#1122 - Langflow Agent invoking Agent example

Introduction 

Here is a simple example of agent invoking agent. I have 2 agents, each backed by tools running on OIC.

• our Order Processing Agent - hopefully well known to you all
• our CRM agent - responsible for managing customer order history

Our CRM agent has access to the following tool - 

My order processing flow has been augmented as follows - 

Note the new actions - 

• API Request
• Custom Component - Agent -> Webhook Body

The Custom Component transforms the agent output into the required format for the API Request.

So which API are we requesting?

Well, here's the second flow - 

The core is the Agent/MCP Tools that update our CRM system, after the customer's order has been processed.

The initial action is the Webhook, which is what the API Request from the first flow calls. The next 2 actions, marshal the incoming data int the correct format - 

I probably don't need 2 here, just showing my langflow inadequacy.

The Parser output is the input to the Agent - 

Note the 2 Chat Output actions; these will show the request and response from the agent.
  
Now I run the first agent, processing the following order -  

create order with the order number 310397 for Irish customer Riobard Crowley, for the product, iBike, price is 456. Customer email is xxx@gmail.com

I check the second flow in Playground -

I check in OIC - 

Summa Summarum

This scenario was simple to implement, thanks to chatGPT. Net, net - no matter which agent topology you have, OIC tools are there for you. 

#1121 - Mistral AI - kicking the tyres of this compelling European LLM

Introduction

So what is it?

From Wikipedia - 

Mistral AI is a premier French AI startup, founded in 2023 and valued over US$14 billion by 2025, that develops high-performance, cost-efficient, and largely open-weight large language models (LLMs). It offers versatile models—including Mistral Large 3 and Codestral—optimized for tasks like text generation, coding, and multilingual reasoning

A European LLM, excellent!

I'm starting with the Experiment for free option -

I get my api key -

and off I go - 

Postman & Mistral AI

Sanity test passed!

OIC & Mistral AI

Now to OIC, here I create a new project and add a REST (invoke) connection -

I also add a REST (trigger) connection -

I create a simple sync integration - 

Input and output are simple strings.

I add the Mistral invoke - 

Here is the full Response structure - 

{
    "id": "33b51594fea144cf8d4669abd35478e9",
    "created": 1772190410,
    "model": "mistral-large-latest",
    "usage": {
        "prompt_tokens": 11,
        "total_tokens": 1137,
        "completion_tokens": 1126,
        "prompt_tokens_details": {
            "cached_tokens": 0
        }
    },
    "object": "chat.completion",
    "choices": [
        {
            "index": 0,
            "finish_reason": "stop",
            "message": {
                "role": "assistant",
                "tool_calls": null,
                "content": "somecontent"
            }
        }
    ]
}

I complete the Map actions - first Map - 

                      I complete the Map actions - second Map - 

I activate and run - 

                              Second sanity test passed.

Langflow, Mistral AI and OIC Tools

                     Note how I use the Mistral AI component, to allow the Agent to use this LLM.

 

                                      I just need to specify the Model name and Mistral API Key.

The rest you've already seen in earlier posts, OIC based MCP Tools etc.

Summa Summarum

Mistral AI is simple to use, be it from Postman or as the LLM for your agent in Langflow. What's in a name? Mistral - a strong, cold, dry, northwesterly wind that blows from southern France into the Mediterranean Sea, particularly in winter and spring. Originating from the Occitan word for "master," this wind is known for its force, often reaching over 100 km/h

 Courtesy of Wikipedia.

 

#1120 - YAPO Langflow consuming OIC Tools

Introduction 

First the acronym - YAPO - Yet Another Post On.

Think of this post as notes for myself on getting a compelling OIC Agent Tools demo together.

However, it may also be interesting for you.

Let's begin - I've installed Rancher on my windows laptop. Then Langflow -

docker cmd is - docker run -p 7860:7860 langflowai/langflow:latest

This may pull the latest langflow, so, to avoid this you can use - 

docker run -p 7860:7860 --pull=never langflowai/langflow:latest

Note the image is ephemeral, i.e. wiped, once you quit Rancher, so you may want to configure persistent storage.

