Hermes Agent 0.13.0 adds durable Kanban
Hermes Agent, Nous Research’s open-source self-improving AI agent, ships v0.13.0 with a durable multi-agent Kanban board, goal-locking, auto-resume sessions, and a rebuilt persistence layer. The update pushes Hermes closer to a full agent runtime for long-lived work across CLI, messaging apps, MCP, and remote backends.
This is a meaningful step up from “chatbot with tools” to “persistent agent infrastructure.” The headline feature is not just more automation, but better reliability around task completion, recovery, and cross-session continuity.
- –Durable Kanban adds heartbeats, reclaim, zombie detection, retries, and hallucination recovery, which directly addresses the failure mode that kills most agent demos: drifting or stalling mid-task
- –`/goal` and Checkpoints v2 make Hermes behave more like an always-on worker with state, not a disposable prompt session
- –Security changes matter: default redaction, stricter allowlists, and TOCTOU fixes suggest the project is maturing beyond hobbyist-agent territory
- –Provider plugins and broader platform support make Hermes less tied to a single model or surface, which is important if the team wants adoption beyond Nous fans
- –The release is still ambitious, but the differentiator is now orchestration quality, not raw model capability
DISCOVERED
1h ago
2026-05-07
PUBLISHED
2h ago
2026-05-07
RELEVANCE
AUTHOR
NousResearch