Hermes Agent Dashboard Update: Multiple Agents Online - Grab Tasks and Start Working Without Opening Multiple Terminal Windows

According to Dynamic Beating monitoring, Nous Research's open-source agent framework Hermes Agent introduced the Kanban multi-agent collaboration system in v0.12.0. The core idea is to place tasks on a shared board, with each agent acting as an independent operating system process to autonomously claim and execute tasks in parallel, replacing the traditional scheduler for in-process sub-agent clusters.

The entire system operates around a local SQLite database. Each task claim is an atomic transaction, where only one agent can win when multiple agents compete for the same task. In case of agent crashes or timeouts, the scheduler automatically recovers tasks and reassigns them through process liveness detection; three consecutive failures trigger a circuit breaker, locking the task automatically for manual intervention to prevent the board from idling.

The framework includes two built-in skills: an orchestrator and a worker. The orchestrator is responsible for breaking down the objectives and assigning them to expert roles (researchers, engineers, auditors, etc.), while the worker is responsible for execution. Tasks can establish parent-child dependencies, with upstream summaries and metadata automatically passed down to downstream agents upon completion. It supports nine collaboration modes, including fan-out parallelism, pipeline, voting arbitration, and human-in-the-loop review.

The accompanying Web dashboard provides real-time event updates via WebSocket, and all operations are accessible through CLI and slash commands on gateways such as Telegram and Discord.