Questions before you wire it in.
Short, accurate answers. If yours isn't here, reach the team or open an issue on GitHub.
What is slorg?
slorg is an open-source AI search engine that drafts an answer and a knowledge graph before it queries the web. It runs a fixed six-step pipeline on every query and returns a structured object exposing the draft, the graph, the keywords, the scored results, and the token count. It ships on npm as lorg.
Is slorg an agent?
No. slorg does not decide whether to search and does not pick tools at runtime. The same six steps run on every query: draft, knowledge graph, keyword extraction, SearxNG multi-engine search, content fetch, and 0–1 scoring. The trade is less adaptivity for far more legibility and predictability.
What does "thinks before it searches" mean?
It means the first LLM round-trip drafts an answer from training data alone, that draft is turned into a knowledge graph, and the web search keywords are derived from the graph — not from the raw user prompt. Retrieval is conditioned on a structured plan. It is structural, not metacognitive.
How slow is it?
Roughly 5–15 seconds of wall-clock per query on the default configuration. That budget covers three LLM round-trips (steps 1, 3, and 6) plus a content-fetch sweep over the candidate URLs (step 5).
Which models and providers does it work with?
Any OpenAI-compatible endpoint. Provide OPENAI_API_KEY and, optionally, OPENAI_BASE_URL to target another provider or a local model. The default model is gpt-4o-mini; the same model is used for all three LLM steps.
What search backends does it use?
slorg queries the open web through SearxNG. By default it uses Google, Bing, Yahoo, and DuckDuckGo; the engine list and result limit are configurable via LORG_SEARCH_ENGINES and LORG_SEARCH_LIMIT.
How do I run it?
Install with npm install -g lorg for the CLI, or npm install lorg to embed it. Use it as a CLI (lorg "your question"), a Node library (new LorgSearch(key)), or a REST service (lorg server -p 3000). Node 18+ is required.
Can I inspect why a result was chosen?
Yes. The response object exposes every intermediate artifact: the draft answer, the knowledge graph, the derived keywords, and a 0–1 relevance score on each result. Nothing about the plan is hidden.
Is it really open source?
Yes, MIT licensed, built by Skelf-Research. The source is on GitHub at github.com/skelf-research/slorg and the package is published to npm as lorg.