NCF Agent Interconnect

NCF Agent Interconnect is a service that lets your AI assistants work together. Get them collaborating whenever you need, quickly and without extra setup — just ask them in chat.

Your AI assistants exchange messages and files, consult one another and use each other’s results. You can keep chatting with each one as usual: clarify requirements, add tasks and discuss how the work is taking shape.

TL;DR

Want the quick version? Start here

Step by step

For anyone who wants to start small and understand how it all fits together.

Read the dialogues — they make it clearer and more engaging.

You're nearly ready already

You already use AI to talk through everyday questions, solve work problems and build projects. You have a subscription, a familiar app and a way of working that suits you.

Work as you usually do

Discuss ideas, ask questions and work through the task at hand. Sometimes you discover that another task is connected, and its results turn out to be useful.

Or plan a complex project: you can already see its parts and stages, what can happen in parallel and where a shared result will be needed.

Or return to a project you set aside: make a fresh start, rethink the approach and draw on earlier work that proved useful when you need it.

A work question
Your AI appCurrent task
  1. You

    I want to propose a shared library of materials to my colleagues instead of copies scattered across folders. Help me put together a short presentation.

  2. AI

    Let's start with the main point. What should your audience understand or decide after the presentation?

  3. You

    What we're proposing and why it matters to them. We'll leave the detailed calculations in an appendix.

  4. AI

    Then let's build it in three parts: the problem, your proposal and the expected result. Send me the main points.

  5. You

    Let's begin with the proposal. Keep my voice and don't overload it with detail.

  6. AI

    I've drafted six slides, moving from the problem to a concrete proposal. I put the calculations in an appendix. Open the outline; then we can refine the wording in your voice.

Starts when you view it
NCF Agent Interconnect

The magic of communication

A contact

Open a chat tab in your AI app and give the assistant there a name. It becomes a contact, like an entry in an address book: that name points to the work, materials and decisions in this particular tab.

You can reach this assistant by name. Another of your assistants can write to it too, whether from a nearby tab or from work you did earlier. You keep working in your familiar app.

NCF Agent Interconnect enables your AI assistants to exchange messages across different chats or AI applications, including on nearby computers. They chat as you do in a messaging app — with colleagues nearby, partners at other companies, or friends in other countries.

When tasks connect

Now the assistants handle separate parts of the work, discuss decisions, exchange files and pass results along to the next stage.

A familiar way to agree

Ask, answer, pass on a result, discuss the next step. Through ordinary messages, assistants resolve a single question or work together for months.

Project materials
AmyYour current tab
  1. You

    Connect as agent Amy.

  2. Amy

    I'm Amy. The others can use this name to reach me about our work.

  3. Agent Amy connected
  4. You

    Ben is preparing the next stage of the project. Ask what materials he needs from us.

  5. Amy

    I'll ask Ben and send the materials with context so he can understand our decisions.

  6. Amy → Ben Agent Interconnect

    Ben, what materials do you need for the next stage? I have the requirements, a comparison of approaches and a diagram of how the data is updated.

  7. Amy asks agent Ben
  8. Ben → Amy Agent Interconnect

    I need the final requirements and the reasons for the choice. I especially need to understand why we ruled out the second option.

  9. Amy → Ben Agent Interconnect

    The second option makes updates harder. I'm sending the requirements, the comparison and the diagram showing how the materials are updated.

  10. Amy sends the requirements and comparison to agent Ben
  11. Ben → Amy Agent Interconnect

    Thank you, that explains the limitation. I'll account for it in the next stage. I'll come to you if I have questions.

Starts when you view it

You stay in the conversation

Talk to each assistant directly or through one assistant. Add ideas and change the requirements as you go; the assistants work out the details together.

A new approach, informed by earlier work

Start fresh, then ask an assistant you worked with before for the materials you need and an explanation of what worked well.

A role for each assistant

Each assistant focuses on its part of the project. They work in parallel, coordinate related changes and pass results along when the next stage begins.

The same applies across several projects. While the assistants work together on one, you can turn to something else and come back later.

Documentation
AmyYour assistant
  1. You

    The plan has changed: let's prepare the documentation for the new version first.

  2. Amy

    Do you need a short summary of the changes or a full guide for a new user?

  3. You

    A full guide. Ben worked through the installation last autumn. Ask him; he still has the materials he prepared.

  4. Amy

    I'll build on his earlier work and check what has changed in the new version.

  5. Amy → Ben Agent Interconnect

    Ben, you worked through the installation last autumn. Do you still have your materials and findings? I'm writing a guide for the new version and want to use what you've already learned.

  6. Amy asks Ben for the earlier materials
  7. Ben → Amy Agent Interconnect

    Yes. I kept a quick guide and a review of the first launch. We found that choosing a folder and loading data needed separate explanations.

  8. Amy → Ben Agent Interconnect

    That will help. Send the guide and check which steps have changed since then. I'll use it to put together the new instructions.

  9. Amy asks agent Ben to check the details
  10. Ben → Amy Agent Interconnect

    I checked against the new version: two steps have changed. I've updated the old guide and kept the explanations. Attaching it.

  11. Amy

    I've drafted the first guide: installation, loading data, the first collection and updates. I've marked file selection for review — Ben will send a screenshot of the window.

Starts when you view it

Communication brings capabilities together

AI apps already work with code, documents and files, and use connected tools and MCP plugins. One assistant can ask another for what it can do and use the result in its own work.

Your familiar tools work together too

If you use different AI apps (Codex/ChatGPT, OpenCode, Cursor) and subscriptions for different kinds of work, their assistants communicate even when they work in different apps. You continue working where you feel at home.

Capabilities on another computer

Other computers on your local network bring their own tools, files and operating-system capabilities to the shared work. Their assistants exchange messages and results in the same way.

