Composio powers 1000+ toolkits, tool search, context management, authentication, and a sandboxed workbench to help you build AI agents that turn intent into action.
Mirrored from https://github.com/Camo67/composio.git
Synced Nov 21, 04:45 UTC
| 1 | import { ChatOpenAI } from '@langchain/openai'; |
| 2 | import { HumanMessage, AIMessage } from '@langchain/core/messages'; |
| 3 | import { ToolNode } from '@langchain/langgraph/prebuilt'; |
| 4 | import { StateGraph, MessagesAnnotation } from '@langchain/langgraph'; |
| 5 | import { Composio } from '@composio/core'; |
| 6 | import { LangchainProvider } from '@composio/langchain'; |
| 7 | // initiate composio |
| 8 | const composio = new Composio({ |
| 9 | apiKey: process.env.COMPOSIO_API_KEY, |
| 10 | provider: new LangchainProvider(), |
| 11 | }); |
| 12 | |
| 13 | // fetch the tool |
| 14 | console.log(`🔄 Fetching the tool...`); |
| 15 | const tools = await composio.tools.get('default', 'HACKERNEWS_GET_USER'); |
| 16 | |
| 17 | // Define the tools for the agent to use |
| 18 | const toolNode = new ToolNode(tools); |
| 19 | |
| 20 | // Create a model and give it access to the tools |
| 21 | const model = new ChatOpenAI({ |
| 22 | model: 'gpt-4o-mini', |
| 23 | temperature: 0, |
| 24 | }).bindTools(tools); |
| 25 | |
| 26 | // Define the function that determines whether to continue or not |
| 27 | function shouldContinue({ messages }: typeof MessagesAnnotation.State) { |
| 28 | const lastMessage = messages[messages.length - 1] as AIMessage; |
| 29 | |
| 30 | // If the LLM makes a tool call, then we route to the "tools" node |
| 31 | if (lastMessage.tool_calls?.length) { |
| 32 | return 'tools'; |
| 33 | } |
| 34 | // Otherwise, we stop (reply to the user) using the special "__end__" node |
| 35 | return '__end__'; |
| 36 | } |
| 37 | |
| 38 | // Define the function that calls the model |
| 39 | async function callModel(state: typeof MessagesAnnotation.State) { |
| 40 | console.log(`🔄 Calling the model...`); |
| 41 | const response = await model.invoke(state.messages); |
| 42 | |
| 43 | // We return a list, because this will get added to the existing list |
| 44 | return { messages: [response] }; |
| 45 | } |
| 46 | |
| 47 | // Define a new graph |
| 48 | const workflow = new StateGraph(MessagesAnnotation) |
| 49 | .addNode('agent', callModel) |
| 50 | .addEdge('__start__', 'agent') // __start__ is a special name for the entrypoint |
| 51 | .addNode('tools', toolNode) |
| 52 | .addEdge('tools', 'agent') |
| 53 | .addConditionalEdges('agent', shouldContinue); |
| 54 | |
| 55 | // Finally, we compile it into a LangChain Runnable. |
| 56 | const app = workflow.compile(); |
| 57 | |
| 58 | // Use the agent |
| 59 | const finalState = await app.invoke({ |
| 60 | messages: [new HumanMessage('Find the details of the user `pg` on HackerNews')], |
| 61 | }); |
| 62 | console.log(`✅ Message recieved from the model`); |
| 63 | console.log(finalState.messages[finalState.messages.length - 1].content); |
| 64 | |
| 65 | const nextState = await app.invoke({ |
| 66 | // Including the messages from the previous run gives the LLM context. |
| 67 | // This way it knows we're asking about the weather in NY |
| 68 | messages: [...finalState.messages, new HumanMessage('what about haxzie')], |
| 69 | }); |
| 70 | console.log(`✅ Message recieved from the model`); |
| 71 | console.log(nextState.messages[nextState.messages.length - 1].content); |