But back in Langflow, start with a blank flow -

Check out the components - 

Add an Agent, and enter your OpenAI API Key and choose a model - 

I am basing this agent on the one I already created in OIC. Check out my previous posts for intel on that. 

I copy the role and guidelines to the Agent Instructions field -

Add the MCP Tools - 

Here is the connect data for the MCP Server -

{

  "mcpServers": {

    "myOICOrderProcessingToolkit": {

      "disabled": true,

      "timeout": 300,

      "type": "stdio",

      "command": "npx",

      "args": [

        "-y",

        "supergateway",

        "--streamableHttp",

    "yourOICMCPURL",

        "--oauth2Bearer",

"yourOAuthToken"

      ]

    }

  }

}

Add MCP Server - and test a tool - 

Connect MCP Tools to Agent -  

Add Chat Input and connect to Prompt - 

Add Chat Output - 

Then test in Playground - 

Looks good. 

Btw. MCP Server settings can be edited here - 

You see mine is marked as Error - 

The token is no longer valid, so I need to replace it.

Reload the page, and we're back in business - 

Click here to select/de-select tools - 

File Processing Example

So much for the simple example - now let's look at something more complex. I have a post about the OIC AI Agent native action, in which I describe the following scenario - read orders from a file, then, invoke the AI Agent for each order. This we will now prototype in Langflow -

Here is the finished flow - 

This may seem complex, but it isn't. I'll start from left to right - 

Read File

The file I read is as follows - 

[

  {

    "orderNr": "1",

    "customer": "NiallC",

    "product": "iCar",

    "unitPrice": 4899,

    "quantity": 2,

    "email": "xxx@gmail.com"

  },

  {

    "orderNr": "2",

    "customer": "RenateC",

    "product": "iBike",

    "unitPrice": 499,

    "quantity": 2,

    "email": "xxx@gmail.com"

  },

  {

    "orderNr": "3",

    "customer": "LuciaC",

    "product": "iSpy",

    "unitPrice": 123,

    "quantity": 2,

    "email": "xxx@gmail.com"

  }

]

I select this file - 

I run Read File - 

Parse Orders JSON

Puts the input into the required format for the next step.

Trigger Batch

This step receives the parsed input and will invoke the agent in "batch" mode.

Orders - DataFrame

This is the next step, putting the input into the required format for the Loop, which is our iterator -

Loop

The loop is connected as follows - 

Format each order into a message, via Order Row - Message,   the output from the latter is the input to the Agent.

Note the loop back from Agent response to the Loop.

Finally, when the 3 orders have been processed, the Loop is done. Then the output is pushed to Chat Output.

We now just need a Chat Input to kick this off - 

Let's run it in Playground -

I check in OIC, my tools provider - 

All have been processed. 

I now export the flow - by export/import the MCP server settings, LLM api key, Read File settings are wiped.

Ping me, if you want a copy.

I can also invoke the flow via an api call -

Here's an extract from the postman response - 

"results":

{"message":

{"text_key": "text",

"data": {

"timestamp": "2026-02-23 16:07:46 UTC",

      "sender": "Machine",

"sender_name": "AI",

"session_id": "YOUR_SESSION_ID_HERE",

      "context_id": "",

"text": "...Dear NiallC, I regret to

inform you that the product iCar is currently out of stock. We apologize for the inconvenience and appreciate your understanding.

Your order cannot be processed at this time due to the product's unavailability.

If you have any other requests or need assistance, please let me know.

The order with order number 2 for customer RenateC for the product iBike priced at

499 has been successfully created in the SAP system with order number SAP-1234567.

The customer has been notified via email.  

... 

Summa Summarum

I used chatGPT to generate most of the flow. Even it has issues with conflicting data types, apparently LangFlow is somewhat idiosyncratic in this respect. Net, net - we got there in the end!