Wiki (navigate-not-cite)

The wiki is a distilled, [[cross-linked]] map of your corpus that you navigate to find evidence. It is the strongest expression of navigate-not-cite: a wiki page is never itself a citation — matching a page resolves down to the real Evidence Units it points at, and only those are cited.

Enable it

Declare the wiki signal. Pages are integrated per-document as you ingest (Karpathy-style incremental upsert) and fully rebuilt on a slow-path refresh.

Python facade — Go and JavaScript have no equivalent

CiteNexus and the ingest / ask /evaluate / reconcile / delete verbs, plus the config layer they read, are the Python facade. The Go and JavaScript packages ship the deterministic core instead — the faithfulness gate, tokenizer, chunker, RRF and Result, at conformance parity — and enter through Ask / ask over an in-memory corpus (quickstart). The Python samples on this page have no port tab because there is nothing to put in it.

rag = CiteNexus("./citenexus-data", embedder=..., generator=...,
                signals=["embedding", "text", "wiki"])
rag.ingest("handbook.pdf")
response = rag.ask("What's our policy on remote work?")

How it works

• Distillation. With an injected wiki distiller (from_config, config.wiki_distill), LLMWikiDistiller produces cross-referenced pages — per-document summaries plus cross-document concept pages — each with [[links]] and sanitized eu_refs back to source units. Without a distiller, pages are deterministic truncation-summaries + keyword lists.
• Retrieval. WikiRetriever matches a light page index and follows one [[link]] hop (at reduced score), then resolves matched pages down to their citable Evidence Units (retrieval signal wiki).
• Maintenance. A lint() pass reports dangling_link, orphan_page, and missing_eu issues so the map stays honest as the corpus changes.

Import from citenexus.wiki (WikiStore, WikiRetriever, LLMWikiDistiller).

It's a single-pass wiki, not a deep DeepWiki

Distillation is one small-model call over a capped (~24k-char) corpus block, producing a flat set of pages with one-hop links. There is no multi-level hierarchy, no iterative refinement, and no page-tree depth beyond that single hop. It’s a genuine navigate-not-cite layer — just don’t expect a deep, multi-level knowledge base yet.