Provider reference

Each section covers one provider in the same order: what it is good for, how to get a key, then the options and quirks. Pass options under providerOptions.<provider> as described in Request options. Common fields (query, limit, url, max_chars) are not repeated below.

Compare providers

The free allowances come from vendor research in October 2026. Vendors change plans often, so confirm on their site. For how limits behave in Polyweb, see Rate limits and budgets.

Provider	search	fetch	Key	Free allowance
brave	yes	no	Required	$5 of credits each month (card required)
exa	yes	yes	Required	$10 monthly grant plus a one-time $10 bonus
firecrawl	yes	yes	Required	1,000 credits per month
keenable	yes	yes	Optional	Keyless is unmetered; a key adds 100,000 requests per month
parallel	yes	yes	Required	$5 of credits each month (card required)
tavily	yes	yes	Optional	1,000 credits per month with a key
tinyfish	yes	yes	Required	search and fetch are free, with rate limits
you	yes	yes	Required	$100 one-time credit plus 100 searches per day

“Optional” means Polyweb ships a built-in public credential, so the provider works with no setup. Add your own key for higher limits.

How to add a key

Provider keys live in the encrypted vault, never in .env. After you get a key from the vendor, add it in the dashboard under Keys and pools, put it in a key pool, and bind the pool to your project. See Provider keys and pools.

Brave

Good for plain web results (search_type: "web") or query-driven context sized for an LLM (search_type: "llm_context"). Brave has no fetch.

Get a key. Create an account in the Brave Search API dashboard and subscribe to a plan, even to use the free credits.

Search options. search_type (web or llm_context), api_version, country, search_lang, ui_lang, offset (0 to 9), freshness, safesearch (off, moderate, strict), spellcheck, text_decorations, result_filter, units (metric, imperial), goggles, extra_snippets, summary, enable_rich_callback, include_fetch_metadata, operators, no_cache, location.

Context options for llm_context only: maximum_number_of_urls, maximum_number_of_tokens, maximum_number_of_snippets, maximum_number_of_tokens_per_url, maximum_number_of_snippets_per_url, context_threshold_mode (strict, balanced, lenient, disabled), enable_local, enable_source_metadata.

Quirks.

• limit becomes count: up to 20 for web, up to 50 for llm_context.
• ui_lang, offset, summary, text_decorations and result_filter work only with web. Sending summary with llm_context fails validation.
• result_filter accepts discussions, faq, infobox, news, query, summarizer, videos, web and locations. goggles takes one to three entries.
• location maps to Brave’s X-Loc-* headers. Give latitude and longitude together.

{
  "query": "fusion power progress",
  "limit": 10,
  "routing": {"mode": "prefer", "providers": ["brave"], "allow_fallback": true},
  "providerOptions": {
    "brave": {"search_type": "llm_context", "maximum_number_of_tokens": 4096}
  }
}

Exa

Good for semantic search, deep research queries and pulling page contents in the same call.

Get a key. Sign up at dashboard.exa.ai. Free credits need no card.

Search options. type (instant, fast, auto, deep-lite, deep, deep-reasoning), category (company, publication, news, personal site, financial report, people), include_domains, exclude_domains, start_published_date, end_published_date, moderation, additional_queries (1 to 10), user_location, compliance (hipaa), output_schema, objective, system_prompt, user_identifier, contents.

contents accepts text (boolean or {max_characters}), highlights (boolean or {query, highlights_per_url, num_sentences, highlights_text_length}), summary (boolean or {query, max_characters}), extras (links, image_links, rich_image_links, rich_links, code_blocks), livecrawl_timeout, max_age_hours, snapshot_as_of, subpages (0 to 100), subpage_target, compliance, user_identifier.

Fetch options. The same extraction fields without the contents wrapper: text, highlights, summary, extras, livecrawl_timeout, max_age_hours, snapshot_as_of, subpages, subpage_target, compliance, user_identifier.

