# Web Search & Extract Providers β€” the honest comparison Built for **Web Search Plus** users configuring `web_search_plus` and `web_extract_plus`. WSP is capability-based: configure **one search-capable provider** and search works; extraction is additive. Pick providers by what their **free tier**, **pricing**, and **capabilities** actually are β€” not by landing-page glitter. **Verified 2026-08-22** against live provider pricing/billing/docs where possible. Prices and free tiers change often; re-confirm at signup before relying on a provider for production. ## Quick picks - **Best no-card starter:** Tavily. - **Best extraction/scrape fallback:** Firecrawl. - **Best generous one-time starter credit:** You.com. - **Best cheap Google-style SERP:** Serper or SerpBase. - **Best semantic discovery:** Exa, but only with the credit caveat. - **Best no-key public experiment:** Keenable public tier, opt-in. - **Best local no-provider-key path:** DonSeTch, explicit-only local stdio provider. - **Best privacy/self-host path:** SearXNG. ## Free-tier types - 🟒 **Forever / recurring** β€” recurring allowance, currently-free API, self-hosted, or keyless public tier. - 🟑 **One-time / trial** β€” signup credits you burn once, then pay. - πŸ”΄ **None** β€” paid from request one / no durable free API tier found. - βšͺ **Depends** β€” gated behind another account or billing system. --- ## Privacy & Terms **Standard self-serve contract audit: 2026-08-22.** Explicit-only is a routing control, not a privacy control. Enterprise ZDR or no-training claims count only when they are activated and contractually binding for the exact endpoint. Search and Fetch/Extract are rated separately because those scopes often differ. - **Caution:** no explicit training licence found, but retention or recipients remain material. - **Review:** the standard public contract is incomplete, ambiguous, or lacks a binding no-training/retention promise. - **High risk:** standard terms expressly permit training/fine-tuning or comparably broad use. - **Tech reject:** not active in WSP because no verified source-only endpoint exists. | Provider | Search | Fetch / Extract | Training, retention & ZDR | Primary sources | |---|---|---|---|---| | Serper | **Review** | **Review** | No explicit Customer-Data training licence found, but activity/analytics data may support AI/ML improvement. No fixed query retention, public ZDR or public DPA found. | [Terms](https://serper.dev/terms) Β· [Privacy](https://serper.dev/privacy) | | SerpBase | **Caution** | N/A | No explicit training licence found. Query text, request metadata and IP data are logged without a published fixed retention period or ZDR. | [Terms](https://serpbase.dev/terms) Β· [Privacy](https://serpbase.dev/privacy) | | Brave Search API | **Caution** | N/A | No training/fine-tuning licence found. Standard query logs: up to 90 days. Search API ZDR is Enterprise/Custom and must be activated. | [Terms](https://api-dashboard.search.brave.com/documentation/resources/terms-of-service) Β· [Privacy](https://api-dashboard.search.brave.com/app/documentation/general/privacy-policy) Β· [ZDR](https://brave.com/blog/search-api-zero-data-retention/) | | Tavily | **Review** | **Review** | Query Data may improve future responses. The β€œAllow Use of Query Data” opt-out matters, but its default and a general no-training promise are unclear; no fixed retention found. | [Terms](https://www.tavily.com/terms) Β· [Privacy](https://docs.tavily.com/documentation/privacy) Β· [Opt-out](https://help.tavily.com/articles/4205958832-understanding-the-allow-use-of-query-data-setting) | | Querit | **High risk** | N/A | Personal data may train, develop and improve AI/ML; API content receives broad perpetual improvement rights. No ZDR or fixed query retention found. | [Terms](https://www.querit.ai/en/terms) Β· [Privacy](https://www.querit.ai/en/privacy) | | Linkup | **High risk** | **High risk** | Standard Terms permit Client Data, including