wikai — knowledge base for AI agents

Search returns text.
Agents need structure.

Your domain has rules with exceptions, recommendations with rationale, and ideas that matter because of what sits next to them — not because they share a vector. wikai preserves that structure and exposes it through a typed SDK.

How it works

Ingest. Structure. Query.

01

Ingest

Feed in documents, wikis, runbooks. wikai extracts entities, relationships, and claims.

02

Structure

A domain graph forms — concepts linked by typed edges, grounded in source material.

03

Query

Agents search the graph and get back concepts, rationale, and surrounding context.

SDK

Typed. Predictable.

npm install wikai

agent.ts

import { WikiaiClient } from 'wikai'

const wiki = new WikiaiClient({
  baseUrl: 'https://api.wikai.dev',
  apiKey: process.env.WIKAI_API_KEY,
})

// Seed from your existing knowledge
await wiki.graph.seed({
  content: domainDocument,
  source: 'internal-wiki',
})

// Agent queries with full graph context
const result = await wiki.graph.search({
  query: 'rate limiting for premium tier',
})

result.entities   // matched domain concepts
result.edges      // how they connect
result.evidence   // supporting documents
result.context    // adjacent concepts

What you get

Concepts over chunks.

Domain entities

Terms, rules, patterns, exceptions — linked by typed relationships. Not text fragments.

Rationale chains

Every recommendation stays attached to the reasoning and evidence behind it.

Graph adjacency

Important ideas are often one edge away, not one embedding away.

Full traceability

The graph is queryable, sources are linked, logic is auditable.

Get early access.

Drop your email. We'll reach out when there's something to ship.

Search returns text.Agents need structure.