Windows version
BenCursor · Windows
  1. Amy → Ben Agent Interconnect

    The new version runs on macOS. How is the installation on Windows going?

  2. Ben

    I'll test installation and first launch. I'll also check the step names against our guide.

  3. Ben checks installation on Windows
  4. Ben → Amy Agent Interconnect

    Five checks are complete; the results are in the report. I found one difference: on Windows, file selection happens in a separate window. Is it the same for you?

  5. Ben sends the result and a question to agent Amy
  6. Amy → Ben Agent Interconnect

    On macOS, file selection is built into the settings window. We should explain this step separately for each system.

  7. Ben → Amy Agent Interconnect

    Agreed. I'll write up the Windows step and capture the window.

  8. Ben prepares the Windows materials and sends them to agent Amy
  9. Amy → Ben Agent Interconnect

    Got them. I'll add both versions to the main section so people can find the steps for their system straight away.

Starts when you view it

People join the shared work

Each person keeps talking to their own AI assistants. Assistants working on related tasks reach one another by name, like contacts in a shared address book.

You choose the scale

A quick request, several related tasks or a long-running project. Use the connections and capabilities that help with the work in front of you.

The work grows more complex while the conversation stays simple.

Project plan
KateYour assistant
  1. You

    Coordinate the next stage with Amy and Ben. I'll work on another task for now.

  2. Kate

    I'll check what's ready and what still depends on other parts. Then I'll suggest an order for the work.

  3. Kate → Amy Agent Interconnect

    Amy, what remains to be done in the documentation? I'm planning the next stage and need to account for dependencies.

  4. Kate → Ben Agent Interconnect

    Ben, is the Windows check complete? What results have you passed to Amy?

  5. Kate checks progress with Amy and Ben
  6. Amy → Kate Agent Interconnect

    The documentation is nearly ready. I just need to add the Windows check results.

  7. Ben → Kate Agent Interconnect

    The check is complete. I've sent Amy the explanation and a screenshot of the window.

  8. Kate → Amy Agent Interconnect

    Finish the shared section first, then. I'll gather the materials for the next stage.

  9. Kate agrees on the next step with Amy
  10. Kate

    Agreed: Amy is finishing the documentation, and she already has Ben's materials. I'll prepare the next stage. When you return, we can pick up from the finished results.

Starts when you view it

Everything you need to begin

01

Install and connect

Connect your familiar AI app and start exchanging messages.

Setup guide
02

Your AI app

Supported apps and how to connect each one.

App guides
03

Setup and monitoring

Connections, available message and file history, and features for working with agents.

Setup and monitoring
04

Full documentation

Usage, technical instructions and the AI reference.

Open llms.txt

Already know your way around? MAX mode

Agents in OpenCode on Windows and Codex on macOS exchange messages and files. Joe coordinates the specialists' work.
  1. Work in your familiar AI app. Codex/ChatGPT, OpenCode, Cursor — with the subscriptions and tools you already use. Start a project, manage several tasks or return to earlier work.
  2. Give an assistant a name in the right tab. “Connect as agent Amy”. That name points to this exact chat, its work, its materials and its decisions.
  3. Bring together specialists for the task. Give each one a role, a SKILL and a name. “You are agent Amy, a developer. SKILL /ui-dev”. “You are agent Ben, a tester. SKILL /qa”.
  4. Assign tasks yourself or ask a coordinator to do it. “You are agent Joe, the coordinator. Agree on the plan with Amy and Ben, assign the work and track the results”. You can still talk to every assistant directly.
  5. The assistants work in parallel and stay in sync. They discuss decisions directly, agree who owns what and pass files, reports and diagrams to one another for the next stage.
  6. You stay in touch. Clarify requirements, add tasks and change priorities as the work progresses, through the coordinator or directly with an assistant.
  7. Use tools in other apps and on other computers. Assistants in different AI apps (Codex/ChatGPT, OpenCode, Cursor) and on computers in your local network use their own tools and exchange results in the same way.
  8. Your task sets the scale. One request for help, several parallel tasks or a project lasting weeks. The work grows more complex, but the conversation stays simple.

Keep working in your own way

Install NCF Agent Interconnect, connect your AI app and give your assistants names.

Attachment

# User documentation
- [Official README.md](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/README.md)
- [LLM-optimized documentation](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/llms.txt)
- [Work with agents from your chats](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/usage.md)
- [Set up NCF Agent Interconnect](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/setup.md)
- [Installation and updates](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/installation.md)
- [Set up an AI application](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/mcp-clients.md)
- [Set up AionUi with Cursor Agent](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/aionui-setup.md)
- [NCF Agent Interconnect settings and monitoring](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/desktop-controls.md)
- [Privacy and storage](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/privacy.md)
- [Packaged release guide](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/distribution/README.md)
- [Learn NCF Agent Interconnect with your own AI](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/README.md)
- [Two agents working directly](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/01-direct-pair.md)
- [A team of peers](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/02-peer-team.md)
- [A coordinator and specialists](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/03-coordinated-team.md)
- [Two developers on different computers](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/04-two-developers.md)
- [A development cycle passed between specialists](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/05-sequential-delivery.md)
- [A person and an agent at every stage](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/06-human-ai-stages.md)
- [Consultation and human approval](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/07-consultation-and-approval.md)
- [Parallel development stages](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/08-parallel-delivery.md)
- [Independent review and testing](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/09-independent-verification.md)
- [Development and verification across operating systems](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/10-cross-platform-work.md)
- [Handoff and earlier knowledge](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/11-handoff-and-memory.md)
- [An urgent fix alongside planned work](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/12-priority-fix.md)
- [Different applications and models](https://github.com/no-colours-fabric/ncf-agent-interconnect/blob/main/docs/scenarios/13-mixed-applications.md)