~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Python academiaanthropicarxivbravedeep-researchencryptionhome-automationhomeserverlocallocal-deep-researchlocal-llmmistralollamaopenaipubmedresearchresearch-toolretrieval-augmented-generationsearxngself-hosted
✨ New Features
- The research form now shows a dismissible warning when SearXNG would silently return no results because its private/localhost instance URL has not been approved by the server operator — including the exact environment line to fix it. A mirror-image warning appears when a local document engine (Paperless-ngx, Elasticsearch) points at a host that does not look private, since document-search queries would then leave the machine. (#5604)
- New env-only operator setting
LDR_SEARCH_PRIVATE_ENGINE_URL_ALLOWLIST: a comma-separated list of exact URL origins (e.g.http://localhost:8080) that public search-engine instance URLs (SearXNG) may use even though they are private/loopback/link-local. This is a finer-grained alternative toLDR_SEARCH_ALLOW_PRIVATE_ENGINE_URLS=true— private fetching is enabled only when the engine's configured URL matches a listed origin, instead of for any private address. Consulted at both save time and runtime; cloud-metadata addresses and non-http(s) schemes remain blocked even if listed, and the setting cannot be edited through the web UI. (#5604)
📝 Other Changes
- Clearer guidance when a self-hosted SearXNG URL is blocked by the private-URL guard: the runtime self-disable now logs at ERROR (was WARNING) and both it and the settings-save rejection name all three operator remedies; docs, installation guides, and
.env.templatenow document the v1.10.3 guard next to everylocalhost:8080example instead of describing the old allow-by-default behavior. (#5604)