Both pages carried design rationale and implementation notes that belong
in a review conversation rather than in documentation: why NDJSON is
sent to the browser instead of SSE, why the endpoint is registered in
webapp.py, why the outgoing proxy is bypassed. They are gone, along with
the troubleshooting section.
The section about public instances no longer opens by saying the plugin
is meant for private instances, which contradicted the instructions that
followed it. It now explains why the server, model and key are user
preferences at all -- so that they can be changed at home without
editing settings.yml -- and what that means once other people can reach
the instance.
"AI Summary" is the plugin, "AI summary" is the text it produces, and an
"instance" is always a SearXNG one; a page that also talks about LLM
servers cannot leave that to context.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
An idle LLM server unloads the model and loads it again on the next
request. Nothing of the response is sent while that load runs, so a
request that arrives on a cold server can exceed read_timeout and fail --
reliably making the first search after an idle period the one that does
not get a summary. That same request is what starts the load, so asking
again succeeds.
Only a server that failed to answer is asked again. A server that did
answer with an error status is not: a wrong API key or an unknown model
name does not become right on a second attempt.
The status and body of such an error response are now logged. Until now
every upstream failure looked identical from the outside -- HTTP 502 with
no indication of whether the key, the model name or the network was at
fault.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Furo renders its own page-local table of contents and reports an error
when a page adds one with contents::. No other page in the tree uses
that directive; these two now follow suit.
The plugin is named "AI Summary" where the name of the plugin is meant,
and "AI summary" where the generated text itself is meant. That also
fixes the entry in the plugin navigation, which takes its label from the
page title.
Two references to the configuration page rendered as its title, the bare
YAML key "ai_summary:", which reads as a typo in a sentence. They now
carry explicit link text.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The documentation was written from the perspective of someone who had
just implemented the plugin: it opened with SSRF caveats and settings
keys, and explained decisions rather than usage.
The admin page now starts from what the feature is, followed by a
four-step quickstart (install Ollama, pull a model, configure, restart)
and a troubleshooting section for the failures that actually occur. The
quickstart repeats the default plugins because a plugins: block replaces
that list instead of merging into it -- following the short version of
the instructions would otherwise switch every other plugin off.
The developer page gains a rendered data flow diagram. The two-request
design -- placeholder first, streamed answer second -- is the part of
this plugin that is hard to convey in prose, and the SSE to NDJSON
change is easier to see than to read about.
The module docstring now describes the module and links to both pages,
instead of restating administration guidance.
Add Jason Witty to AUTHORS.rst.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Grounding is on by default, and it sends the top search results along
with the query. Which data leaves the instance depends on where the LLM
server runs -- per the privacy-by-design directive the reader has to be
able to see that from the documentation.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three changes to the ai_summary plugin:
- The *AI Summary* preferences tab is only rendered when the plugin is
activated in settings.yml. An instance that does not offer AI
summaries no longer shows an AI tab at all. The gate is the
administrator setting, not the user opt-out, because the per user
on/off switch lives inside that tab -- hiding it on opt-out would
leave no way to opt back in.
- Users can configure an API key for their own LLM server
(ai_summary_api_key). The administrator key is still only sent to
base_url and the user key only to a server the user configured, so
neither key can be captured through the other. The setting is marked
secret: credentials are excluded from the preferences URL, which users
copy around to transfer or share their preferences.
- Grounding summaries on the search results is now the default; the
extra cost of the longer prompt is moderate.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
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>
* [mod] client/simple: client plugins
Defines a new interface for client side *"plugins"* that coexist with server
side plugin system. Each plugin (e.g., `InfiniteScroll`) extends the base
`ts Plugin`. Client side plugins are independent and lazy‑loaded via `router.ts`
when their `load()` conditions are met. On each navigation request, all
applicable plugins are instanced.
Since these are client side plugins, we can only invoke them once DOM is fully
loaded. E.g. `Calculator` will not render a new `answer` block until fully
loaded and executed.
For some plugins, we might want to handle its availability in `settings.yml`
and toggle in UI, like we do for server side plugins. In that case, we extend
`py Plugin` instancing only the information and then checking client side if
[`settings.plugins`](https://github.com/inetol/searxng/blob/1ad832b1dc33f3f388da361ff2459b05dc86a164/client/simple/src/js/toolkit.ts#L134)
array has the plugin id.
