LangGraph Binding

TrikHub provides first-class LangChain/LangGraph integration. Convert triks to LangChain tools with one function call.

Quick Start


import { loadLangChainTriks } from '@trikhub/gateway/langchain';
import { ChatOpenAI } from '@langchain/openai';
 
// Load triks and create tools
const { tools } = await loadLangChainTriks();
 
// Bind to model
const model = new ChatOpenAI().bindTools(tools);
 
// Use in your chain
const response = await model.invoke([
  { role: 'user', content: 'Search for AI articles' }
]);

Using Passthrough Content

The recommended way to integrate with LangGraph is to use an onPassthrough callback to handle content that should bypass the agent.

To prevent prompt injection, passthrough content should be handled separately and kept outside of the main Agent’s history.


import { loadLangChainTriks } from '@trikhub/gateway/langchain';
 
let lastPassthroughContent = '';
 
const { tools, gateway, loadedTriks } = await loadLangChainTriks({
  // Handle passthrough content
  onPassthrough: (content) => {
    lastPassthroughContent = content;
  },
});
 
const model = new ChatOpenAI().bindTools(tools);
 
// ...
// Regular model invocation. Make sure to reset `lastPassthroughContent` on each new cycle
// ...
 
if (lastPassthroughContent) {
  console.log('Delivering directly to user:', lastPassthroughContent);
}
 
const assistantMessage = result.messages[result.messages.length - 1];
console.log(`Assistant response: ${assistantMessage.content}\n`);

Options

Option	Type	Description
`onPassthrough` / `on_passthrough`	`(content) => void`	Callback for passthrough content
`debug`	`boolean`	Enable debug logging

Advanced: Manual Tool Creation

For more control, use createLangChainTools / create_langchain_tools directly:


import { TrikGateway } from '@trikhub/gateway';
import { createLangChainTools } from '@trikhub/gateway/langchain';
 
const gateway = new TrikGateway();
await gateway.loadTriksFromConfig();
 
// Manual session management
const sessions = new Map<string, string>();
 
const tools = createLangChainTools(gateway, {
  getSessionId: (trikId) => sessions.get(trikId),
  setSessionId: (trikId, sessionId) => sessions.set(trikId, sessionId),
  onPassthrough: (content) => console.log(content),
  debug: true,
});

Tool Naming

Trik actions become LangChain tools with normalized names:

Trik Action	LangChain Tool Name
`@demo/article-search:search`	`demo_article_search__search`

To parse tool names back:


import { parseToolName } from '@trikhub/gateway/langchain';
 
const { trikId, actionName } = parseToolName('demo_article_search__search');
// trikId: '@demo/article-search'
// actionName: 'search'

Next Steps

Check a complete Agent & Trik integration in the Quickstart
Learn how to create your own triks