Repository · workstreams
Workstream: Agent Memory
Status: Planned · Component: Enable agentic swarms
Goal
A persistent, queryable knowledge store that agents read from and write to over url://, backed by an approved storage mechanism, so a lesson learned by one agent is immediately available to the whole fleet. It is the concrete, buildable form of knowledge compounding: turn a fleet's collective experience into a compounding asset instead of a per-agent, per-session memory that evaporates.
Why it matters for swarms
A swarm that cannot share what it learns re-solves the same problems over and over. Agent Memory is what makes the fleet's experience accumulate — the difference between N independent agents and one organism that gets smarter with every task. It builds directly on SearchableBucket, which already provides query-by-content document storage over url://searchable-bucket/; SearchableBucket is a concrete step toward this larger, fleet-wide memory. The External Brain is the human-facing counterpart — a personal knowledge graph for a person rather than a shared memory for a fleet — and the two share substrates without sharing a store.
Graduation
This is early-stage: the fleet-wide store itself is not built (SearchableBucket exists as a substrate but not the shared-memory product on top of it), and it has no Documentation Repository project page. It remains a workstream until agents in a real swarm durably read and write shared knowledge across sessions over url://, at which point it becomes a first-class project with a Documentation Repository project doc and an ALL_PROJECTS.md entry.