* [mod] client/simple: rebuild static
The settings are currently an untyped key/value structure, whose types are
dynamically built at runtime. The construction process of this structure
is *hand-crafted*.
In the long term, we want a static typing of this structure, based on a standard
tool. The ``msgspec.Struct`` structures are suitable as a standard tool.
This patch makes a first step towards static typing and implements the "brand"
section using ``msgspec.Struct`` structures.
BTW: searx/settings_defaults.py - ``git_url`` and ``git_branch`` had been
removed in aee613d256, this is a leftover.
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
TL;DR; For all the issues that comes with HTTP POST I recommend instance
maintainers to switch to GET and lock the property in the preferences:
```yaml
server:
method: GET
preferences:
lock:
- method
```
We don't want this in the defaults of the SearXNG distributions for the pros vs
cons listed in this discussion:
- https://github.com/searxng/searxng/pull/3619
In customizing it should be decided which plugin modules should be loaded and
which should not.
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
This patch brings two major changes:
- ``Result.filter_urls(..)`` to pass a filter function for URL fields
- The ``enabled_plugins:`` section in SearXNG's settings do no longer exists.
To understand plugin development compile documentation:
$ make docs.clean docs.live
and read http://0.0.0.0:8000/dev/plugins/development.html
There is no longer a distinction between built-in and external plugin, all
plugins are registered via the settings in the ``plugins:`` section.
In SearXNG, plugins can be registered via a fully qualified class name. A
configuration (`PluginCfg`) can be transferred to the plugin, e.g. to activate
it by default / *opt-in* or *opt-out* from user's point of view.
built-in plugins
================
The built-in plugins are all located in the namespace `searx.plugins`.
.. code:: yaml
plugins:
searx.plugins.calculator.SXNGPlugin:
active: true
searx.plugins.hash_plugin.SXNGPlugin:
active: true
searx.plugins.self_info.SXNGPlugin:
active: true
searx.plugins.tracker_url_remover.SXNGPlugin:
active: true
searx.plugins.unit_converter.SXNGPlugin:
active: true
searx.plugins.ahmia_filter.SXNGPlugin:
active: true
searx.plugins.hostnames.SXNGPlugin:
active: true
searx.plugins.oa_doi_rewrite.SXNGPlugin:
active: false
searx.plugins.tor_check.SXNGPlugin:
active: false
external plugins
================
SearXNG supports *external plugins* / there is no need to install one, SearXNG
runs out of the box.
- Only show green hosted results: https://github.com/return42/tgwf-searx-plugins/
To get a developer installation in a SearXNG developer environment:
.. code:: sh
$ git clone git@github.com:return42/tgwf-searx-plugins.git
$ ./manage pyenv.cmd python -m \
pip install -e tgwf-searx-plugins
To register the plugin in SearXNG add ``only_show_green_results.SXNGPlugin`` to
the ``plugins:``:
.. code:: yaml
plugins:
# ...
only_show_green_results.SXNGPlugin:
active: false
Result.filter_urls(..)
======================
The ``Result.filter_urls(..)`` can be used to filter and/or modify URL fields.
In the following example, the filter function ``my_url_filter``:
.. code:: python
def my_url_filter(result, field_name, url_src) -> bool | str:
if "google" in url_src:
return False # remove URL field from result
if "facebook" in url_src:
new_url = url_src.replace("facebook", "fb-dummy")
return new_url # return modified URL
return True # leave URL in field unchanged
is applied to all URL fields in the :py:obj:`Plugin.on_result` hook:
.. code:: python
class MyUrlFilter(Plugin):
...
def on_result(self, request, search, result) -> bool:
result.filter_urls(my_url_filter)
return True
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
Typification of SearXNG
=======================
This patch introduces the typing of the results. The why and how is described
in the documentation, please generate the documentation ..
$ make docs.clean docs.live
and read the following articles in the "Developer documentation":
- result types --> http://0.0.0.0:8000/dev/result_types/index.html
The result types are available from the `searx.result_types` module. The
following have been implemented so far:
- base result type: `searx.result_type.Result`
--> http://0.0.0.0:8000/dev/result_types/base_result.html
- answer results
--> http://0.0.0.0:8000/dev/result_types/answer.html
including the type for translations (inspired by #3925). For all other
types (which still need to be set up in subsequent PRs), template documentation
has been created for the transition period.
