LibreChat web search

Two URLs and one key in the config you already have

LibreChat reads its web search settings from librechat.yaml, and each value can point to a variable in .env. SpecStream answers SearXNG's JSON at /searxng and Firecrawl's scrape at /firecrawl/v2/scrape, so both providers LibreChat already knows run on SpecStream.

# librechat.yaml
webSearch:
  searchProvider: searxng
  scraperProvider: firecrawl
  searxngInstanceUrl: '${SEARXNG_INSTANCE_URL}'
  firecrawlApiUrl: '${FIRECRAWL_API_URL}'
  firecrawlApiKey: '${FIRECRAWL_API_KEY}'
  jinaApiKey: '${JINA_API_KEY}'   # your reranker stays as it is

# .env
SEARXNG_INSTANCE_URL=https://api.specstream.ai/searxng/YOUR_SPECSTREAM_KEY
FIRECRAWL_API_URL=https://api.specstream.ai/firecrawl
FIRECRAWL_API_KEY=<YOUR_SPECSTREAM_KEY>

The key goes in the SearXNG address, so leave SEARXNG_API_KEY empty. LibreChat adds /search to the instance URL and /v2/scrape to the Firecrawl URL. These are the two calls it sends:

# The search LibreChat sends: it adds /search to the instance URL
curl "https://api.specstream.ai/searxng/$SPECSTREAM_API_KEY/search?q=latest+LibreChat+release&format=json"

# The page read LibreChat sends: it adds /v2/scrape to the Firecrawl URL
curl -X POST https://api.specstream.ai/firecrawl/v2/scrape \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" -H "Content-Type: application/json" \
  -d '{"url": "https://www.librechat.ai/docs"}'

LibreChat blocks private and loopback addresses for web search, and https://api.specstream.ai is public, so it needs no entry in allowedAddresses.

Let each user bring a key

Leave SEARXNG_INSTANCE_URL out of .env and LibreChat asks each user for it in the chat interface. Each user then pastes https://api.specstream.ai/searxng/ followed by their own SpecStream key, and each search is billed to that key.

Already on Tavily? Change one URL

SpecStream answers Tavily's search call in Tavily's JSON. Point TAVILY_SEARCH_URL at SpecStream and send the SpecStream key in TAVILY_API_KEY. The URL is the whole endpoint, so it ends in /search.

# librechat.yaml
webSearch:
  searchProvider: tavily
  tavilySearchUrl: '${TAVILY_SEARCH_URL}'

# .env
TAVILY_SEARCH_URL=https://api.specstream.ai/tavily/search
TAVILY_API_KEY=<YOUR_SPECSTREAM_KEY>

$0.005 a search and $0.0008 a page

A search of up to 10 results costs $0.005, and the SearXNG route always returns 10. Each page the scraper reads costs $0.0008, which is $0.80 per 1,000 pages, and a page with no text costs nothing. A call stopped by a 429 charges nothing.

There is no monthly plan, no credit pack that expires and no reset on the first of the month. Credit you buy never expires.

The most accurate web search API in the world

On SimpleQA Verified the SpecStream Search API scores 97.9 F-score on all 1,000 official questions, 978 of them correct. On WideSearch it scores 64.0 item F1 and 43.4 row F1 on all 200 official tasks. The Extract API scores 0.908 F1 on WCXB. Every score and its method is on the benchmarks page.

About one second a search, measured under load

Every drop-in route runs on the fast search. On November 9, 2026 production took 1, 5, 10 and 20 searches a second through it and answered 718 of 720. The median was 0.82 s at 1 search a second and 1.07 s at 20 a second, and the 95th percentile stayed between 1.09 and 1.73 s. One account can send up to 8,800 searches a minute; a call over that gets a 429 with Retry-After and is not charged.

You decide what is kept, on each key

Give LibreChat a key with zero_retention and a search made with it keeps no request text and no answer after the reply. It is not in the history and its log lines carry neither. The charge keeps the account, the key, the request ID, the price and the time. spend_cap_cents is the most the key can spend.

curl -X POST https://api.specstream.ai/v1/account/keys \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" -H "Content-Type: application/json" \
  -d '{"name": "production", "scopes": ["account:read", "finds:spend"], "spend_cap_cents": 500000, "zero_retention": true}'

Search, pages and rows on one key

The same key reads up to 20 pages or documents a call with the Extract API at $0.80 per 1,000 pages, and a page with no text is not charged. Deep Search takes a request in plain words and returns every match on the live web as structured rows, at 5¢, 15¢ or $1 a row, with a source URL on every value.

Serper

Search sold in credit packs: $1.00 per 1,000 in the pack of 50,000, with credits valid for 6 months and 2,500 free queries to start. LibreChat fixes Serper's search address in its code. SpecStream searches the live web at $5 per 1,000 with credit that never expires, and reads the pages too.

Tavily

Search and extract billed in credits at $0.008 each on pay as you go, with 1,000 free credits a month. A basic extract costs 1 credit for every 5 successful URLs. SpecStream answers Tavily's own search call at https://api.specstream.ai/tavily/search at $0.005 a search.

Firecrawl

The scraper LibreChat names first. The free plan reads 1,000 pages a month and the Hobby plan costs $19 a month. Set FIRECRAWL_API_URL to https://api.specstream.ai/firecrawl and each page costs $0.0008, from the same key as your searches.

Keenable

A keyless search and fetch service. Its public endpoints are rate limited, and a key lifts the cap. SpecStream has no shared public limit: one account sends up to 8,800 searches a minute, and the search scores 97.9 F-score on all 1,000 SimpleQA Verified questions.