Agent Mesh

The background layer of scheduled agents. EA, family, content, SRE, research, finance, AI ops, creative.

The agent mesh is the part of Arkeus that runs without Ryan in the room. It is a set of scheduled agents, each with a scoped responsibility, running on launchd cycles in the background. EA handles morning scans and logistics. Family watches Janelle's calendar and Greyson's schedule. Content runs the writing pipeline. SRE monitors system health. Research does deep dives. Finance tracks expenses and subscriptions. AI ops maintains the tooling layer.

Each agent has its own specification file, its own decision log, its own corrections file, and its own state. They do not share memory directly. When one agent needs to route information to another, it writes to a well-defined section of the target agent's state file with full provenance. The runner scans that section on the next cycle and processes the routed items with the same write policy discipline as any other input.

The mesh runs on schedules, not on event loops. A typical agent cycle is 20 to 60 minutes long, and the schedule is managed by launchd on macOS. This is deliberate: event-driven agents are harder to reason about, harder to recover from crashes, and easier to accidentally turn into infinite loops. Schedule-driven agents are boring, which is what you want from background infrastructure.

The mesh is designed to surface only what needs human attention. If an agent cycle completes cleanly, it writes a single line to its log and ends. If it surfaces something that matters — a pending recruiter message, a calendar conflict, a failing health check — it routes the surface to Telegram or the morning brief.

The mesh is the body of Arkeus. Interactive sessions are the head. The head is where diagnosis and prescription happen; the body is where the work gets done quietly, on schedule, under governance, with a paper trail.

Related
← back to glossary