Quirks. limit becomes numResults. deep-lite, deep and deep-reasoning have lower vendor rate limits than instant, fast and auto. Streaming is not exposed.

{
  "url": "https://example.com/report",
  "routing": {"mode": "prefer", "providers": ["exa"], "allow_fallback": true},
  "providerOptions": {"exa": {"text": {"max_characters": 5000}}}
}

Firecrawl

Good for clean markdown from hard pages, with optional browser actions and structured extraction. Search can scrape each result too.

Get a key. Sign up at firecrawl.dev. The free plan needs no card.

Search options. sources (web, images, news, alexandria), domain_tools, tool_detail (compact, summary, full), categories (developer, gov, research, pdf, as strings or {type} objects), include_domains, exclude_domains, tbs, location, country, safe, timeout, ignore_invalid_urls, highlights, enterprise (one-element list: anon or zdr), scrape_options, threat_protection.

Fetch options. formats, only_main_content, only_clean_content, include_tags, exclude_tags, max_age, min_age, headers, wait_for, mobile, skip_tls_verification, timeout, parsers (pdf, image), actions, location, remove_base64_images, block_ads, proxy, store_in_cache, lockdown, check_prompt_injection, redact_pii, profile, zero_data_retention, domain_tools, tool_detail.

Quirks.

• formats takes strings or objects with type, prompt, schema, selector, attributes, full_page, quality, viewport, mode, question and query.
• include_domains and exclude_domains are mutually exclusive. gov cannot be combined with another category.
• threat_protection is a search option. Fetch rejects it as an unknown field.
• scrape_options accepts the fetch fields above for each result.
• Only Search and Scrape are exposed. Crawl, batch scrape, monitors and agents are not.

{
  "url": "https://example.com/docs",
  "routing": {"mode": "prefer", "providers": ["firecrawl"], "allow_fallback": true},
  "providerOptions": {"firecrawl": {"formats": ["markdown"], "only_main_content": true}}
}

Keenable

Good for trying Polyweb with no signup. It works keyless, and a key adds a monthly allowance and removes the hourly cap.

Get a key. Optional. Create one in the Keenable console.

Search options. mode (pro or realtime), site, acquired_after, acquired_before, published_after, published_before, query_time, snippet_max_length (180 to 10000).

Fetch options. live (boolean, request a fresh page instead of an indexed copy), prompt (up to 2000 characters, guides extraction).

Quirks. limit becomes max_results, and max_chars is passed to the vendor. realtime mode and live fetches can cost more than one credit per call.

{
  "query": "solar panel prices",
  "limit": 5,
  "routing": {"mode": "prefer", "providers": ["keenable"], "allow_fallback": true},
  "providerOptions": {"keenable": {"mode": "pro", "snippet_max_length": 500}}
}

Parallel

Good when you can describe the goal in words. You give an objective and keyword search_queries, and Parallel returns excerpts.

Get a key. Create an account at platform.parallel.ai.

Search options. search_queries, objective, mode (turbo, fast, basic, advanced), max_chars_total, session_id, client_model, advanced_settings.

advanced_settings holds source_policy (include_domains, exclude_domains, after_date), fetch_policy (max_age_seconds of at least 600, timeout_seconds, disable_cache_fallback), excerpt_settings.max_chars_per_result, location and max_results.

Fetch options. objective, search_queries, max_chars_total, session_id, client_model, and advanced_settings with fetch_policy, excerpt_settings.max_chars_per_result and full_content (boolean or {max_chars_per_result}). Fetch uses Parallel Extract, and Polyweb turns the single url into the vendor’s URL list.

Quirks.

• A non-zero top-level extra_bundles is rejected with bad_request. See Request options.
• Parallel supports OAuth keys, but that grant only reads the account balance (balance:read). An OAuth key cannot serve search or fetch, and a binding of one cannot enable either.

{
  "query": "perovskite solar cell efficiency",
  "limit": 5,
  "routing": {"mode": "prefer", "providers": ["parallel"], "allow_fallback": true},
  "providerOptions": {
    "parallel": {"objective": "Find recent efficiency records", "mode": "fast"}
  }
}

Tavily

Good for agent-oriented search with depth, topic and date controls. It works keyless for search and fetch.