prompts, to be used for training and improvement. Public ZDR/BYOC claims require a signed controlling contract. | [Terms](https://www.linkup.so/terms-of-use) Β· [Privacy](https://www.linkup.so/privacy-policy) Β· [Data privacy](https://docs.linkup.so/pages/security-and-privacy/data-privacy) | | Exa | **High risk** | **High risk** | Query Data may be used for training and fine-tuning; Terms grant broad input/output improvement rights. ZDR is Enterprise-only. | [Terms](https://exa.ai/assets/Exa_Labs_Terms_of_Service.pdf) Β· [Privacy](https://exa.ai/privacy-policy) Β· [ZDR](https://exa.ai/blog/zdr-search-engine) | | Firecrawl Cloud | **Review** | **Review** | Cloud policy covers caching, indexing, analytics and product improvement; no fixed API-content retention. Firecrawl documents Enterprise ZDR for scrape, crawl and search; confirm contractual activation, plan coverage and endpoint-specific terms before sensitive use. Self-hosting is a separate operator-controlled path. | [Terms](https://www.firecrawl.dev/terms-of-service) Β· [Privacy](https://www.firecrawl.dev/privacy-policy) Β· [ZDR](https://www.firecrawl.dev/glossary/web-scraping-apis/zero-data-retention-web-scraping) Β· [Self-host](https://docs.firecrawl.dev/contributing/self-host) | | Parallel | **High risk** | **High risk** | Customer Terms expressly permit queries and outputs to train and improve AI/ML models. | [Customer Terms](https://parallel.ai/customer-terms) Β· [Privacy](https://parallel.ai/privacy-policy) | | You.com | **Review** | **Review** | Self-serve lacks a standard no-training or fixed-retention promise. Enterprise ZDR can cover Search `/v1/search` and Answer `/v1/answer`, but not Contents/Extract or Research. | [Terms](https://you.com/terms) Β· [Privacy](https://you.com/privacy) Β· [ZDR](https://you.com/docs/administration/zero-data-retention) | | SearXNG self-hosted | **Caution** | N/A | You control instance logs and retention, but every enabled upstream engine receives the query. Proxy logs, autocomplete and favicon services add data paths. | [Self-hosting](https://docs.searxng.org/own-instance.html) Β· [Engines](https://docs.searxng.org/admin/settings/settings_engines.html) Β· [Egress](https://docs.searxng.org/admin/settings/settings_outgoing.html) | | Keenable | **Review** | **Review** | Processes queries, fetch URLs, content and telemetry. No public no-training promise, fixed input retention, ZDR, DPA or complete subprocessor scope verified. | [Privacy](https://keenable.ai/privacy) Β· [Trust](https://trust.keenable.ai/) | | DonSeTch local stdio | **Review** | **Review** | No DonSeTch cloud, but search engines receive queries and target sites receive fetches. Local browser/runtime state can persist data. AGPL-3.0-only; install separately. | [Project](https://github.com/dondai44423/donsetch) Β· [License](https://github.com/dondai44423/donsetch/blob/main/LICENSE) | | Octen via Monid | **High risk** | N/A | Octen expressly permits ML training on input/output; Monid adds a broker recipient without verifiable public Terms, Privacy, retention or DPA pages at audit time. | [Octen Terms](https://octen.ai/terms-of-service) Β· [Octen Privacy](https://octen.ai/privacy-policy) Β· [Monid API](https://docs.monid.ai/api/overview.html) | | TinyFish | **High risk** | N/A in WSP | BYOK: use your own TinyFish account/API key; WSP does not provide, pool, proxy or share TinyFish credentials. Standard Terms allow Customer Data to train, fine-tune, evaluate and improve AI/ML models. Search usage records include query details/results; no fixed deletion period is published. | [Terms](https://www.tinyfish.ai/terms) Β· [Privacy](https://www.tinyfish.ai/privacy-policy) Β· [Usage records](https://docs.tinyfish.ai/api-reference/list-search-usage) | | Perplexity API *(not active)* | **Tech reject** | **Tech reject** | Strong contractual no-training clause and Chat Completions ZDR, but no verified source-only WSP endpoint. | [API Terms](https://www.perplexity.ai/hub/legal/perplexity-api-terms-of-service) Β· [Privacy & security](https://docs.perplexity.ai/docs/resources/privacy-security) | | Kilo / Perplexity bridge *(not active)* | **High risk + tech reject** | N/A | Additional recipient, broad perpetual improvement rights and downstream model terms; also no verified source-only WSP endpoint. | [Terms](https://kilo.ai/terms) Β· [Privacy](https://kilo.ai/privacy) | This is operational guidance, not legal advice. Re-check the linked primary sources before sensitive or production use. --- ## 🟒 Forever / recurring Recurring allowance, self-hosted, or currently free/keyless. Still read the caveats: Brave needs a card; Keenable can change; SearXNG and DonSeTch need local infrastructure. ### Tavily β€” *search + extract + research* - **🟒 forever** β€” 1,000 API credits / month, recurring, no credit card. - **Pricing:** Pay-as-you-go $0.008 / credit; Project about $30/mo for 4,000 credits; higher plans reduce unit cost. - **Best for:** Convenient no-card starter for users who want search + extract in one account. - **Watch out:** Credits are not always equal to calls; deeper research-style features can consume more credits. - **Key:** `TAVILY_API_KEY` Β· **Source:** https://docs.tavily.com/documentation/api-credits, https://docs.tavily.com/documentation/keyless ### Firecrawl β€” *search + extract + JS/crawl* - **🟒 forever** β€” 1,000 credits / month, recurring β€” no cost, no card, no hassle. - **Pricing:** Hobby $16–19/mo for 5k credits; Standard about $83–99/mo for 100k; Growth about $333–399/mo for 500k. Search costs 2 credits / 10 results; scrape/crawl/map are commonly 1 credit per page/call. - **Best for:** Best extraction/scraping fallback, especially JS-heavy pages and crawl workflows. - **Watch out:** Excellent for extraction; less ideal as the main cheap search backend because search burns credits quickly. - **Key:** `FIRECRAWL_API_KEY` Β· **Source:** https://www.firecrawl.dev/pricing.md, https://docs.firecrawl.dev/features/scrape ### Brave Search β€” *search + news + local* - **🟒 forever / card caveat** β€” $5 in free credits every month, recurring. - **Pricing:** Search $5 / 1k requests; Answers $4 / 1k queries plus token charges; autosuggest/spellcheck priced separately. - **Best for:** Independent broad web index plus news, images and videos. - **Watch out:** Brave docs require a payment card because API usage can generate costs. Treat as recurring free credits, not no-payment onboarding. - **Key:** `BRAVE_API_KEY` Β· **Source:** https://api-dashboard.search.brave.com/documentation/pricing, https://brave.com/search/api/ ### Querit β€” *search (multilingual, real-time)* - **🟒 forever** β€” 1,000 API requests / calendar month on the Free tier. - **Pricing:** Pay-as-you-go US$4 / 1,000 API requests; Enterprise custom. Paid activation requires valid payment info. - **Best for:** Multilingual / real-time search candidate. - **Watch out:** WSP exposes it as search-only. Querit's public privacy terms permit personal data to train, develop and improve AI/ML; see the Privacy & Terms section. - **Key:** `QUERIT_API_KEY` Β· **Source:** https://www.querit.ai/en/subscription, https://www.querit.ai ### SearXNG β€” *search (meta, self-hosted)* - **🟒 self-hosted** β€” Free if you self-host; no vendor API bill. - **Pricing:** Open-source/self-hosted. Your cost is the server, maintenance and upstream engine reliability. - **Best for:** Self-hosting control, no central SaaS vendor account and metasearch across selected engines. - **Watch out:** Self-hosted does not mean no-egress: every enabled engine still receives the query. Public instances are fragile and not production-grade. - **Key:** `searxng.base_url` config (legacy `SEARXNG_INSTANCE_URL` still supported) β€” also the backbone of the WSP 3.x `self_hosted` profile. Β· **Source:** https://docs.searxng.org/admin/installation.html, https://github.com/searxng/searxng ### Keenable β€” *search + extract* - **🟒 keyless rate-limited / authenticated