captar

Runtime budget and policy enforcement SDK for LLM calls.

captar helps you stop expensive or runaway model calls before execution by enforcing:

• spend limits
• max call counts
• repeated-call loop detection
• reserve/commit cost accounting

Install

npm i captar

Quick Start

import OpenAI from "openai";
import { captar } from "captar";

const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });

const client = captar(openai, {
  budget: { limitUsd: 10, window: "day" },
  policy: {
    maxCalls: 100,
    maxRepeatedCalls: 5,
  },
});

const result = await client.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "Hello" }],
});

Status

• Package: Active
• Stability: Early production-ready core
• Provider support: OpenAI-first
• Runtime support: Node.js 20+
• Module formats: ESM + CJS
• Types: Included (.d.ts)

Error Handling

Blocked requests throw CaptarBlockedError with a structured decision object:

• budget_exceeded
• max_calls_exceeded
• loop_detected

API

captar(client, options)

Wraps a provider client and enforces runtime policy.

options.budget

• limitUsd: number
• window: "hour" | "day" | "session"

options.policy (optional)

• maxCalls?: number
• maxRepeatedCalls?: number

options.hooks (optional)

• onEvent(event)
• onBlocked(event)

Development

pnpm install
pnpm --filter captar build
pnpm --filter captar test
pnpm --filter captar lint
pnpm --filter captar typecheck

Repository

Source: https://github.com/8dazo/captor