Turn URLs, PDFs, GitHub repos, sitemaps and support content into a grounded, cited support assistant in minutes. Ship it as a chat widget, an iframe, a REST endpoint, or an MCP server for AI agents.
Ingest URLs, PDFs, DOCX, GitHub repos, sitemaps, OpenAPI specs, RSS feeds and YouTube transcripts. Hybrid search — vector + BM25 + reranking.
A floating chat bubble in one line of HTML, an iframe, or a public share link. Origin allow-list, custom accent colour, light / dark theme.
Claude Code, Cursor, Windsurf, Continue.dev and any MCP client can query your bots through the built-in MCP server. Streaming, citations, scoped keys.
Streaming chat, raw similarity search with metadata filters, and full source management — all via OpenAPI 3.1. Generate typed clients in any language.
Anthropic Claude, OpenAI, Gemini — pick the model per bot or per persona. Cost prediction before you spend, full usage metering in millicents.
Workspaces, roles, invitations, audit log. Postgres RLS isolates every row by tenant. Bring teammates, give them scoped API keys, share the bots.
A streaming REST API with source citations, an OpenAPI 3.1 spec for typed SDKs, and a first-class MCP server so coding agents can query your knowledge directly. Scoped API keys, per-key rate limits, and full usage metering included.
// Stream a chat reply with source citations
const res = await fetch(
'https://api.kb.app/bots/<botId>/chat',
{
method: 'POST',
headers: {
Authorization: 'Bearer kb_live_...',
'Content-Type': 'application/json',
},
body: JSON.stringify({
message: 'How do I configure webhooks?',
stream: true,
include_sources: true,
}),
},
);
for await (const event of res.body) {
process.stdout.write(decodeChunk(event));
}Point it at your docs, embed it in one line, and let it answer customers and teammates with cited sources.
Get started free