You can now use YaCy as a search tool within an agentic LLM framework,
i.e. using MCP clients in N8N or OpenAI AgentBuilder.
The tool address is /tools
The YaCy retro-pixel-art images had been designed in a time
where displays had not been "retina displays" and
device pixel ratios (DPR) of 2 (everything must be scaled up
by 100% to have the same size).
Because the YaCy images have a fixed-width size, they appear
now tiny compared on how they appeared in the past.
Our 6-pixel font is not readable any more on high-res
displays because with a DPR-2 they are scaled down to 50%.
Experiments with DPR-agnostic auto-scaling of the images did
not show a beautiful result. Instead, we now simply give
the graphics 20% more pixel space and also scale up the font
from 6 pixel to 8 pixel. They still have a nice retro-style
look.
This is a complete re-design of the serverObjects data structure which
holds all data that is submitted during http post requests to YaCy.
Before the change, post attributes had been stored to Strings which
cannot be larger than 2GB. Furthermore, byte[] uploads had been encoded
to b64 Strings to fit into this data structure. Those strings are now
replaced by a new data structure, ChunkedBytes which is an object that
can hold more than 2GB data using a list of byte[] objects. All required
streaming functions are implemented and streaming from http post upload
into this data structure works. The b64 encoding has been removed. The
ZIM and WARC reader make use of the new data structure.
This does a direct scan of the dataset repositories at huggingface. At
this time there is one repo which has the String "YaCy-Pack" inside the
repo name. All repositories that have that String in its name are
listed. A download button exist which downloads the YaCy Pack to the
hold path.
- removed forced setting of the host attribute, let the library do that.
Setting the host attribute caused failure in case of redirects
- removed disabling of cookies, some pages require cookies to accept the
request.
