guide sections, exclude obsolete price pages and send the resulting text to its existing search system. On a normal day an updated guide enters the next collection cycle. The same cost applies to our modest pilot assumption, not to unrestricted internet crawling.

The official tutorial documents local operation on one machine and an unpacked binary installation. Nutch 1.23 uses Java 17. The tagged release log requires Hadoop 3.5 or newer for clustered operation; a local pilot does not need a Hadoop cluster. No universal numerical hardware minimum is published in the tutorial. Our 8 GiB host is a sizing assumption that must be tested against crawl state, extraction memory and retained content.

There is also a compatibility trap. The tutorial's integration table still targets Solr 8.11.4. Do not assume Nutch 1.23 and the newest Solr major release will work together without integration testing. Ask the provider to demonstrate the supported combination and its patch plan. Choosing compatible, maintained components is part of the job.

The download page offers 1.23. Its date conflicts with other official pages, so we use the August 7, 2026 date in the tagged change log and disclose that basis. The security page explains fixes in 1.23, including removal of the old unauthenticated REST server. An old crawler tutorial should never become a reason to expose that service.

For a managed collection alternative, Algolia Grow includes 10,000 crawls monthly and charges $0.80 per 1,000 extra crawls. Our fixed low and high volume assumption is 100,000 monthly crawls: (100,000 minus 10,000) / 1,000 × $0.80 = $72 monthly. This is a volume illustration for a repeatedly refreshed collection, not a claim that the two small example sites need that many crawls. At much lower volume the included allowance can make the managed alternative cheaper still.

The collector comparison is $4,462 to $9,921 self hosted versus 36 × $72 = $2,592 managed, so self hosting costs $1,870 to $7,329 more. Break even is never. Both sides exclude search serving, search requests, record overages, interface work and content editing. Adding Solr to the self hosted side or Algolia search to the managed side requires another line in the worksheet.

Start with a short seed list and strict URL filters. Respect site rules, identify the crawler honestly and use a conservative request schedule. Do not bypass logins, paywalls or bot challenges. Restrict network access so a public-page collection job cannot reach private resources. Skip Nutch if the content platform already exports clean records to the search provider; that may be simpler than collecting rendered pages.

Needs your IT department or a provider

ManifoldCF connects repositories to a search engine while carrying document permissions. A provider must map identities and access rules correctly. The published deployment guide is still for 2.30 while downloads list 2.31. It describes a single-process evaluation setup and database-backed services and agents for larger operation; it gives no universal hardware minimum. A small assessment can start on one host plus an existing index, but production sizing must include the database, connectors and index. Our oversight allowance is 6 to 16 hours monthly, plus integration.

An eighteen person property manager connecting restricted maintenance records may justify ManifoldCF. Ask what happens when an employee loses access or a document becomes private. Stack Internal Organization is a managed knowledge alternative with contact-sales pricing, not guaranteed connector parity. SearchStax Core's published $2,499 monthly annual-billing floor covers managed indexing, not the ManifoldCF permission work. Neither should be presented as a complete replacement without a demonstrated connector plan.

Solr Operator manages Solr in Kubernetes. Its operator instructions require Kubernetes resources, Solr and coordination services. The local tutorial puts several tiny nodes on one laptop; that is a demonstration, not minimum resilient production hardware. The smallest sensible business context is an existing supported Kubernetes platform with persistent storage and a provider. Our additional oversight allowance is 6 to 16 hours monthly; underlying platform operation needs its own allocation.

A thirty person ecommerce brand already operating Kubernetes may justify the operator. A small retailer should not buy a cluster just to install search. Compare SearchStax Core at its verified $2,499 monthly annual-billing floor and ask who handles storage restores and failed nodes. Read the storage configuration: default storage choices must not be mistaken for a durable backup plan.

Your developer's toolbox

Lucene Core provides Java indexing and search code inside applications. Lucene.Net brings that family of capabilities to .NET; ask whether your vendor uses the beta or an older stable version. PyLucene exposes Lucene to Python while embedding a Java runtime, adding another dependency to maintain. StormCrawler supplies components for building crawler applications; your developer owns the collection behavior. Tika extracts text and metadata from documents. It does not decide who may see the text or certify that an uploaded file is harmless.

For a small law office, check attorney-client confidentiality before indexing matter files or sending them to a service. For a clinic, check HIPAA obligations before storing patient material in an index. Those checks do not establish compliance; the software choice cannot do that by itself.

The shared cost worksheet

Our assumptions use the series research read October 29, 2026: $68.09 per hour for administration, derived from the BLS administrator wage of $47.66 divided by the 70.0% wages share of compensation. This estimates employee compensation. An outside provider bills more; get a quote.

DigitalOcean monthly prices are $24 for 2 vCPU and 4 GiB RAM, or $48 for 4 vCPU and 8 GiB RAM. Backblaze B2 is $6.95 per decimal TB per month, with the first 10 GB free. Our low and high backup assumptions are 50 and 100 GB retained, including versions: (50 minus 10) × $0.00695 = $0.278, or (100 minus 10) × $0.00695 = $0.6255 monthly. We assume no chargeable excess transfer or API calls. Check both providers' allowances before moving large files.

