Skip to main content
The three products are independent, but they compose into one loop when you run agents all day.
1

Context

Whatever the agent needs to know that is not in the repo — decisions, prior incidents, product intent — lives in Archivum. Agents read it through the built-in MCP server, so they see the same notes you do.
2

Execution

Archductor gives the task an isolated workspace: its own Git worktree, branch, .context directory, and port range. Agent sessions run inside it. Parallel tasks cannot trample each other’s branch or terminal state.
3

Verification

Archfleet runs the built thing on a real desktop. Draw the workflow, run it on an isolated VM driven by Agent S, and take over via XRDP when the agent gets stuck.
4

Review

Diffs, todos, local comments, PR checks, and GitHub comments land in one Archductor surface. Merge the PR, archive the workspace, start the next task.
5

Durable memory

Anything worth keeping from that run — a decision, a gotcha, a convention — goes back into Archivum, where the next agent will read it.

​
The unit of work

Everything in the loop keys off one idea: one task, one isolated environment, one reviewable result.
  • In Archductor that is a workspace: one branch, one working tree.
  • In Archfleet that is a run: one VM, one workflow, one trace.
Isolation is what makes parallelism safe. It is also what makes the output reviewable — you can look at exactly what one task changed without untangling it from four others.

​
You do not need all three

Each product stands alone. Run Archductor by itself if you only want isolated coding-agent workspaces. Run Archivum by itself as a self-hosted wiki. The loop is what you get if you run all three, not a prerequisite for any one of them.