The future is multiplayer

the short story:

In the future we are going to be checking in all of our conversations, I have started trying to figure out what that might be shaped like with loch. I would love to see other people’s take on the same idea.

the slightly less short story:

It is inevitable that multiplayer agent chats are the future of software development. In my private harness I have had multiplayer chats for months. It unlocks a few very useful tools:

1. Being able to ask follow up/design questions directly in-context of the original creator (the agent can act as a proxy for the person who implemented it)
2. Being able to pick up investigations / projects where other people left off
3. Being able to make slight tweaks to someone’s implementation, without needing to involve them (fixing an import path or a styling issue in-context without bothering the original submitter)
4. Being able to do other things I haven’t thought of yet

The more I feel around the edges of this problem, the more it feels like we are going to need a git-like tool for managing conversations. Software has always worked best by building sharp, specialized software at each layer of abstraction (rather than single vertically integrated monoliths)

The tool is going to need to do these basic things:

1. Store conversation history indefinitely
   1. Both sides of the conversation
   2. Enough metadata to be able to attribute every message (action) in the system to an actor
2. Allow for N-actors in a conversation
   1. Conversation participants can be agents or humans
3. Allow for forking conversations
   1. Two actors should be able to continue a conversation in parallel, without causing any immediate problems
   2. Conversations will need to be mergeable
      1. Multiple merging strategies will be supported
4. Allow for storing file / binary artifacts alongside conversations & tools
   1. Unclear if git is suitable for this half of the task

The tool should not:

1. Be an editor or chat harness itself
   1. The tool exists to manage other chat histories, not to run them itself
   2. The tool is primarily used to handle the business logic of the chat history manipulation, not to be a human based interface on top of that
      1. With the exception that some choices will need to be made in favor of the upstream UX
2. Have a favoritism or opinion on harness (allow any harness to be used as long as it has a binary log format)

Requirements last edited 11/5/2026

Using this set (probably evolving over time) of requirements I am spinning up a new project called loch. Loch will be an exploration into what a tool shaped like this would be. If you are still reading this message then loch is still experimental, and should not be used for anything other than curiosity.