Three submissions ask for web search documentation that covers deploying the whole stack, and for a fully self-hosted option.
Reports (verbatim, sanitized)
| Page |
Report |
.../object_structure/web_search |
"add docker compose configuration example to deploy the whole system with docker compose up" |
/docs/features/web_search |
"Add (link to or direct) docker configuration documentation to add web search services to the deployment" |
.../object_structure/web_search |
"This setup does not allow for fully locally hosted web search setup. to support fully locally hosted web search you would need to support tools like crawl4ai and a locally hosted reranker. please add documentation to set that up" |
Current state
web_search.mdx was substantially improved by #732, which added searxngSearchOptions, the SSRF/allowedAddresses section, and engine guidance. Two gaps remain:
- No compose example. The SearXNG setup guide is a Docker Desktop click-through. There is no
docker-compose.override.yml fragment that brings up SearXNG alongside LibreChat, even though allowedAddresses and the service-name URL are exactly the parts people get wrong.
- No fully self-hosted path. Per
webSearchSchema in packages/data-provider/src/config.ts, the scraper providers are Firecrawl / Serper / Tavily and the rerankers are Jina / Cohere. Firecrawl and Jina can be self-hosted or given a custom URL (firecrawlApiUrl, jinaApiUrl), but that combination is undocumented, and there is no crawl4ai integration at all.
Item 2 is partly an upstream capability question (does LibreChat want a crawl4ai scraper?) and partly a docs question (can a Firecrawl + Jina self-hosted combination be documented today?). The docs half is actionable now; the crawl4ai request belongs upstream if it is wanted.
Discord feedback IDs: 1526235582680793151, 1526235208259735644, 1526251900801126614.
Three submissions ask for web search documentation that covers deploying the whole stack, and for a fully self-hosted option.
Reports (verbatim, sanitized)
.../object_structure/web_search/docs/features/web_search.../object_structure/web_searchCurrent state
web_search.mdxwas substantially improved by #732, which addedsearxngSearchOptions, the SSRF/allowedAddressessection, and engine guidance. Two gaps remain:docker-compose.override.ymlfragment that brings up SearXNG alongside LibreChat, even thoughallowedAddressesand the service-name URL are exactly the parts people get wrong.webSearchSchemainpackages/data-provider/src/config.ts, the scraper providers are Firecrawl / Serper / Tavily and the rerankers are Jina / Cohere. Firecrawl and Jina can be self-hosted or given a custom URL (firecrawlApiUrl,jinaApiUrl), but that combination is undocumented, and there is no crawl4ai integration at all.Item 2 is partly an upstream capability question (does LibreChat want a crawl4ai scraper?) and partly a docs question (can a Firecrawl + Jina self-hosted combination be documented today?). The docs half is actionable now; the crawl4ai request belongs upstream if it is wanted.
Discord feedback IDs: 1526235582680793151, 1526235208259735644, 1526251900801126614.