Trusera SDK for JavaScript/TypeScript

Beta — This SDK is under active development. Expected GA: April 2026.

The official TypeScript/JavaScript SDK for monitoring AI agents with Trusera. Provides transparent HTTP interception, policy enforcement, and comprehensive observability for AI agent applications.

Key Features

• Transparent HTTP Interception: Zero-code instrumentation via fetch, axios, and undici
• Policy Enforcement: Runtime evaluation against Cedar policies with configurable enforcement modes
• LangChain.js Integration: First-class support for LangChain.js callbacks with optional Cedar enforcement
• ESM + CJS: Dual-format output (built with tsup) -- works in any Node.js or Bun environment
• Rich Event Tracking: Track LLM calls, tool executions, data access, and custom events
• Batched Transmission: Automatic event batching and retry logic
• Type-Safe: Full TypeScript support with strict typing

Installation

npm install trusera-sdk

Quick Start

Basic Usage (HTTP Interceptor)

The simplest way to get started is with the HTTP interceptor:

import { TruseraClient, TruseraInterceptor } from "trusera-sdk";

// Create client
const client = new TruseraClient({
  apiKey: "tsk_your_api_key_here",
  agentId: "my-agent-123",
});

// Install interceptor
const interceptor = new TruseraInterceptor();
interceptor.install(client, {
  enforcement: "log", // or "warn" or "block"
});

// All fetch calls are now automatically tracked
await fetch("https://api.github.com/repos/anthropics/claude");
await fetch("https://api.openai.com/v1/chat/completions", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({ model: "gpt-4", messages: [] }),
});

// Cleanup
await client.close();
interceptor.uninstall();

LangChain.js Integration

Track LLM calls, tool executions, and chain steps automatically:

import { ChatOpenAI } from "langchain/chat_models/openai";
import { TruseraClient, TruseraLangChainHandler } from "trusera-sdk";

const client = new TruseraClient({ apiKey: "tsk_xxx" });
const handler = new TruseraLangChainHandler(client);

const model = new ChatOpenAI({
  callbacks: [handler],
});

await model.invoke("What are the top AI security risks?");

await client.close();

Axios and Undici Interception

The interceptor automatically detects and patches axios and undici if they are installed. No additional configuration is needed:

import axios from "axios";
import { TruseraClient, TruseraInterceptor } from "trusera-sdk";

const client = new TruseraClient({ apiKey: "tsk_xxx" });
const interceptor = new TruseraInterceptor();
interceptor.install(client, { enforcement: "warn" });

// axios requests are now tracked automatically
await axios.get("https://api.openai.com/v1/models");

// undici requests too (if undici is installed)
// import { request } from "undici";
// await request("https://api.openai.com/v1/models");

await client.close();
interceptor.uninstall();

Libraries are detected via require() at install-time. If a library is not installed, it is silently skipped with no errors.

LangChain.js with Cedar Enforcement

Add Cedar policy enforcement to tool and LLM calls:

import { TruseraClient, TruseraLangChainHandler, CedarEvaluator } from "trusera-sdk";

const evaluator = new CedarEvaluator();
await evaluator.loadPolicy(`
  forbid (principal, action == Action::"*", resource)
  when { resource.hostname == "langchain" };
`);

const handler = new TruseraLangChainHandler(client, {
  enforcement: "block",
  cedarEvaluator: evaluator,
});

// Denied tool/LLM calls throw in block mode

Manual Event Tracking

For custom instrumentation:

import { TruseraClient, EventType, createEvent } from "trusera-sdk";

const client = new TruseraClient({ apiKey: "tsk_xxx" });

// Track tool call
client.track(
  createEvent(EventType.TOOL_CALL, "github.search_repos", {
    query: "AI security",
    language: "TypeScript",
  })
);

// Track LLM invocation
client.track(
  createEvent(EventType.LLM_INVOKE, "openai.gpt4", {
    model: "gpt-4",
    tokens: 150,
    temperature: 0.7,
  })
);

// Track data access
client.track(
  createEvent(EventType.DATA_ACCESS, "database.users.read", {
    table: "users",
    query: "SELECT * FROM users WHERE id = ?",
  })
);

await client.close();

Configuration

TruseraClient Options

interface TruseraClientOptions {
  /** API key for authenticating with Trusera (required) */
  apiKey: string;

  /** Base URL for Trusera API (default: https://api.trusera.io) */
  baseUrl?: string;

  /** Agent identifier (auto-registered if not provided) */
  agentId?: string;

  /** Interval in ms to auto-flush events (default: 5000) */
  flushInterval?: number;

  /** Max events per batch (default: 100) */
  batchSize?: number;

  /** Enable debug logging (default: false) */
  debug?: boolean;
}

Interceptor Options

interface InterceptorOptions {
  /** How to handle policy violations (default: "log") */
  enforcement?: "block" | "warn" | "log";

  /** Cedar policy service URL for runtime policy checks */
  policyUrl?: string;

  /** URL patterns to exclude from interception (regex strings) */
  excludePatterns?: string[];

  /** Enable debug logging (default: false) */
  debug?: boolean;
}

Enforcement Modes

The interceptor supports three enforcement modes for policy violations:

log (default)

Logs violations silently, allows all requests to proceed.

interceptor.install(client, { enforcement: "log" });

warn

Logs warnings to console but allows requests to proceed.

interceptor.install(client, { enforcement: "warn" });
// Outputs: [Trusera] Policy violation (allowed): Unauthorized API access

block

Throws an error and blocks the request.

interceptor.install(client, { enforcement: "block" });
// Throws: Error: [Trusera] Policy violation: Unauthorized API access