Doc of the fields used in Templates
===================================
The template documentation is the basis for the typing and is the first complete
documentation of the results (needed for engine development). It is the
"working paper" (the plan) with which further typifications can be implemented
in subsequent PRs.
- https://github.com/searxng/searxng/issues/357
Answer Templates
================
With the new (sub) types for `Answer`, the templates for the answers have also
been revised, `Translation` are now displayed with collapsible entries (inspired
by #3925).
!en-de dog
Plugins & Answerer
==================
The implementation for `Plugin` and `Answer` has been revised, see
documentation:
- Plugin: http://0.0.0.0:8000/dev/plugins/index.html
- Answerer: http://0.0.0.0:8000/dev/answerers/index.html
With `AnswerStorage` and `AnswerStorage` to manage those items (in follow up
PRs, `ArticleStorage`, `InfoStorage` and .. will be implemented)
Autocomplete
============
The autocompletion had a bug where the results from `Answer` had not been shown
in the past. To test activate autocompletion and try search terms for which we
have answerers
- statistics: type `min 1 2 3` .. in the completion list you should find an
entry like `[de] min(1, 2, 3) = 1`
- random: type `random uuid` .. in the completion list, the first item is a
random UUID
Extended Types
==============
SearXNG extends e.g. the request and response types of flask and httpx, a module
has been set up for type extensions:
- Extended Types
--> http://0.0.0.0:8000/dev/extended_types.html
Unit-Tests
==========
The unit tests have been completely revised. In the previous implementation,
the runtime (the global variables such as `searx.settings`) was not initialized
before each test, so the runtime environment with which a test ran was always
determined by the tests that ran before it. This was also the reason why we
sometimes had to observe non-deterministic errors in the tests in the past:
- https://github.com/searxng/searxng/issues/2988 is one example for the Runtime
issues, with non-deterministic behavior ..
- https://github.com/searxng/searxng/pull/3650
- https://github.com/searxng/searxng/pull/3654
- https://github.com/searxng/searxng/pull/3642#issuecomment-2226884469
- https://github.com/searxng/searxng/pull/3746#issuecomment-2300965005
Why msgspec.Struct
==================
We have already discussed typing based on e.g. `TypeDict` or `dataclass` in the past:
- https://github.com/searxng/searxng/pull/1562/files
- https://gist.github.com/dalf/972eb05e7a9bee161487132a7de244d2
- https://github.com/searxng/searxng/pull/1412/files
- https://github.com/searxng/searxng/pull/1356
In my opinion, TypeDict is unsuitable because the objects are still dictionaries
and not instances of classes / the `dataclass` are classes but ...
The `msgspec.Struct` combine the advantages of typing, runtime behaviour and
also offer the option of (fast) serializing (incl. type check) the objects.
Currently not possible but conceivable with `msgspec`: Outsourcing the engines
into separate processes, what possibilities this opens up in the future is left
to the imagination!
Internally, we have already defined that it is desirable to decouple the
development of the engines from the development of the SearXNG core / The
serialization of the `Result` objects is a prerequisite for this.
HINT: The threads listed above were the template for this PR, even though the
implementation here is based on msgspec. They should also be an inspiration for
the following PRs of typification, as the models and implementations can provide
a good direction.
Why just one commit?
====================
I tried to create several (thematically separated) commits, but gave up at some
point ... there are too many things to tackle at once / The comprehensibility of
the commits would not be improved by a thematic separation. On the contrary, we
would have to make multiple changes at the same places and the goal of a change
would be vaguely recognizable in the fog of the commits.
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
All favicons implementations have been documented and moved to the Python
package:
searx.favicons
There is a configuration (based on Pydantic) for the favicons and all its
components:
searx.favicons.config
A solution for caching favicons has been implemented:
searx.favicon.cache
If the favicon is already in the cache, the returned URL is a data URL [1]
(something like `data:image/png;base64,...`). By generating a data url from
the FaviconCache, additional HTTP roundtripps via the favicon_proxy are saved:
favicons.proxy.favicon_url
The favicon proxy service now sets a HTTP header "Cache-Control: max-age=...":
favicons.proxy.favicon_proxy
The resolvers now also provide the mime type (data, mime):
searx.favicon.resolvers
[1] https://developer.mozilla.org/en-US/docs/Web/HTTP/Basics_of_HTTP/Data_URLs
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
jasonwitty