yantrix

JavaScript/TypeScript SDK for Yantrix agent infrastructure.

Policy enforcement · Semantic memory · Full audit traces — for any JS/TS agent runtime.

npm install yantrix

Quick start

import { configure, run } from "yantrix";

configure({ apiKey: process.env.YANTRIX_API_KEY! });

const result = await run({ intent: "research competitor pricing" });
console.log(result.traceId);     // ytx_run_0ba9b36a
console.log(result.status);      // "completed" | "policy_denied"
console.log(result.totalCostUsd); // 0.156

Client API

import { Yantrix } from "yantrix";

const ytx = new Yantrix({
  apiKey: "ytx_...",
  agentId: "my_agent",
  budgetUsd: 5.0,         // daily cap
  fallbackOnError: true,  // proceed if Yantrix unreachable
});

// Run an intent
const result = await ytx.run({ intent: "summarize latest AI news" });

// Policy check only (no execution)
const policy = await ytx.evaluatePolicy({ action: "send_email", estimatedCostUsd: 0.01 });
if (policy.decision === "deny") throw new Error(policy.reason);

// Memory
await ytx.remember({ key: "research_001", content: "OpenAI cut prices 50%" });
const memories = await ytx.recall({ query: "OpenAI pricing", topK: 3 });
await ytx.forget("research_001");

// Traces
const trace = await ytx.getTrace("ytx_run_0ba9b36a");
const recent = await ytx.listTraces(10);

// Budget
await ytx.setBudget(10.0);
const { today, remaining, limit } = await ytx.getSpend();

LangChain.js

import { wrapTools, YantrixCallbackHandler } from "yantrix/langchain";
import { TavilySearchResults } from "@langchain/community/tools/tavily_search";

// Option 1: wrap tools (full governance)
const tools = wrapTools(
  [new TavilySearchResults()],
  { apiKey: "ytx_...", agentId: "langchain_agent" }
);

// Option 2: callback handler (observability only)
const handler = new YantrixCallbackHandler({ apiKey: "ytx_..." });
const agent = await createReactAgent({ tools, llm, callbacks: [handler] });

Vercel AI SDK

import { wrapTools, yantrixMiddleware } from "yantrix/vercel-ai";
import { generateText, wrapLanguageModel } from "ai";
import { openai } from "@ai-sdk/openai";

// Option 1: wrap tools
const tools = wrapTools(myTools, { apiKey: "ytx_...", agentId: "vercel_agent" });
const { text } = await generateText({ model: openai("gpt-4o"), tools, prompt });

// Option 2: model middleware (intercepts all LLM calls)
const model = wrapLanguageModel({
  model: openai("gpt-4o"),
  middleware: yantrixMiddleware({ apiKey: "ytx_..." }),
});

What happens on every call

Your agent action
      ↓
Policy Engine  — is this allowed? within budget?
      ↓ allow
Semantic Memory — inject relevant past context
      ↓
Execution       — your tool/task runs normally
      ↓
Trace stored    — full audit log
      ↓
Result returned

Full trace visible at trace.yantrix.ai.

Configuration

| Option | Type | Default | Description | |--------|------|---------|-------------| | apiKey | string | — | Your Yantrix API key (ytx_...) | | agentId | string | "js_agent" | Agent identifier for scoping | | budgetUsd | number | 0 | Daily spend cap in USD | | timeout | number | 30000 | Request timeout in ms | | fallbackOnError | boolean | true | Proceed if Yantrix unreachable | | baseUrl | string | https://runtime.yantrix.ai | Override base URL |

Environment variable: YANTRIX_API_KEY (used by global configure() if no key passed).

Links

• Yantrix Runtime API
• Trace Dashboard
• Python SDK
• GitHub