Advanced Usage

Policy-Based Enforcement

Connect to a Cedar policy service for runtime access control:

interceptor.install(client, {
  enforcement: "block",
  policyUrl: "https://policy.trusera.io/evaluate",
});

// This request will be evaluated against Cedar policies
// If denied, an error will be thrown before the request executes
await fetch("https://api.stripe.com/v1/customers");

Exclude Internal URLs

Prevent interception of internal or infrastructure URLs:

interceptor.install(client, {
  excludePatterns: [
    "^https://api\\.trusera\\.io/.*", // Exclude Trusera API
    "^https://internal\\.company\\.com/.*", // Exclude internal services
    "^http://localhost.*", // Exclude localhost
  ],
});

Agent Registration

Register a new agent programmatically:

const client = new TruseraClient({ apiKey: "tsk_xxx" });

const agentId = await client.registerAgent("my-chatbot", "langchain");
console.log(`Agent registered: ${agentId}`);

// Use the returned ID for future sessions
const client2 = new TruseraClient({
  apiKey: "tsk_xxx",
  agentId: agentId,
});

Manual Flush Control

Control when events are sent:

const client = new TruseraClient({
  apiKey: "tsk_xxx",
  flushInterval: 999999, // Disable auto-flush
  batchSize: 50,
});

// Track events
client.track(createEvent(EventType.TOOL_CALL, "test"));

// Manually flush when ready
await client.flush();

Event Types

The SDK tracks six core event types:

| Event Type | Description | Example Use Cases | | -------------- | ---------------------------------------- | ------------------------------------ | | TOOL_CALL | Tool or function call by the agent | API calls, file operations, searches | | LLM_INVOKE | LLM inference invocation | OpenAI, Anthropic, local models | | DATA_ACCESS | Data read/write operations | Database queries, file I/O | | API_CALL | Outbound HTTP API calls | Third-party APIs, webhooks | | FILE_WRITE | File write operations | Log files, cache writes | | DECISION | Decision points or chain steps | Agent reasoning, workflow branches |

API Reference

TruseraClient

Methods

• constructor(options: TruseraClientOptions): Create a new client
• track(event: Event): void: Queue an event for transmission
• flush(): Promise<void>: Immediately send all queued events
• registerAgent(name: string, framework: string): Promise<string>: Register a new agent
• close(): Promise<void>: Flush events and shutdown client
• getQueueSize(): number: Get current queue size
• getAgentId(): string | undefined: Get the current agent ID

TruseraInterceptor

Methods

• install(client: TruseraClient, options?: InterceptorOptions): void: Install HTTP interceptor for fetch, axios (if available), and undici (if available)
• uninstall(): void: Restore original fetch, eject axios interceptors, and restore undici functions

TruseraLangChainHandler

Methods

• constructor(client: TruseraClient, options?: LangChainHandlerOptions): Create handler for LangChain callbacks with optional Cedar enforcement
• getPendingEventCount(): number: Get count of incomplete events
• clearPendingEvents(): void: Clear all pending events

LangChainHandlerOptions

interface LangChainHandlerOptions {
  enforcement?: "block" | "warn" | "log";
  cedarEvaluator?: CedarEvaluator;
}

When enforcement is "block" and a Cedar policy denies a tool/LLM call, the handler throws an Error. In "warn" mode it logs via console.warn. In "log" mode (default) violations are tracked silently.

Utility Functions

• createEvent(type: EventType, name: string, payload?, metadata?): Event: Create a well-formed event
• isValidEvent(obj: unknown): boolean: Type guard to validate events

Examples

Example: Agent with Tools

import { TruseraClient, TruseraInterceptor, EventType, createEvent } from "trusera-sdk";

const client = new TruseraClient({ apiKey: process.env.TRUSERA_API_KEY });
const interceptor = new TruseraInterceptor();

interceptor.install(client, {
  enforcement: "warn",
  excludePatterns: ["^http://localhost.*"],
});

// Tool execution wrapper
async function executeTool(name: string, input: any) {
  const startTime = Date.now();

  try {
    // Tool logic here
    const result = await fetch("https://api.example.com/tool", {
      method: "POST",
      body: JSON.stringify(input),
    });

    const duration = Date.now() - startTime;

    client.track(
      createEvent(EventType.TOOL_CALL, `tool.${name}.success`, {
        input,
        duration_ms: duration,
      })
    );

    return result;
  } catch (error) {
    client.track(
      createEvent(EventType.TOOL_CALL, `tool.${name}.error`, {
        input,
        error: error.message,
      })
    );
    throw error;
  }
}

await executeTool("search", { query: "AI security" });
await client.close();
interceptor.uninstall();

Example: Multi-Agent System

const agent1 = new TruseraClient({
  apiKey: "tsk_xxx",
  agentId: "planner",
});

const agent2 = new TruseraClient({
  apiKey: "tsk_xxx",
  agentId: "executor",
});

// Track coordination events
agent1.track(
  createEvent(EventType.DECISION, "coordination.delegate", {
    target_agent: "executor",
    task: "web_search",
  })
);

agent2.track(
  createEvent(EventType.TOOL_CALL, "search.execute", {
    delegated_from: "planner",
  })
);

await Promise.all([agent1.close(), agent2.close()]);

Requirements

• Node.js >= 18.0.0
• TypeScript >= 5.0 (for TypeScript users)

License

Apache 2.0 - see LICENSE for details.

Support

• Documentation: https://docs.trusera.io
• Issues: GitHub Issues
• Email: [email protected]

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

Related Projects

• trusera-sdk-python - Python SDK
• ai-bom - AI Bill of Materials scanner
• trusera-platform - Full monitoring platform

Built with care by the Trusera team. Making AI agents safe and observable.