Getting Started

Integrate Execlave into my project using the JavaScript/TypeScript SDK. Here's what you need to do: 1. Install the SDK:   npm install @execlave/sdk 2. Add this environment variable to my .env file:   EXECLAVE_API_KEY=<my-api-key> 3. Create a file called "execlave.ts" (or .js) in my project's lib/ or utils/ folder with this initialization code: import { Execlave, AgentPausedError, PolicyBlockedError } from '@execlave/sdk'; export const ag = new Execlave({  apiKey: process.env.EXECLAVE_API_KEY!,  baseUrl: process.env.EXECLAVE_BASE_URL || 'https://api.execlave.com',  environment: process.env.NODE_ENV || 'development',  debug: process.env.NODE_ENV !== 'production',}); // Register agent on startup (idempotent — safe to call multiple times)export async function initAgent() {  return ag.registerAgent({    agentId: 'my-agent',    name: 'My AI Agent',    type: 'chatbot',    platform: 'custom',  });} // Wrap any LLM call with automatic tracingexport function traceCall<T>(fn: (input: string) => Promise<T>, agentId = 'my-agent') {  return ag.wrap(fn, { agentId });} // Export error types for catch blocksexport { AgentPausedError, PolicyBlockedError }; 4. In my main application entry point (e.g. index.ts, app.ts, or server.ts), call initAgent(): import { initAgent } from './lib/execlave';await initAgent(); 5. Wrap my LLM calls with BOTH pre-execution policy enforcement AND tracing. The enforcePolicy() call is synchronous and MUST run before the LLM call — it raises PolicyBlockedError if any block-mode policy fires (e.g. prompt injection). import { ag, AgentPausedError, PolicyBlockedError } from './lib/execlave'; async function handleUserMessage(userMessage: string) {  const trace = ag.startTrace({ agentId: 'my-agent' });  trace.setInput(userMessage);   try {    // Pre-execution policy check — MUST run before the LLM call.    // Throws PolicyBlockedError if a block-mode policy is violated.    await ag.enforcePolicy({ agentId: 'my-agent', input: userMessage });     const response = await myLLM.call(userMessage);    trace.setOutput(response).setModel('gpt-4').finish();    return response;  } catch (error) {    if (error instanceof PolicyBlockedError) {      trace.setOutput('[BLOCKED BY POLICY]').finish('error', error.message);      // Decide what the user sees — generic refusal is safest:      return 'Your input was blocked by our content policies.';    }    if (error instanceof AgentPausedError) {      trace.finish('error', error.message);      return 'Service temporarily unavailable.';    }    trace.finish('error', String(error));    throw error;  }} 6. Add graceful shutdown to catch process exit: process.on('SIGTERM', async () => {  await ag.shutdown();  process.exit(0);}); The SDK has zero runtime dependencies and automatically:- Buffers and batch-flushes traces in the background- Respects kill-switch (throws AgentPausedError if agent is paused)- Caches policy enforcement decisions (60s TTL)- Implements circuit breaker (3 failures → 60s backoff) IMPORTANT: tracing alone does NOT block LLM calls. Trace ingestion creates incidents post-hoc. To block requests before they reach the LLM, you MUST call ag.enforcePolicy() and handle PolicyBlockedError as shown above.