allowance** β€” Keyless public pool plus an API key with a monthly free credit allowance. Allowance size is unpublished. - **Pricing:** Unauthenticated: 1,000 requests/hour, max 10 rps per IP, not credit-metered. Authenticated: 10 rps/org, metered against a monthly free allowance, then paid packages. Search and fetch use separate credit SKUs. 402 when allowance and purchased credits are spent. - **Best for:** Opt-in keyless fallback and an independent search/fetch endpoint. - **Watch out:** Authenticated use is no longer β€œcurrently free.” Keep public/keyless mode opt-in. - **Key:** `KEENABLE_API_KEY` β€” keyless mode additionally requires `KEENABLE_ALLOW_PUBLIC=1` (or `keenable.allow_public: true`); without it Keenable does not auto-route in WSP. Β· **Source:** https://docs.keenable.ai/credits.md, https://docs.keenable.ai/rate-limits ### DonSeTch β€” *search + extract (local stdio MCP)* - **🟒 local / no provider API key** β€” DonSeTch metasearches and extracts through a separately installed local stdio MCP process. - **Pricing:** Open source under AGPL-3.0-only. There is no provider request fee; your cost is the machine, bandwidth, optional browser runtime and maintenance. - **Best for:** Local key-free metasearch, optional browser-backed extraction and a fallback outside commercial provider APIs. - **Watch out:** Not zero-setup, offline or unlimited. Search-engine availability, anti-bot behavior and latency vary; DonSeTch still makes outbound requests. WSP keeps it explicit-only unless you deliberately enable auto-routing. - **Endpoint:** `DONSETCH_BIN=/absolute/path/to/donsetch` Β· **Source / credit:** [DonSeTch](https://github.com/dondai44423/donsetch). Tested with DonSeTch 3.6.1. ### Linkup β€” *search + fetch/extract + citations* - **🟒 recurring credits / eligible accounts** β€” Professional-email signups receive $20 credit; eligible accounts are topped back up to $20 each month. - **Pricing:** Search $0.005–$0.006 / request for standard/fast; deep $0.05–$0.055. Fetch $0.001 without JS, $0.005 with JS. Research $0.25–$2.50 / request. - **Best for:** Clean cited extraction, sourced answers, low-cost fetch. - **Watch out:** Eligibility matters, and Linkup's standard Terms permit Client Data including prompts to be used for training. ZDR/BYOC requires a controlling contract. - **Key:** `LINKUP_API_KEY` Β· **Source:** https://docs.linkup.so/pages/documentation/platform/pricing, https://docs.linkup.so/pages/changelog/usd-pricing --- ## 🟑 One-time / trial credits Good for trying WSP quickly. You burn the credit, then pricing matters. ### You.com β€” *search + extract + research* - **🟑 one-time + You.com day cap** β€” $100 free API credit on a key. You.com also lists 100 queries/day on its own no-key Free plan β€” that is not the WSP path. - **Pricing:** Search $5 / 1k calls; Contents $1 / 1k pages; Research $12 / 1k; Answer API $5 / 1k. - **Best for:** Large starter credit with search, contents/extract and research-style endpoints. - **Watch out:** WSP uses `YOU_API_KEY`, not You.com’s no-key Free plan. Self-serve has no standard no-training or fixed-retention promise. Enterprise ZDR covers Search and Answer, not Contents/Extract or Research. - **Key:** `YOU_API_KEY` Β· **Source:** https://you.com/pricing ### Parallel β€” *search + extract + citations* - **🟒 recurring** β€” Up to 5,000 requests / month free, plus up to $80 signup credit and $5 free credits / month. - **Pricing:** Search turbo/fast $1 / 1k with 10 results; basic/advanced $5 / 1k. Extra results $1 / 1k. Extract $1 / 1k URLs. WSP 4.0.2 defaults Search to `fast`. Task API $5–$2,400 / 1k runs. - **Best for:** Ranked URLs, compressed excerpts, direct page extraction and agent-friendly citations. - **Watch out:** Standard Customer Terms expressly permit queries and outputs to train and improve AI/ML models. From 4.0.2, a configured key also joins automatic routing unless `auto_allow.parallel=false`. Search defaults to `fast`. - **Key:** `PARALLEL_API_KEY` Β· **Source:** https://www.parallel.ai/pricing, https://docs.parallel.ai/getting-started/pricing.md ### Serper β€” *search + maps/news/shopping/scholar + extract* - **🟑 one-time** β€” 2,500 free credits, no credit card. - **Pricing:** Volume packs: $1.00 / 1k at 50k credits down to $0.30 / 1k at 12.5M credits; credits valid 6 months. - **Best for:** Cheap, fast Google-like SERP with rich verticals. - **Watch out:** Extraction runs via Serper’s scrape endpoint and consumes credits like search calls; heavy extraction workloads may fit a dedicated extract provider better. - **Key:** `SERPER_API_KEY` Β· **Source:** https://serper.dev/ ### Exa β€” *search + extract (semantic)* - **🟒 $20 signup + $10 / month** β€” No payment method required. Exa lists $20 on signup and $10 credits every month. - **Pricing:** Search $7 / 1k requests; Contents $1 / 1k pages; Deep Search $12–$15 / 1k; Answer $5 / 1k; extra results above 10 cost $1 / 1k. - **Best for:** Neural/semantic discovery, similar pages, AI-optimized contents. - **Watch out:** Exa's standard privacy terms permit Query Data to be used for training and fine-tuning. Enterprise ZDR is not the self-serve default. - **Key:** `EXA_API_KEY` Β· **Source:** https://exa.ai/pricing, https://exa.ai/docs/reference/billing.md ### SerpBase β€” *search + images/news/maps* - **🟑 one-time** β€” 100 free searches to start, no card; no-key preview capped to 5 organic links. - **Pricing:** Starter Boost $3 / 10k ($0.30 / 1k) expires after one month. Standard prepaid packs $0.50–$0.30 / 1k and do not expire. - **Best for:** Very cheap Google-like SERP fallback. - **Watch out:** Search-only in WSP; explicit/fallback-only rather than a complete search+extract stack. - **Key:** `SERPBASE_API_KEY` Β· **Source:** https://serpbase.dev/, https://serpbase.dev/docs --- ## βšͺ Brokered access / no free-tier claim ### Octen via Monid β€” *source search* - **βšͺ no free-tier claim** β€” A Monid API key with wallet balance is required. - **Pricing:** Access and billing use Monid's prepaid wallet; see Monid for current pricing and terms. - **Best for:** Explicit source search with ranked links, highlights, native recency and include/exclude-domain filters. - **Watch out:** Octen's standard Terms expressly allow ML training on input/output; Monid adds a second recipient without verifiable public legal pages. Explicit-only does not fix that privacy risk. - **Key:** `MONID_API_KEY` Β· **Source:** https://docs.monid.ai/api/run.html, https://octen.ai/terms-of-service ### TinyFish β€” *source search* - **βšͺ account required** β€” Search requests do not consume credits, but your own TinyFish account and API key are required. WSP does not provide, pool, proxy or share TinyFish credentials. - **Pricing:** Published Search limits start at 30 requests/minute for Free and PAYG accounts; verify current eligibility before relying on it. - **Best for:** Direct ranked web/news results with snippets, native freshness, locale and domain filtering. - **Watch out:** TinyFish's standard Terms permit Customer Data to train, fine-tune, evaluate and improve AI/ML models. WSP integrates Search only, omits purpose/fetch, and keeps it explicit-only β€” that reduces scope, not the contractual risk. - **Key:** `TINYFISH_API_KEY` Β· **Sources:** https://docs.tinyfish.ai/search-api, https://www.tinyfish.ai/terms, https://www.tinyfish.ai/privacy-policy --- ## Not registered in WSP 3.x Listed for completeness and pricing context only: since WSP 3.0 these answer endpoints are **not registered** (source-only charter). Configured keys are ignored, not deleted. They are not part of the 15-search / 9-extract WSP surface. ### Perplexity Sonar API β€” *search + cited answers* - **πŸ”΄ none** β€” No durable free API tier found. - **Pricing:** Search/Sonar request fees roughly $5–$22 / 1k depending context/model, plus token pricing. - **Best for:** Search-grounded answers with Perplexity synthesis and citations. - **Watch out:** Not