LangChain search tools

TavilySearch, before and after

langchain-tavily sends every call to self.api_base_url or TAVILY_API_URL, and TavilySearch builds its wrapper with api_base_url when you pass it. The tool, its arguments and the agent that calls it stay as they are. Before and after:

from langchain_tavily import TavilySearch

# reads TAVILY_API_KEY and calls https://api.tavily.com
search = TavilySearch(max_results=5)
search.invoke({"query": "open source vector databases"})

import os
from langchain_tavily import TavilySearch

search = TavilySearch(
    max_results=5,
    tavily_api_key=os.environ["SPECSTREAM_API_KEY"],
    api_base_url="https://api.specstream.ai/tavily",
)
search.invoke({"query": "open source vector databases"})

TavilyExtract takes the same api_base_url. Its pages come back in Tavily's extract format, at $0.0008 a page.

Every class, one argument

ExaSearchRetriever hands exa_base_url to the Exa client as base_url:

import os
from langchain_exa import ExaSearchRetriever

# before: ExaSearchRetriever(k=5)
retriever = ExaSearchRetriever(
    k=5,
    exa_api_key=os.environ["SPECSTREAM_API_KEY"],
    exa_base_url="https://api.specstream.ai/exa",
)

JinaSearch reads its address from the base_url of its JinaSearchAPIWrapper, which adds the query straight after the address with no separator. So the value ends in ?q=:

import os
from langchain_community.tools import JinaSearch
from langchain_community.utilities import JinaSearchAPIWrapper

# before: JinaSearch(), calling https://s.jina.ai/
tool = JinaSearch(search_wrapper=JinaSearchAPIWrapper(
    api_key=os.environ["SPECSTREAM_API_KEY"],
    base_url="https://api.specstream.ai/jina-search?q=",
))

BingSearchAPIWrapper defaults to Microsoft's Bing Web Search API, retired on August 11, 2025. Give it bing_search_url, or set BING_SEARCH_URL:

import os
from langchain_community.utilities import BingSearchAPIWrapper

# before: BingSearchAPIWrapper(), calling https://api.bing.microsoft.com/v7.0/search
bing = BingSearchAPIWrapper(
    bing_subscription_key=os.environ["SPECSTREAM_API_KEY"],
    bing_search_url="https://api.specstream.ai/v7.0/search",
)

PerplexitySearchRetriever has no address argument. It builds the Perplexity client, which reads PERPLEXITY_BASE_URL:

# before: only PPLX_API_KEY was set
export PERPLEXITY_BASE_URL=https://api.specstream.ai/perplexity
export PPLX_API_KEY=$SPECSTREAM_API_KEY

ParallelWebSearchTool takes base_url, whose default is https://api.parallel.ai:

import os
from langchain_parallel import ParallelWebSearchTool

# before: ParallelWebSearchTool()
tool = ParallelWebSearchTool(
    api_key=os.environ["SPECSTREAM_API_KEY"],
    base_url="https://api.specstream.ai/parallel",
)

FirecrawlLoader passes api_url to the Firecrawl client. Its scrape, crawl and map modes all ran on SpecStream:

import os
from langchain_firecrawl import FirecrawlLoader

# before: FirecrawlLoader(url=..., mode="scrape"), calling https://api.firecrawl.dev
loader = FirecrawlLoader(
    url="https://example.com",
    mode="scrape",
    api_key=os.environ["SPECSTREAM_API_KEY"],
    api_url="https://api.specstream.ai/firecrawl",
)
docs = loader.load()

What each address answers

Each address answers in that API's own request and reply form. Every parameter and field is in its guide: Tavily, Tavily Extract, Exa, Jina Search, Bing Web Search v7, Perplexity, Parallel and Firecrawl.

For SearxSearchWrapper and a first LangChain agent with web search, read web search in LangChain.

The Tavily call underneath TavilySearch, from plain Python:

import os, requests

reply = requests.post(
    "https://api.specstream.ai/tavily/search",
    headers={"Authorization": f"Bearer {os.environ['SPECSTREAM_API_KEY']}"},
    json={"query": "open source vector databases", "max_results": 5},
).json()

for result in reply["results"]:
    print(result["title"], result["url"], result["score"])

$5 per 1,000 searches, with no daily limit

A search of up to 10 results costs $0.005 on every address above, and each started block of 10 more adds $0.006. A page from TavilyExtract or a FirecrawlLoader scrape costs $0.0008, and a page with no text is not charged. There is no monthly plan and no daily limit. A busy call gets a 429 with Retry-After and charges nothing.

The most accurate web search API, in about one second

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.

A search call on these addresses runs the fast search. Measured on production on November 9, 2026, its median was 0.82 s at 1 search a second and 1.07 s at 20 a second, the 95th percentile stayed between 1.09 and 1.73 s, and it answered 718 of 720. There is no limit per key and no daily limit.

You decide what is kept, on each key

Create the key with zero_retention and a search made with it keeps no request text and no answer after the reply. spend_cap_cents is the most the key can spend, so an agent in a loop stops at the cap you set.

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

Search, pages and rows on one key

One SpecStream key serves all eight classes. The Extract API reads up to 20 pages or documents a call at $0.80 per 1,000 pages. Deep Search returns every match for a request in plain words as structured rows, at 5¢, 15¢ or $1 a row, with a source URL on every value. The most a run can cost is approved before it starts.

Tavily

Billed in credits: $0.008 a credit on pay as you go, 1 credit a basic search and 2 an advanced one. On a monthly plan, requests stop when the month's credits run out. SpecStream answers the same TavilySearch at $0.005 a search, and credit you buy never expires.

Exa

$7 per 1,000 requests on auto and fast, $4 on lite, and the base price covers the first 10 results. SpecStream answers ExaSearchRetriever at $5 per 1,000 searches of up to 10 results.

Perplexity Search API

$5.00 per 1,000 successful requests. SpecStream costs the same $5 per 1,000 behind PERPLEXITY_BASE_URL and scores 97.9 F-score on all 1,000 SimpleQA Verified questions.

Parallel Search

$1 to $5 per 1,000 requests with 10 results each, and a limit of 600 searches a minute. SpecStream answers ParallelWebSearchTool with no limit per key and no daily limit.

Firecrawl

Sold by monthly plan: Hobby is $19 a month for 5,000 credits, and a search costs 2 credits per 10 results. SpecStream answers FirecrawlLoader with no plan: $0.0008 a scraped page and $0.005 a search.