@kynth/ai-tools
Kynth Core as ready-made agent tools — parse documents, fight chargebacks, analyze contracts, match POs, screen leads, and run cited research from inside your agent loop. One import for the Vercel AI SDK, one for LangChain.
Every tool is a thin wrapper over the official @kynth/api client: auth, timeouts, and typed errors behave exactly like a direct SDK call, and failed calls never burn credits. Each tool's description carries its exact price, so cost-aware agents can budget calls. The full machine-readable rate card lives at api.kynth.studio/pricing.json.
Install
npm i @kynth/ai-tools @kynth/api
Get a key (and 500 free credits every month, no card) at api.kynth.studio.
Vercel AI SDK
import { generateText } from "ai";
import { KynthCore } from "@kynth/api";
import { kynthTools } from "@kynth/ai-tools";
const client = new KynthCore({ apiKey: process.env.KYNTH_API_KEY! });
const { text } = await generateText({
model,
tools: kynthTools(client),
prompt: "Build a representment packet for this dispute: …",
});
LangChain
import { KynthCore } from "@kynth/api";
import { kynthLangchainTools } from "@kynth/ai-tools/langchain";
const client = new KynthCore({ apiKey: process.env.KYNTH_API_KEY! });
const tools = kynthLangchainTools(client);
const agent = createReactAgent({ llm, tools });
Tools
kynth_chargeback | Dispute details → representment narrative, evidence checklist, win likelihood |
kynth_compare | Two document versions → every material change + risk notes |
kynth_contract | Contract → parties, term, renewal, obligations, risk flags |
kynth_statement | Bank/card statement → normalized transactions |
kynth_po_match | Invoice vs PO vs receipt → 3-way match + discrepancies |
kynth_dunning | Overdue invoice → escalation-appropriate collection sequence |
kynth_screen | Lead + your ICP → web-grounded qualification verdict |
kynth_research | Company/topic → multi-source, citation-backed brief |
kynth_parse | Any document (PDF/image/text) → structured, validated JSON |
kynth_tables | Any document → every table as clean headers + rows |
Document tools accept fileUrl, raw text, or fileBase64 + fileMimeType.
Also available: the MCP server (npx @kynth/api-mcp), the Python SDK, and CrewAI tools.