Platform
The AI-native knowledge platform
Wikantik is built from the ground up for two kinds of readers: the humans who write and browse it, and the AI agents that query, cite, and reason over it. These are the nine platform capabilities that make that possible.
MCP for AI agents
Two native Model Context Protocol servers — 21 read-only retrieval tools and 26 write/analytics tools — so your agents query and curate your wiki directly.
Hybrid retrieval
BM25 + dense embedding similarity fused with weighted Reciprocal Rank Fusion, fail-closed to BM25. Ask in plain language and get the right page.
RAG context bundles
Retrieval assembled for you: ranked, de-duplicated, version-pinned-cited sections and session briefings for coding agents. Context with receipts — never a synthesized answer.
Knowledge graph
LLM-extracted entities with co-mention and typed-relation edges, pgvector-backed embeddings, and a curated inclusion policy. Agents reason over meaning, not HTML.
Ontology & SPARQL
Pages, clusters, tags, and entities projected into a queryable RDF/OWL model — SPARQL, dereferenceable IRIs, and Turtle dumps, governed by a SHACL gate and shared human + AI curation.
Page graph
Real wikilink edges, rename-stable canonical IDs, and cluster-hub membership. Navigate your wiki by shape, not just by search.
Agent-grade content
Every page serves two readers: a clean human read and a token-budgeted machine projection with runbook types, verification metadata, and derived agent hints.
Structural spine
A machine-queryable index of clusters, tags, types, and canonical IDs so agents navigate the wiki by shape — not by guessing at keywords.
Connectors
Six source types — filesystem, web crawler, sitemap, RSS/Atom, Google Drive, GitHub, Confluence — continuously sync into provenance-marked pages that flow into search, RAG, and the knowledge graph.