usable through WSP 3.x at all; shown here only for market/pricing context. - **Key:** `PERPLEXITY_API_KEY` Β· **Source:** https://docs.perplexity.ai/docs/getting-started/pricing, https://docs.perplexity.ai/docs/sonar/models/sonar --- ### Kilo Code β€” Perplexity bridge β€” *search + answers* - **βšͺ depends** β€” Depends on your Kilo account and billing. - **Pricing:** Usage is billed through the configured Kilo/provider account; no standalone WSP free tier. - **Best for:** Convenience if you already use Kilo Code and want Perplexity-style answers through that account. - **Watch out:** Not usable through WSP 3.x at all; billing context only. - **Key:** `KILOCODE_API_KEY` Β· **Source:** https://kilo.ai/docs/ai-providers/kilocode, https://kilo.ai/ --- ## Capability & free-tier matrix | Provider | Search | Extract | Free tier | Headline free | Starting price | |---|:--:|:--:|---|---|---| | Tavily | βœ… | βœ… | 🟒 forever | 1,000 API credits / month, recurring, no credit card. | Pay-as-you-go $0.008 / credit | | Firecrawl | βœ… | βœ… | 🟒 forever | 1,000 credits / month, recurring β€” no cost, no card, no hassle. | Hobby $16–19/mo for 5k credits | | Brave Search | βœ… | β€” | 🟒 forever / card caveat | $5 in free credits every month, recurring. | Search $5 / 1k requests | | Querit | βœ… | β€” | 🟒 forever | 1,000 API requests / calendar month on the Free tier. | Pay-as-you-go US$4 / 1,000 API requests | | SearXNG | βœ… | β€” | 🟒 self-hosted | Free if you self-host; no vendor API bill. | Open-source/self-hosted. Your cost is the server, maintenance and upstream engine reliability. | | Keenable | βœ… | βœ… | 🟒 keyless / authenticated allowance | Keyless public pool plus monthly authenticated credit allowance (size unpublished). | Unauthenticated: 1,000/hour, 10 rps/IP, unmetered. Authenticated: monthly free allowance then paid packages. | | DonSeTch | βœ… | βœ… | 🟒 local / no provider key | Separate local stdio MCP process; no provider request fee. | Your machine, bandwidth, browser runtime and maintenance. | | Octen via Monid | βœ… | β€” | βšͺ no free-tier claim | Monid API key and wallet balance required. | See Monid for current pricing and terms. | | TinyFish | βœ… | β€” | βšͺ account required | Search requests do not consume credits; API access required. | Published Search limit starts at 30 requests/minute. | | You.com | βœ… | βœ… | 🟑 one-time | $100 free API credit on a key; You.com also lists 100 queries/day no-key on its own site. | Search $5 / 1k; Answer $5 / 1k | | Parallel | βœ… | βœ… | 🟒 recurring | Up to 5,000 requests / month free, plus up to $80 signup and $5 / month. | Search turbo/fast $1 / 1k; basic/advanced $5 / 1k | | Serper | βœ… | βœ… | 🟑 one-time | 2,500 free credits, no credit card. | Volume packs: $1.00 / 1k at 50k credits down to $0.30 / 1k at 12.5M credits | | Exa | βœ… | βœ… | 🟒 $10 / month | $20 signup + $10 credits / month, no card. | Search $7 / 1k requests | | SerpBase | βœ… | β€” | 🟑 one-time | 100 free searches to start, no card; no-key preview capped to 5 organic links. | Starter Boost $3 / 10k ($0.30 / 1k) expires after one month. Standard prepaid packs $0.50–$0.30 / 1k and do not expire. | | Linkup | βœ… | βœ… | 🟒 recurring credits / eligible accounts | Professional-email signups receive $20 credit; eligible accounts are topped back up to $20 each month. | Search $0.005–$0.006 / request for standard/fast | | Perplexity Sonar API | β€” (not in WSP 3.x) | β€” | πŸ”΄ none | No durable free API tier found. | Search/Sonar request fees roughly $5–$22 / 1k depending context/model, plus token pricing. | | Kilo Code β€” Perplexity bridge | β€” (not in WSP 3.x) | β€” | βšͺ depends | Depends on your Kilo account and billing. | Usage is billed through the configured Kilo/provider account | *Compiled for Web Search Plus Β· 15 active search providers / 9 extract providers (plus two unregistered answer APIs listed for context) Β· pricing snapshot 2026-08-22 Β· privacy/terms audit 2026-08-22. This is a starting map, not a contract.