For one server, our low and high labor assumptions are 4 and 12 setup hours, then 1 and 3 administration hours per month. For a small cluster they are 24 and 60 setup hours, then 6 and 16 hours per month. These are planning ranges, not measured installation times. Custom development, data cleanup, migration, training, compliance reviews and major incidents are excluded. Budget those separately on both sides.

Setup = hours × $68.09. Monthly = server charges + backups + administration hours × $68.09. The 36 month total = setup + 36 × monthly. Savings = alternative total minus self hosted total. Break even = setup difference divided by monthly savings, rounded up, only when monthly savings are positive. Displayed totals are rounded to whole dollars; calculations retain cents. Prices are US dollars before tax, read October 29, 2026.

All three worked examples assume one server at both ends and 50 to 100 GB retained backup data. SaaS incremental setup is zero by assumption; custom integration and moderation are excluded on both sides. Fixed search and crawl volumes are explicitly assumptions, not observed customer traffic. The one-server examples have no availability cluster.

Setup time when Claude or ChatGPT does the typing

We keep the HVAC team's Answer example: one $24 server, its existing SMTP allowance and 50 to 100 GB of retained backups. Our AI assumptions use 50% to 80% of manual setup hours and 75% to 100% of monthly admin hours. Prompting, review, testing and corrections are included; these are not measured installations. Coding studies cannot establish Answer deployment savings.

Claude Pro and ChatGPT Plus each cost $20 monthly, before tax, in the October 29 research. We charge a separate setup month, then allocate $5 monthly for 36 months only when other work uses the remaining 75%. Sole use adds $540. Included coding access has usage limits; price extra usage before starting.

Manual 36 months = S x L + 36 x (H + A x L)
AI 36 months = S x q x L + T0 + 36 x (H + A x r x L + Tm)
API fee = (uncached input x input rate + cache writes x write rate
           + cache reads x read rate + output x output rate) / 1,000,000

Here S and A are manual hours, q and r the AI shares, L is $68.09 hourly, H is unchanged hosting plus backup, and T0/Tm are setup/monthly tool fees. We calculate before rounding totals to whole dollars.

Our cumulative API assumption is 200,000 to 1,000,000 setup input tokens and 20,000 to 100,000 output; monthly budgets are half those amounts, including retries. Low input is 30% ordinary, 10% short cache writes and 60% cache reads; high input is uncached. Opus setup/monthly fees are $0.764/$0.382 low and $6/$3 high; Astra costs $1.97/$0.985 and $15/$7.50. The kit supplies the rates.

Claude Code or Codex CLI can draft Answer's Compose configuration and database restore commands. Desktop Commander on the server or through authorized SSH can install Docker, edit SMTP settings and inspect container logs; our base uses its free local server. Claude in Chrome can inspect Cloudflare DNS, the provider panel and Answer's admin UI. ChatGPT's cloud browser uses a separate browser; confirm availability.

Use a new server and approve commands or restrict a sandbox; never paste secrets into chat. Watch dashboard actions, review account permissions and test a database/upload restore. Use subscriptions for supervised supported tools and API keys for custom pipelines; you still own backups, updates and acceptance.

The conclusion improves only at the low end: subscription assistance raises Answer's saving over Discourse from $2 to $551 and moves break even to month 8. High cases still lose; their unchanged admin hours can make assistance cost more than manual work.

Risks and support

Read the Answer security page, Solr security page, Nutch security page and Tika advisories, ManifoldCF advisories and ASF security guidance. Isolate document parsing, limit upload sizes and keep extraction components updated. Public search should expose only approved content, including snippets, previews and cached text.

Check release dates, mailing lists and retirement notices before relying on a project. Answer support does not promise an instant response; Solr's community is not an on-call contract. SearchStax publishes its managed offering on its own site, so it is a real place to request a support quote. Ask which components the contract actually covers.

The Apache License 2.0 permits commercial use and modification subject to its terms, including distribution notices. It provides no warranty or trademark permission. This is a practical summary, not legal advice.

The useful first acceptance test is simple: find the right answer, fail to find a private document from an unauthorised account, and remove an obsolete answer completely. The small-business budgeting perspective in The $97 Launch applies here: fund a specific improvement that a person can verify.

Download the kit

The search and knowledge kit includes the Answer starter plan, provider cards, complete project table and filled cost worksheet.

Related reading

• Private document search and AI answers: three Apache paths
• ManifoldCF: permission aware document search and setup kit
• Apache Tika: searchable business documents and setup kit
• Apache projects for small business
• Workflow automation
• Websites and content systems
• Security and identity
• Building a self hosted AI server

Fact-check notes and sources

1. Apache Answer. Read October 29, 2026, 21:12 UTC.

2. 2.0.3; 2026-10-09. Read October 29, 2026, 21:37 UTC.

