Hermes Kanban keeps agent work durable
Hermes Kanban is a local SQLite-backed board inside Hermes Agent that keeps task state, dependencies, run history, and handoff metadata across profiles. The tutorial shows four practical patterns: solo feature shipping with parent-child task promotion, parallel specialist queues, PM-to-engineer-to-reviewer pipelines with retries and blockers, and crash recovery when workers fail or trip the circuit breaker. The core value is persistence: every attempt is stored as a run, downstream workers inherit structured context, and the board stays consistent across the dashboard, CLI, and worker tool calls.
The real product here is not a kanban UI, it is a persistence and coordination layer for agents that makes failure and retry first-class.
- –Durable state beats ephemeral chat for anything with handoffs, dependencies, or review cycles.
- –Structured metadata on each run is the important part: it gives downstream workers context without forcing humans to re-explain decisions.
- –The retry/block/recover flow is well suited to agentic work where partial progress, reviewer feedback, and crash recovery are normal.
- –Using one shared SQLite database across dashboard, CLI, and worker tools keeps the workflow simple and local instead of turning it into infrastructure.
- –Best fit: solo builders, specialist queues, and small multi-role pipelines that need traceability more than heavyweight project management.
DISCOVERED
45d ago
2026-05-04
PUBLISHED
45d ago
2026-05-04
RELEVANCE
AUTHOR
AICodeKing
