Architecting Resilient AI Agent Systems with a Central Heartbeat and Brokerless Communication

The Challenge with Dispersed AI Agent Scheduling

Many AI agent systems initially adopt a simple architecture: numerous scheduled tasks, each running in its own process, often triggered by cron. While straightforward for a small number of agents, this approach quickly leads to operational opaqueness as the system scales. Identifying failures becomes difficult, as a lack of output is indistinguishable from "nothing to report," making it hard to ascertain system health and task execution status.

Consolidating Scheduling with a Single Subsystem Heartbeat

The core architectural improvement described is the consolidation of per-task scheduling into a single subsystem heartbeat. Instead of individual cron jobs, a central loop dispatches a fixed set of subsystems, each with its own cadence and back-off logic. This design provides a unified point of visibility, allowing operators to easily monitor what tasks are firing, what's due, and what's failing, replacing a "scavenger hunt across a scheduler" with a clear operational view. This approach enables isolation and graceful degradation for failing tasks, preventing them from quietly starving neighbors.

[ON] OK a2a every 240m Agent-to-Agent — brain export & peer comms
[ON] OK a2a_server every 15m A2A protocol server — realtime peer comms
[ON] OK cognition every 30m world model + prediction + self model, one pass
[ON] OK darwin every 15m evolutionary skill evolution
[ON] OK desktop every 1440m desktop evidence ledger — integrity check
[ON] OK infra every 5m browser, ssh, plugins, sync
[ON] OK learning every 7m internet + active + self-model learning
[ON] OK meta every 30m reflection, rewriting, research
[ON] OK outreach every 180m social & github outreach
[ON] OK security every 240m cve scan, bugbot, pen-test
[ON] OK senses every 30m senses & biological rhythm
[ON] OK swarm every 30m multi-agent coordination
[ON] OK system every 5m self-healing, resources, cache

In-Process Tools and State Management

To avoid the overhead and complexities of inter-process communication, the agent's capabilities are exposed as native, in-process tools. This design allows subsystems to directly import and execute functions, sharing the same world model, memory, and locks. This eliminates the cost of process spawning on every tick and facilitates efficient state sharing, crucial for maintaining a consistent understanding across different agent functionalities. Durable state is managed in files rather than in-memory context, ensuring resilience against crashes.

Brokerless, Verified Agent-to-Agent Communication Mesh

• Identity as a Key: Messages are cryptographically signed by the sender and verified by the receiver *before any handler runs*. This decentralizes trust, removing the need for a central router and ensuring secure communication even if peers are compromised.
• Dumb Relay: Peers drop signed envelopes into a shared, passive relay. Listeners pick up and validate messages, preventing the relay from becoming a single point of failure or coordination.
• Capabilities, Not Credentials: Agents hold scoped tokens (e.g., `write:/tmp/x`, `call:send_email`) rather than long-lived secrets. Authorization checks occur at the tool boundary, meaning an injected agent cannot mint new tokens, bolstering security.