A self-hosted research assistant with auditable sources
You need answers you can defend. slorg exposes the draft, the graph, the keywords, and a score per source, so every answer comes with its own paper trail.
The problem
Hosted answer engines hand you a fluent summary and a list of footnotes, but the planning is invisible. When an answer is wrong you cannot tell whether the model misunderstood the topic, searched for the wrong terms, or was misled by a bad source — the failure is opaque.
With slorg
slorg returns every intermediate artifact. You can read the draft it planned around, inspect the knowledge graph, see exactly which keywords hit the web, and check the 0–1 score on each result. When something looks off, you can point at the step that produced it.
How it flows
- Draft + graph — Inspect the step-1 draft and step-2 graph to confirm the model understood the question.
- Keywords — Read the graph-derived keywords to see what actually got searched.
- Scores — Sort results by their 0–1 relevance score and drop the weak ones.
- Self-host — Run it against your own SearxNG and OpenAI-compatible endpoint for full control.
What you get
- Every answer ships with an inspectable plan
- Failure modes are attributable to a specific step
- No dependence on a hosted black box
try it
Install with
npm install lorg and follow the
quickstart, or read how the pipeline works on the
architecture page.