Get a key. Optional. Sign up at tavily.com. Keys start with tvly-.

Search options. search_depth (basic, advanced, fast, ultra-fast), chunks_per_source (1 to 3), topic (general, news, finance), time_range (day, week, month, year, d, w, m, y), start_date, end_date, days, include_domains, exclude_domains, include_answer (boolean, basic or advanced), include_raw_content (boolean, markdown or text), include_images, include_image_descriptions, include_published_date, filter_by_published_date, include_domains_mode (restrict or prefer), language, filter_by_language, safe_search, auto_parameters, include_favicon, exact_match, country.

Fetch options. extract_depth (basic or advanced), format (markdown or text), query, chunks_per_source, include_images, include_favicon, timeout (1 to 60 seconds).

Quirks. The shared mode supplies the default depth, and an explicit search_depth wins. Conflicting basic and advanced values are rejected. Polyweb sets the endpoint, max_results, URL list and usage reporting itself. Crawl, map and research are not exposed.

{
  "query": "fusion power news",
  "limit": 5,
  "routing": {"mode": "prefer", "providers": ["tavily"], "allow_fallback": true},
  "providerOptions": {"tavily": {"search_depth": "advanced", "topic": "news"}}
}

Tinyfish

Good when you want search and fetch without a per-call charge. Both calls need a key.

Get a key. Create one at agent.tinyfish.ai. No card is needed to start.

Search options. purpose (up to 2000 characters), location, language, include_domains, exclude_domains, recency_minutes, after_date, before_date, domain_type (web, news, research_paper), pub_year_min, pub_year_max, page (zero-based, 0 to 10).

Fetch options. purpose, format (markdown, html, json), links, image_links, iframe_links, ttl, per_url_timeout_ms (1 to 110000), if_none_match, if_modified_since, include_etag_and_last_modified, include_selectors (1 to 20), exclude_selectors (1 to 20), highlights.

highlights accepts query, max_snippets (1 to 20), max_characters (100 to 20000) and include_full_page_text.

Quirks.

• Tinyfish has no result-count parameter. It returns its own page size (currently 10), and limit trims the list.
• recency_minutes cannot be combined with after_date or before_date. pub_year_min must not exceed pub_year_max.
• max_chars trims the content after the fetch. Each fetch sends one URL.

{
  "query": "perovskite efficiency 2026",
  "limit": 5,
  "routing": {"mode": "prefer", "providers": ["tinyfish"], "allow_fallback": true},
  "providerOptions": {"tinyfish": {"domain_type": "research_paper", "pub_year_min": 2023}}
}

You.com

Good for web search where you control how much page text comes back: short highlights or full pages.

Get a key. Create an account at you.com/platform. Keys look like ydc-sk-....

Search options. freshness, offset (0 to 9), country, language, safesearch (off, moderate, strict), knowledge (core), include_domains, exclude_domains, boost_domains, strict_operators, extraction, crawl_timeout (1 to 60), livecrawl (web, news, all), livecrawl_formats (html, markdown).

extraction accepts mode (highlights or full_page), source (blend, cache, fetch), highlights.total_max_tokens (1024 to 8192) and full_page.extraction_formats (html, markdown).

Fetch options. formats (markdown, html, metadata), crawl_timeout (1 to 60), max_age.

Quirks.

• limit becomes count.
• include_domains and exclude_domains cannot be combined. boost_domains cannot be combined with include_domains.
• highlights settings need extraction.mode: "highlights", and full_page settings need extraction.mode: "full_page".
• livecrawl is deprecated by You.com but still accepted.

{
  "query": "open source vector databases",
  "limit": 5,
  "routing": {"mode": "prefer", "providers": ["you"], "allow_fallback": true},
  "providerOptions": {"you": {"extraction": {"mode": "highlights"}}}
}