3. Apache Lucene Core. Read October 29, 2026, 21:13 UTC.

4. 10.5.2; 2026-10-04. Read October 29, 2026, 21:00 UTC.

5. Apache Lucene.Net. Read October 29, 2026, 21:13 UTC.

6. 4.8.0-beta00018 (Beta); 2026-06-22. Read October 29, 2026, 21:40 UTC.

7. Apache ManifoldCF. Read October 29, 2026, 21:13 UTC.

8. 2.31; 2026-07-30. Read October 29, 2026, 21:37 UTC.

9. Apache Nutch. Read October 29, 2026, 21:13 UTC.

10. 1.23; 2026-08-07 (release log). Read October 29, 2026, 21:39 UTC.

11. Apache PyLucene. Read October 29, 2026, 21:40 UTC.

12. Apache Solr. Read October 29, 2026, 21:13 UTC.

13. 10.0.0 current major; 9.11.0 previous major; 2026-03-03; 2026-10-04. Read October 29, 2026, 21:37 UTC.

14. Apache Solr Operator. Read October 29, 2026, 21:13 UTC.

15. Apache Stormcrawler. Read October 29, 2026, 21:13 UTC.

16. 3.7.0; 2026-07-29. Read October 29, 2026, 21:16 UTC.

17. Apache Tika. Read October 29, 2026, 21:13 UTC.

18. news. Read October 29, 2026, 21:38 UTC.

19. installation guide. Read October 29, 2026, 21:37 UTC.

20. FAQ. Read October 29, 2026, 21:38 UTC.

21. Discourse Pro. Read October 29, 2026, 21:38 UTC.

22. Solr 10 requirements. Read October 29, 2026, 21:39 UTC.

23. production guide. Read October 29, 2026, 21:39 UTC.

24. Algolia Grow. Read October 29, 2026, 21:37 UTC.

25. SearchStax. Read October 29, 2026, 21:38 UTC.

26. official tutorial. Read October 29, 2026, 21:39 UTC.

27. download page. Read October 29, 2026, 21:39 UTC.

28. security page. Read October 29, 2026, 21:40 UTC.

29. published deployment guide. Read October 29, 2026, 21:38 UTC.

30. Stack Internal Organization. Read October 29, 2026, 21:37 UTC.

31. operator instructions. Read October 29, 2026, 21:39 UTC.

32. local tutorial. Read October 29, 2026, 21:39 UTC.

33. storage configuration. Read October 29, 2026, 21:39 UTC.

34. BLS administrator wage. Read October 29, 2026, 21:12 UTC.

35. 70.0% wages share of compensation. Read October 29, 2026, 21:17 UTC.

36. DigitalOcean monthly prices. Read October 29, 2026, 21:10 UTC.

37. Backblaze B2. Read October 29, 2026, 21:10 UTC.

38. Answer security page. Read October 29, 2026, 21:38 UTC.

39. Solr security page. Read October 29, 2026, 21:38 UTC.

40. Tika advisories. Read October 29, 2026, 21:40 UTC.

41. ManifoldCF advisories. Read October 29, 2026, 21:38 UTC.

42. ASF security guidance. Read October 29, 2026, 21:39 UTC.

43. Answer support. Read October 29, 2026, 21:40 UTC.

44. Solr's community. Read October 29, 2026, 21:40 UTC.

45. Apache License 2.0. Read October 29, 2026, 21:39 UTC.

46. Claude pricing. Read October 29, 2026, 23:31:08 UTC.

47. ChatGPT pricing. Read October 29, 2026, about 23:27 UTC in a regular browser.

48. Codex with a ChatGPT plan. Read October 29, 2026, 23:31:16 UTC.

49. Claude API pricing. Read October 29, 2026, 23:31:14 UTC.

50. OpenAI API pricing. Read October 29, 2026, 23:31:13 UTC.

51. OpenAI prompt caching. Read October 29, 2026, 23:32:00 UTC.

52. Codex CLI. Read October 29, 2026, 23:32:01 UTC.

53. Desktop Commander. Read October 29, 2026, 23:31:09 UTC.

54. Desktop Commander pricing. Read October 29, 2026, 23:31:08 UTC.

55. Desktop Commander permissions and installation. Read October 29, 2026, 23:31:41 UTC.

56. Claude in Chrome. Read October 29, 2026, 23:31:11 UTC.

57. Claude in Chrome permissions. Read October 29, 2026, 23:31:09 UTC.

58. ChatGPT cloud browser. Read October 29, 2026, 23:32:01 UTC.

59. METR coding study. Read October 29, 2026, 23:30:58 UTC.

60. Anthropic consumer terms. Read October 29, 2026, 23:31:08 UTC.

61. OpenAI terms. Read October 29, 2026, about 23:27 UTC in a regular browser.

Apache, the Apache feather logo and the Apache project names are trademarks of The Apache Software Foundation. This site is not affiliated with or endorsed by the ASF. Other product names are used for identification only.