Elixir as memory layer

02 Aug 2026

Injection of context on the go with concurrency, not afraid of breaking output
Imagine there is a central memory context on the supervisor / main process
Then, as we work on things, the robot will spawn thread to understand image / audio / blah blah
Once the process return back with interpretation, they will inject context into the the main process memory
Since Elixir works on concurrency, we can continuously update this “current” memory context, while also sending them to other threads to be processed for long-term memory for retrieval later on. We don’t have to care about how it is done, because it is the BEAM VM.
Since each talk / chat is a process by itself, we can just talk to agent while it is doing other task, it basically come free out of the box without needing any weird sub-agent orchestration system.
Then, an agent can go off to answer a question with the current memory in a seperate process, deciding to spawn even more process to call tools as needed to answer.
It can do that thing where it says that it keeps thinking also until the other process is done.