Telegraph in testing
Coordinating and managing responsibility across different streams, models and developers — so work done by many hands, human and otherwise, still has one record and one owner.
In testing. We run it on our own work every day. If you want to try it on yours, write to shaked@ambitgraph.com.
What it is for
The problem
One agent writing code is manageable. Several, across weeks and models and people, is not. Work happens in chat windows, the reasoning goes when the window closes, and nobody can say afterwards why a thing was done or who decided it.
Streams multiply faster than anyone can hold them in their head.
What Telegraph does
It puts a coordinating layer between you and the work. Every stream reports through it, every decision lands on disk, and every piece of the codebase has someone answerable for it.
You deal with one thing at a time instead of ten windows.
How it works
One thing at a time
You talk to one coordinator. It brings you the decisions that need you and answers the rest itself.
Work is done by seats
Each seat is a session with a name and a job, reporting through the coordinator rather than straight at you.
The record lives on disk
Anything that matters is written down, so it outlives the chat it happened in.
Responsibility rolls up
When someone leaves, the code they owned rolls up to whoever holds it next rather than becoming nobody's.
Across models
Streams are not tied to one assistant. Different models can hold different seats and still report into the same record.
On top of the map
Responsibility is tracked against the same code Omniscient already draws, so ownership has a shape you can look at.
Try it
Telegraph is in testing and not on the package index yet. We use it to run our own work. If you are running more than one stream over one codebase and want to try it, write and say so.