@reminix/langgraph

Reminix agents for LangGraph. Serve any LangGraph agent as a REST API.

Ready to go live? Deploy to Reminix Cloud for zero-config hosting, or self-host on your own infrastructure.

Installation

npm install @reminix/langgraph @langchain/langgraph

This will also install @reminix/runtime as a dependency.

Quick Start

Thread Agent (chat-style)

import { createReactAgent } from '@langchain/langgraph/prebuilt';
import { ChatOpenAI } from '@langchain/openai';
import { LangGraphThreadAgent } from '@reminix/langgraph';
import { serve } from '@reminix/runtime';

const llm = new ChatOpenAI({ model: 'gpt-4o' });
const graph = createReactAgent({ llm, tools: [] });
const agent = new LangGraphThreadAgent(graph, { name: 'my-agent' });
serve({ agents: [agent] });

Workflow Agent (multi-step with interrupt/resume)

import { LangGraphWorkflowAgent } from '@reminix/langgraph';
import { serve } from '@reminix/runtime';

const graph = buildWorkflowGraph(); // your LangGraph compiled graph
const agent = new LangGraphWorkflowAgent(graph, { name: 'my-workflow' });
serve({ agents: [agent] });

Your agent is now available at:

• POST /agents/{name}/invoke - Execute the agent

API Reference

new LangGraphThreadAgent(graph, options?)

Create a LangGraph thread agent for chat-style interactions.

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | graph | CompiledGraph | required | A LangGraph compiled graph | | options.name | string | "langgraph-agent" | Name for the agent (used in URL path) | | options.description | string | "langgraph thread agent" | Description shown in agent metadata | | options.instructions | string | — | System instructions prepended to messages | | options.tags | string[] | — | Tags for categorizing/filtering agents | | options.metadata | Record<string, unknown> | — | Custom metadata merged into agent info |

Returns: LangGraphThreadAgent - A Reminix thread agent instance

The thread agent:

1. Converts incoming messages to LangChain message format
2. Prepends a SystemMessage if instructions is set
3. Invokes the graph with { messages: [...] }
4. Extracts the last AI message from the response
5. Returns it in the Reminix response format

new LangGraphWorkflowAgent(graph, options?)

Create a LangGraph workflow agent for multi-step execution with interrupt/resume support.

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | graph | CompiledGraph | required | A LangGraph compiled graph | | options.name | string | "langgraph-workflow-agent" | Name for the agent (used in URL path) | | options.description | string | "langgraph workflow agent" | Description shown in agent metadata | | options.instructions | string | — | Stored in metadata (not injected into graph execution) | | options.tags | string[] | — | Tags for categorizing/filtering agents | | options.metadata | Record<string, unknown> | — | Custom metadata merged into agent info |

Returns: LangGraphWorkflowAgent - A Reminix workflow agent instance

The workflow agent:

1. Streams the graph and collects per-node outputs as steps
2. Maps GraphInterrupt to pendingAction with status "paused"
3. Accepts resume input to continue interrupted graphs via Command
4. Returns structured {status, steps, result?, pendingAction?} output

Endpoint Input/Output Formats

Thread Agent — POST /agents/{name}/invoke

Request:

{
  "input": {
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }
}

Response:

{
  "output": "Hello! How can I help you today?"
}

Workflow Agent — POST /agents/{name}/invoke

Request:

{
  "input": {"task": "process data"}
}

Response:

{
  "output": {
    "status": "completed",
    "steps": [
      {"name": "fetch_data", "status": "completed", "output": {"records": 10}},
      {"name": "process", "status": "completed", "output": {"summary": "done"}}
    ],
    "result": {"summary": "done"}
  }
}

Resume a paused workflow:

{
  "input": {
    "task": "process data",
    "resume": {"step": "approve", "input": {"approved": true}}
  }
}

Streaming

For streaming responses (thread agent only), set stream: true in the request:

{
  "input": {
    "messages": [{"role": "user", "content": "Hello!"}]
  },
  "stream": true
}

The response will be sent as Server-Sent Events (SSE).

Runtime Documentation

For information about the server, endpoints, request/response formats, and more, see the @reminix/runtime package.

Deployment

Ready to go live?

• Deploy to Reminix Cloud - Zero-config cloud hosting
• Self-host - Run on your own infrastructure

Links

• GitHub Repository
• LangGraph.js Documentation

License

Apache-2.0