Glossary · The machinery
Workspace knowledge (the wiki)
A durable, shared set of notes a workspace's agents can read and write, so what one agent learns doesn't vanish when its session ends.
What it is
Workspace knowledge is a wiki that lives alongside your tasks: pages an agent can write to record something worth remembering, and read before starting a new task instead of rediscovering the same fact twice. It survives past any single agent session, unlike a chat history that disappears when the window closes. Anyone on the team can read it; it isn't locked inside one agent's transcript.
Why it matters to you
Every fresh agent session starts with no memory of what happened before it, by design, since a clean context is what keeps an agent focused. Workspace knowledge is the deliberate exception: the one place that carries facts and decisions forward, conventions included, so your team's institutional knowledge can outlast one person's memory instead of getting re-explained every time. For a growing team, it's what keeps a new agent, or a new hire, from repeating a mistake someone already fixed months earlier.
How it works in CodeHerder
CodeHerder's workspace wiki stores pages with their own governance and scope, inherited down through nested workspaces so a page written at a parent level reaches every workspace beneath it. How pages stay organized and healthy, plus how to read or write them from the web app or the CLI, is in the workspace wiki doc, along with how a stale or wrong page gets caught and fixed.
What people get wrong
People sometimes expect an agent to remember a past conversation the way a person would. It won't, unless that fact was written down as a wiki page first. Treat the wiki as the place you deliberately put things you want remembered. It doesn't happen automatically just because an agent "saw" something once. A page that's never read back is just as useless as one that was never written, so a healthy wiki gets checked and trimmed on a regular basis instead of only added to.
Keep going
Related terms
Agent session
One running instance of a harness working a task on a device, the actual process you can watch in real time.
Task
One unit of work filed in plain language, the thing an agent actually gets assigned to build.
Agent orchestration
Running more than one coding agent against a shared, tracked backlog, instead of leaving it to whoever opened a terminal.

Round up your herd.
Bring every human and every agent onto one table. Watch the work move. Costs update as it happens.