* ## Routing and runtime reference Search providers ## Start with one. Add more when you need them. Web Search Plus works with 15 search providers and can read pages through 9 of them. You only need one provider to get started. **No-card starter** **Tavily** Recurring monthly credits; review the Query Data opt-out before sensitive use. **Cited extract** **Linkup** Fast cited fetch, but standard Terms permit training on Client Data. **Scrape fallback** **Firecrawl** Strong JS extraction; Cloud retention and exact ZDR scope need review. **Starter credit** **You.com** Enterprise Search ZDR does not cover self-serve Contents/Extract. **Local key-free** **DonSeTch** Explicit-only metasearch + page reading through a local stdio MCP process. [Compare privacy, terms + pricing β†’](/providers#privacy-terms) Pricing snapshot: 2026-08-22 Β· privacy/terms audit: 2026-08-22. search + extract search only ### Search + extract Β· 9 providers **Tavily** research search + extract **Exa** semantic discovery search + extract **Linkup** grounded citations search + extract **Firecrawl** scrape-ready content search + extract **Parallel** auto when keyed Β· fast search + extract **You.com** web + code search + extract **Keenable** independent index search + extract **Serper** Google-style results + news/extract search + extract **DonSeTch** local explicit-only stdio search + extract ### Search only Β· 6 providers **Brave** broad web **Querit** multilingual **SerpBase** Google SERP **SearXNG** self-hosted **Octen via Monid** explicit-only **TinyFish** explicit privacy trade-off **Octen via Monid** Octen supplies the source-search results; Monid provides API access and billing. It stays off by default and runs only when you request `provider="octen"` or deliberately opt it into routing. **TinyFish** Direct source search with your own TinyFish API key, but standard Terms permit training and fine-tuning on Customer Data. It stays explicit-only; read the [Privacy & Terms matrix](/providers#privacy-terms) before use. Local & keyless paths ## Local and keyless provider boundaries No commercial provider key does not always mean local. Choose the boundary you actually want: a self-hosted search endpoint, a local Search + Extract sidecar, or an opt-in public keyless service. D local stdio Β· explicit ### DonSeTch Local metasearch and optional browser-backed extraction through a separately installed stdio MCP process. Search Extract Browser / PDF / OCR Runs On your machine as a local stdio MCP process. Egress Public search engines and the target websites you fetch. Routing Explicit-only by default; never silently added to auto or fallback. DONSETCH_BIN=/absolute/path/to/donsetch [DonSeTch](https://github.com/dondai44423/donsetch) Β· independently installed AGPL-3.0-only local provider S self-hosted Β· auto-ready ### SearXNG Your own metasearch endpoint for broad web discovery without a commercial WSP provider account. Search Extract Metasearch Runs At an endpoint you operate and configure in WSP. Egress Upstream search engines selected by your SearXNG instance. Routing Included in the `self_hosted` automatic search pool. searxng.base_url=https://sx.example Search-only in WSP Β· your infrastructure, policy and availability K remote Β· keyless opt-in ### Keenable A public Search + Extract endpoint that can work without an API key, but does not run locally. Search Extract Public tier Runs As a remote public service operated by Keenable. Egress Your query or URL is sent to an unauthenticated third-party endpoint. Routing Off until you explicitly enable the public keyless tier. KEENABLE_ALLOW_PUBLIC=1 Per-IP limits and no SLA Β· convenient fallback, not local infrastructure **Local β‰  offline** SearXNG and DonSeTch keep orchestration and compute under your control, but searches and fetches still reach the public web from your machine. Compare operating cost, privacy and limits in the [provider guide β†’](/providers)