Servers that require authentication (e.g. vLLM or llama.cpp started with
--api-key, or an LLM server behind an authenticating reverse proxy) can
now be configured with an ai_summary.api_key, sent as "Authorization:
Bearer".
The key is administrator configuration only: there is no preference for
it, and it is only sent to the configured base_url. Users can point the
ai_summary_server preference at a server of their own, and such a server
must not be handed the instance API key -- otherwise every user of the
instance could capture it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Talk to the LLM server via GET /v1/models and POST /v1/chat/completions
(SSE) instead of Ollama's native API. Any OpenAI compatible server now
works (Ollama, vLLM, llama.cpp, LM Studio, Hugging Face TGI, ...);
Ollama serves this API natively, existing setups keep working unchanged.
The Ollama specific keep_alive option is dropped, the ai_summary.grounding
setting is added as instance wide default of the grounding preference.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Add an optional, disabled-by-default plugin that shows an AI generated
summary at the top of the result page, generated by a (local) Ollama
server:
- async: the result page is never delayed; a client plugin streams the
answer (NDJSON over a new /ai_summary endpoint) into a placeholder
answer with a typing indicator, collapsed behind a More button, with
an inline follow-up chat
- trigger: first page of general searches only, skipped when an infobox
or instant answer already answers the query
- grounding (per-user preference): send the top result snippets as
context, the model answers from them instead of its own knowledge
- configuration: new AI Summary preferences tab (server URL, model,
grounding) with instance defaults in a new ai_summary: settings
section; all three preferences can be locked for public instances
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The GSA headers that were introduced in PR #5644 unfortunately no longer
work (#6359).
The Google engines
- google.py
- google_videos.py
do not work anymore either, but we'll leave it in the code for now:
- In google.py, central functions like get_google_info(..) are provided, which
are also used by other modules.
- We will probably need a Google HTML (and video) engine again very soon.
Related:
- https://github.com/searxng/searxng/issues/6359
- https://github.com/searxng/searxng/pull/6364
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
Adds a new google cse engine as an alternative to the currently broken html
google engine.
It works by generating a token which is cached for 1 hour and it only uses the
public CSE id from blackle.com. We do have two others but I think just
hardcoding one will be enough.
Will be broken in 2027 due to google deprecating it but I it will be good while
it lasts.
I found I can get up to 20 results per page. Setting safesearch and language
works and I just copied the traits from the google engine.
Also set old google engine as inactive as it is completely broken for now.
FindFiles.net is a specialized file search engine designed to help you search
files online with precision. Unlike traditional search engines that mainly index
web pages, FindFiles focuses on finding real files on the internet - including
PDFs, documents, archives, videos, datasets, and more. [1]
[1] https://findfiles.net
The code that reads the value of variable `x` from `embed.js`, decodes
it to ASCII and based on that sets `window["tuskheader"]` and `window["tuskkey"]`
is attached below. The only real way to figure out what this is doing is
by stepping through it with the debugger, otherwise it's almost hopeless.
```js
function fe() {
const B = pe => pe.map(_e => String.fromCharCode(_e)).join(''),
ae = window,
o = ae.x;
if (o?.length) {
const pe = o.length / 2;
for (let _e = 0; _e < pe; _e++) ae[B(o[_e])] = B(o[pe + _e]);
ae.x = void 0
}
}
```
Minimal script for testing the engine:
```py
import random
from json import loads
import requests
resp = requests.get("https://api.tusksearch.com/revcontent/embed.js")
data = loads(resp.text[6:])
def _decode(text: list[int]) -> str:
return "".join([chr(x) for x in text])
header = _decode(data[3])
value = _decode(data[4])
resp = requests.get(
"https://api.tusksearch.com/Search/Web?q=test&p=1&l=center&nextArgs=&prevArgs=",
# "https://api.tusksearch.com/Search/Image?q=test&p=1&l=center",
headers={
header: value,
'x-lon': str(random.random() * 90),
'x-lat': str(random.random() * 90),
},
)
print(resp.text)
```
Category for searching personal blogs and websites.
Useful if searching for interesting articles on a topic
rather than the mainstream Wikipedia etc. results.
Drop outdated engine attributes: supported_languages, language_aliases
Complete, normalize and document the type definitions for the engine-module and
engine-class.
For the ``engine.about`` section of the configuration, a type check is performed
based on structure ``searx.enginelib.EngineAbout``.
The property ``engine.about.language`` no longer exists; existing values have
been migrated to ``engine.language``.
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
The results seem to be from Brave (i.e. they are exactly
the same). But it doesn't have any strict rate-limits,
so that's nice.
News support time ranges, but apart from that, unfortunately it doesn't
support any advanced features like safesearch or languages.
T-Online_ is a German news portal.
It gets its web results from Google, image results from Flickr and videos results
from YouTube.
For images and videos, it additionally returns result from its
news catalog. However, for pagination we have to specify the result
type (e.g. either videos from YouTube or from T-Online), so we use
flickr/youtube there instead of tonline because the tonline results
are usually irrelevant.
Add support for https://luxxle.com
Localization is not yet supported because it doesn't seem to work on their
website either, no matter which language I select, it only returns English web
results
RawWeb is a search engine for personal websites / blog posts.
It has its own index and the personal websites were selected
by hand. Results are quite good for what it is imo. [^1]
[^1]: https://github.com/0x2E/RawWeb.org
- adds support for https://reloado.com (german)
- as it has its own index, the results are hit or miss and mostly German,
but still worth integrating imo
Add support for https://startsiden.abcnyheter.no, a netherlandish search engine
that probably uses Google or Bing? idk it also returns English results, but
e.g. ``test`` returns mostly results from netherlands.
S1Search provides various different search services, which all seem
to be somewhat based on Google and Yahoo. The site looks kinda suspicious,
but the results are fine.
You can find a list of their engines by using a subdomain finder like
https://web-toolbox.dev/en/tools/subdomain-lookup and search for `s1search.co`.
Chatnoir is an open source search engine developed by universities, based on
CommonCrawl (and others). It's uncommented by default - we don't want to
overload the universities with bot traffic that targets SearXNG (sad truth why
we can't have nice things anymore)
jasonwitty