websearch-skill

The setting, two ways

websearch-skill reads WEBSEARCH_SEARXNG_URL as the base address of a SearXNG and adds /search, /healthz and /config to it. It sends no key header, so the SpecStream key goes in the address as its last path part. Export it for one shell:

export WEBSEARCH_SEARXNG_URL=https://api.specstream.ai/searxng/<YOUR_SPECSTREAM_KEY>
websearch init --skip-searxng
websearch web-search "open source vector database 2026"

Or put it in the settings file websearch-skill reads from any directory, so each new session has it. An exported variable wins over the file, and the file is outside your repository, so the key stays out of git.

mkdir -p ~/.config/websearch
echo 'WEBSEARCH_SEARXNG_URL=https://api.specstream.ai/searxng/<YOUR_SPECSTREAM_KEY>' >> ~/.config/websearch/.env
websearch init --skip-searxng

With uvx and no install, the commands are uvx websearch-skill init --skip-searxng and uvx websearch-skill web-search "...". Every field of the reply is in the SearXNG migration guide.

What `websearch init --skip-searxng` does

websearch-skill asks agents to run websearch init first in every session. Plain init installs and starts a local SearXNG. With --skip-searxng it starts none, and the doctor still checks the address you set: /healthz, then /config, then one search with format=json.

SpecStream answers /healthz with OK and /config with its one engine, and charges neither. The doctor's test search is one search at $0.001. A full sweep sends one more search for each ddgs provider that returned nothing, and websearch init --quick skips that sweep.

Check the key before you save it

This is the call websearch-skill makes for a search. Run it once with your key and you see the results[] it reads. A wrong key gets a 401.

curl "https://api.specstream.ai/searxng/$SPECSTREAM_KEY/search?q=open+source+vector+databases&format=json"

$1 per 1,000 searches, with nothing to subscribe to

Each web-search with the address set is one SpecStream search at $0.001, and so is the doctor's test search at session start. healthz and config cost nothing. A search that returns no results costs nothing, a call stopped for load is not charged, and 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. 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. There is no limit per key and no daily limit; a busy call comes back with Retry-After and is not charged.

You decide what is kept, on each key

The key sits in an address, so give this tool its own. Create it with spend_cap_cents, the most the key can spend, and an agent that loops stops at the cap you set. With zero_retention, a search made with it keeps no request text and no answer after the reply.

curl -X POST https://api.specstream.ai/v1/account/keys \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" -H "Content-Type: application/json" \
  -d '{"name": "websearch-skill", "scopes": ["account:read", "finds:spend"], "spend_cap_cents": 500, "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 runs Deep Search: a request in plain words returns every match on the live web as rows, at 5¢, 15¢ or $1 a row, with the most a run can cost quoted before it starts.

ddgs, the keyless default

A Python library that runs inside websearch-skill and reads nine providers: Google, Brave, DuckDuckGo, Yandex, Yahoo, Startpage, Mojeek, Wikipedia and Grokipedia. It needs no key and no account. With SpecStream as the second engine, each search also gets the results of a search API that scores 97.9 on SimpleQA Verified, at $0.001.

Self-hosted SearXNG

The metasearch server websearch-skill starts with websearch searxng up: the first run clones SearXNG and builds a virtualenv, about 15 to 30 seconds and a few hundred MB, and the server then runs on your machine. SpecStream answers the same three addresses from https://api.specstream.ai/searxng/<key> with nothing to install or keep running. The whole comparison is on SearXNG alternatives.

Brave Search API

A search API on Brave's own index at $5 per 1,000 requests, with $5 in free credits every month. websearch-skill lists keyed engines such as Brave as planned, so it takes no Brave key today. SpecStream works in it now through the SearXNG address, at $1 per 1,000 searches.

Tavily

A search API billed in credits at $0.008 a credit on pay as you go. websearch-skill lists it as planned too. On SpecStream a search is $0.001 with no credits to count.

Exa

A search API at $7 per 1,000 requests on auto, also on websearch-skill's planned list. SpecStream answers today at $1 per 1,000 searches of up to 10 results.