AgentWatch TypeScript SDK

Drop-in OpenAI SDK wrapper with built-in LLM governance, budget enforcement, and telemetry.

Installation

npm install aw-sdk

Quick Start

import { WatchedOpenAI } from "aw-sdk";

const client = new WatchedOpenAI({
  apiKey: "your-openai-api-key",
  agentwatchApiKey: "aw_live_...",
  agentwatchProject: "checkout-service",
  agentwatchTeam: "payments-eng",
  agentwatchSessionId: "ci-run-123",
  agentwatchSessionBudgetUsd: 2.00,
  agentwatchEnforcementMode: true,
});

// Standard OpenAI API usage
const response = await client.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Refactor this module..." }],
});

Features

• Budget Enforcement: Synchronous pre-call budget checks. Blocks requests exceeding session limits.
• Telemetry: Async background logging of all API calls to AgentWatch.
• Risk Detection: PII and secret scanning on all prompts and completions.
• Streaming Support: Full support for stream: true with post-stream telemetry.
• Fail-Open: If AgentWatch infrastructure is down, requests proceed normally.

Options

| Option | Type | Default | Description | |--------|------|---------|-------------| | agentwatchApiKey | string | required | Your AgentWatch API key | | agentwatchProject | string | null | Project name for attribution | | agentwatchTeam | string | null | Team name for attribution | | agentwatchSessionId | string | crypto.randomUUID() | Session identifier | | agentwatchSessionBudgetUsd | number | null | Per-session budget limit in USD | | agentwatchEnforcementMode | boolean | false | Enable budget enforcement | | agentwatchEnforcementFailOpen | boolean | true | Fail open if budget check unavailable |

License

MIT