KTree

ServicesAI and integrations

AI agents that work inside your ERP, beside your team.

We connect large language models (LLMs) to the ERP, the inbox and the documents you already use, and put AI agents to work on named tasks next to your people. For businesses where staff still chase open items, retype documents and wait for reports, and where a chatbot on the side has not changed that.

Request a quote

From our clients

A real estate company runs AI agents we built alongside its operations team.

What you get

What we hand over

  1. 01

    An MCP server for your ERP or application

    MCP (Model Context Protocol) is the open standard that lets an AI assistant read from, and act in, a business system. We add an MCP server to your ERP, web application or mobile app back end, with a login, so each person sees and changes only what their role allows. Your staff can then ask the system a question in plain words from Claude or ChatGPT, for example which tickets are open.

  2. 02

    Agents for named tasks, with approval steps

    An AI agent is a program that is given a goal and the tools to reach it, and then works through the steps by itself. We have built agents that keep checking items that are due and inform the right people, and an agent that checks signed documents, compares the signature with the previous one, moves the work to the next stage and emails the team when something is wrong. Where the result matters, the agent prepares the work and a person approves it: human-in-the-loop.

  3. 03

    Answers from your own documents and data, with the source

    RAG (retrieval-augmented generation) means the AI first finds the right pages and records, then writes the answer from them. Vector search finds them by meaning, not only by exact words. Staff ask one question across manuals, policies, tickets and ERP records and see where the answer came from. Managers can ask the ERP for numbers and a chart without waiting for a report to be built.

  4. 04

    Tests, limits and a record of what the AI did

    Evals are tests on real examples that measure how often the AI is right; we run them before go-live. Guardrails are the limits on what it may do after go-live. Every step an agent takes is written to a log, and you get a monthly cost report with a spending limit.

How it works

How an agent is put to work safely

An agent is useful only if you can trust what it touches. So it gets the same permissions as a member of staff, it shows its work, and a person signs off before anything important is posted. We describe a set of steps like this, run by agents and people together, as an agentic workflow.

  1. 1The agent connects to your ERP through an MCP server and uses its functions as tools (tool calling). It works with the permissions of a user role, so it can read and change only what that role allows.
  2. 2It does the routine part: reads the documents, checks the open items, matches the records and drafts the entry, the reply or the reminder.
  3. 3A person approves or changes the draft before it is posted or sent. Unclear cases go to a person to check.
  4. 4Every step is logged. The agent is tested on real examples before go-live, and limits on what it may do stay in place afterwards.

Your options

Your options, compared honestly

  • Alternative

    Keep doing it by hand

    Costs nothing to set up and is the right choice when the volume is small. As the volume grows, the checking, chasing and retyping take more of your team’s day, and open items get missed.

  • Alternative

    A generic chatbot or an off-the-shelf AI add-on

    Quick to switch on and good for general writing and questions. It does not know your records or your rules unless someone connects it to your systems and sets what it may do.

  • Alternative

    An AI consultancy without ERP experience

    Strong on models and prototypes. The harder part is usually inside the ERP: the data, the permissions and the workflow stages, and that knowledge has to come from somewhere.

  • With KTree

    Build it with KTree

    The team that builds and supports ERP systems also builds the MCP server and the agents, so the AI works with your real records, roles and approval steps. We start with one process on your own data, and you own the code.

Technology

What we build it with

Ways to work with us

A project, a team, or support for what you already run

Project

We build it end to end

A defined scope, delivered in releases you can test. We own delivery from discovery to go-live.

Best for a new system or a rebuild

Dedicated team

Our engineers, your roadmap

Odoo, React, Node, React Native, AWS and PHP engineers who work as part of your team, month by month.

Best for a product that keeps growing

Support

Keep it running

Maintenance, upgrades and urgent fixes for ERP, web and cloud systems, including ones we did not build.

Best for a system already live

Industries

Where we have done this

Related reading

From our engineers

Before you ask

Questions about AI and agents

Short, straight answers. Anything else, ask in the form below and an engineer will reply.

Is our data sent outside the company?

It depends on the model you choose. With a hosted model, the text the AI needs for each request goes to that provider. If your data must not leave, we run open-weight models (AI models anyone may download and run) on your own servers or private cloud, so no record leaves the company. We write down which data each AI feature uses before it goes live.

What is MCP, and why does it matter?

MCP, the Model Context Protocol, is an open standard for connecting AI models to business systems. One MCP server for your ERP lets any compatible assistant or agent use it safely, so you are not tied to one AI vendor. We have built an MCP server that lets users talk to their ERP from Claude or ChatGPT, and one that publishes website content from any AI tool that supports MCP.

Can AI be added to the ERP we already run?

Usually, yes. We put an MCP server in front of the ERP’s own APIs, so AI agents can read and write records with the permissions you set. The system you run stays in place. On Odoo we set up the AI agents that come with recent versions and add what the standard product does not cover. On ERPNext we add an assistant, and for Microsoft Dynamics 365 we build agents in Microsoft’s Copilot tools.

Who approves what the AI does?

Your people do. For each task we agree with you what the agent may do by itself, such as adding a comment or sending an internal alert, and what must wait for a person, such as posting an entry or emailing a customer. The agent prepares the work, and a manager approves or changes it before anything is sent. Every action is recorded against the user or agent that made it.

How do you know the answers are right?

We test before go-live on real examples from your own data and measure how often the AI is right. Answers from documents show the source page, so a person can check them. Fields the AI is not sure of go to a person. After go-live the log shows every step, so a wrong answer can be traced and the test set extended.

What does it cost to run?

The running cost depends on how many requests are made, how much text each one reads, and which model answers it. We keep it down by sending simple questions to a small model and hard ones to a large model, and by reusing answers that repeat. You see a monthly cost report and can set a spending limit. The build is priced separately, as a small first project on one process.

Request a quote

Tell us what you are building.

A senior engineer replies within one working day.

  1. 1An engineer reads it. Not a sales desk.
  2. 2A short call. To understand the scope and what you run today.
  3. 3A written estimate. Scope, team and timeline on paper.
Address
Sanali Info Park, Road No. 2, Banjara Hills, Hyderabad 500034
What do you need?
Used only to reply to you. No newsletters.