Skip to main content
Version: latest (master)

LangChain.js basic agent

Source: node/langchain-js-basic-agent

What this example demonstrates​

A LangChain.js-style agent that integrates Agent Assembly for tool governance. Every tool call is routed through a local policy before execution:

  • get_weather is allowed — it executes and logs output.
  • delete_file is denied — it is blocked at the policy layer.

The governance point: wrapping an agent's tools with withAssembly() enforces policy on every tool call the agent makes.

The framework / library​

LangChain.js-style agent. The example depends only on @agent-assembly/sdk (version 0.0.1-alpha.9.1); it deliberately avoids @langchain/core so it runs fully offline in CI with no API keys.

How it works​

src/tools.ts defines a TOOLS map of mock tool functions. src/policy.ts defines the allow/deny rules and an in-process GatewayClient. src/index.ts wraps the tools with withAssembly and runs each one. The get_weather call passes the policy check and runs; the delete_file call is denied, so withAssembly throws PolicyViolationError before the tool body executes.

Prerequisites & running it​

See Preparing the runtime environment for the shared prerequisites. Then:

cd node/langchain-js-basic-agent
pnpm install
pnpm start

Tests run in offline mode — no gateway or API keys required. For real-provider mode, copy .env.example to .env and fill in your values:

cp .env.example .env
# Edit .env: set AAASM_GATEWAY_URL and optionally OPENAI_API_KEY

Then pnpm start connects to the real gateway and real OpenAI API if configured.

Code walkthrough​

The policy allows the read-only lookup and denies the destructive tool:

src/policy.ts
export const POLICY_RULES: PolicyRule[] = [
  { tool: "get_weather", action: "allow", reason: "Read-only external data lookup — safe to execute." },
  { tool: "delete_file", action: "deny", reason: "Destructive filesystem operations require human approval." },
];

index.ts wraps the tools and handles the denied call:

src/index.ts
const tools = withAssembly(
  {
    get_weather: { execute: async (args) => TOOLS.get_weather(args).output },
    delete_file: { execute: async (args) => TOOLS.delete_file(args).output },
  },
  { gatewayClient: createPolicyGatewayClient(), agentId: "langchain-js-example-agent" }
);

console.log(`  [ALLOW] ${await tools.get_weather.execute({ location: "Taipei" })}`);

try {
  await tools.delete_file.execute({ path: "/etc/hosts" });
} catch (err) {
  if (err instanceof PolicyViolationError) {
    console.log(`  [BLOCKED] ${err.message}`);
  }
}

Notes & caveats​

:::note No @langchain/core dependency The example uses a LangChain.js-style agent without installing @langchain/core, so it runs offline in CI with no API keys. :::

:::caution Gateway connection refused If a real gateway is unreachable, omit AAASM_GATEWAY_URL to use offline mode. :::

Expected behavior​

=== LangChain.js-style Agent Assembly Example ===

Running allowed tool: get_weather
  [ALLOW] Weather in Taipei: 22°C, partly cloudy. [mock]

Running denied tool: delete_file
  [BLOCKED] Tool 'delete_file' blocked: Destructive filesystem operations require human approval.

Done. Tool calls governed by withAssembly + the local policy.