AnyModelBot web search without a SearXNG server

The change is one field in Settings

AnyModelBot is "a team of AI bots for Windows, macOS, and Linux, running on your computer", and its bots search the web for research. Its Web search settings say "Free search by default. A key you add is used first", and one of the choices is SearXNG, with a SearXNG address field. Before and after:

# Settings, Web search, before: AnyModelBot's free search
Web search: Auto

# Settings, Web search, after: SpecStream, no instance to run
Web search: SearXNG
SearXNG address: https://api.specstream.ai/searxng/<your SpecStream key>

The setting holds for every bot in the app. AnyModelBot sends no key of its own to a SearXNG address, and it drops anything after a ? when it saves the address, so the SpecStream key goes in the path, and SpecStream reads it from there. A trailing /search is removed too, so .../searxng/<key>/search works the same.

A key in an address, made for this one use

A key in an address can end up in logs, such as a proxy's access log or a debug line. So make a key for AnyModelBot alone under Agents & API in the workspace, with a spending limit. If the address leaks, revoke that key: a revoked key gets a 401 from then on.

curl -X POST https://api.specstream.ai/v1/account/keys \
  -H "Authorization: Bearer $SPECSTREAM_API_KEY" -H "Content-Type: application/json" \
  -d '{"name": "anymodelbot", "scopes": ["account:read", "finds:spend"], "spend_cap_cents": 2000, "zero_retention": true}'

spend_cap_cents is the most the key can spend, here $20, and with zero_retention a search made with it keeps no request text and no answer after the reply.

SearXNG's JSON, so AnyModelBot's parser reads it

AnyModelBot calls <address>/search?q=<query>&format=json with an accept: application/json header and nothing else. SpecStream reads q, answers format=json, and returns page 1. AnyModelBot keeps each result with an http address and reads its title, url, content and publishedDate.

The same call in JavaScript, the way AnyModelBot sends it:

const key = process.env.SPECSTREAM_API_KEY;
const query = "Series A funding announcements this week";
const reply = await fetch(
  `https://api.specstream.ai/searxng/${key}/search?q=${encodeURIComponent(query)}&format=json`,
  { headers: { accept: "application/json" } },
).then((r) => r.json());

for (const result of reply.results.slice(0, 8)) {
  console.log(result.title, result.url, result.content);
}

Every parameter and field of the route is in the SearXNG migration guide, and how SpecStream compares with running SearXNG is on the SearXNG page.

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

AnyModelBot sends no page number, so each search is page 1: 10 results at $0.005, of which it keeps 8 by default. There is no monthly plan and no daily limit. Credit you buy never expires, and a new account starts with $10 of credit, which covers 2,000 searches. A call stopped by a 429 charges nothing.

The most accurate web search API, in about one second

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

Every drop-in call runs the fast search. 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, its 95th percentile stayed between 1.09 and 1.73 s, and it answered 718 of 720 calls.

Search, pages and rows on one key

The same account 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 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, and the most a run can cost is approved before it starts.

Auto, the free default

With no key, AnyModelBot asks Exa's public MCP server, then reads Bing's and DuckDuckGo's result pages, and puts the last one that worked first. A backend that times out, is rate limited or answers with a human check is set aside for 10 minutes. SpecStream answers every search from one address, with no daily limit, at $0.005.

Your own SearXNG instance

The same SearXNG address pointed at a server you run on Docker or Podman, with the json format turned on and a Valkey database for the limiter. Its results are those of the engines it queries. SpecStream answers the same call with no server, at $0.005 a search.

Tavily

AnyModelBot sends the query and up to 20 results to https://api.tavily.com/search. Tavily gives 1,000 credits a month, then bills $0.008 a credit on pay as you go, 1 credit a basic search. SpecStream costs $0.005 a search, and its credit never expires.

Exa

AnyModelBot asks Exa for the auto search with highlights of up to 300 characters. Exa bills $7 per 1,000 auto searches after $10 of free credits a month, and allows 10 search queries per second. SpecStream costs $5 per 1,000 and has no limit per key.

Keenable, Bocha and Zhipu

Each takes its own key in the same settings. AnyModelBot asks Keenable for up to 20 results, Bocha for up to 50 with a summary, and Zhipu's standard search with the query cut to 70 characters. SpecStream reads the whole query and answers it from the live web, at $0.005 a search.