On May 7, 2026, NousResearch shipped Hermes Agent v0.13.0 — “The Tenacity Release”. It’s not a patch. Since v0.12.0, the project absorbed 864 commits, 588 merged PRs, and 295 community contributors who closed 282 issues. If you haven’t looked at Hermes since yoDEV’s previous coverage, this release is another beast.
What is Hermes (short version)
Hermes runs persistently on your own server. Each time it completes a task, it pauses, reflects, and auto-generates a skill file — a captured procedure it can reuse the next time that same problem appears. Memory is divided into three layers:
- A persistent snapshot of your identity and agent state
- A full-text search index with SQLite FTS5 from each past session
- Procedural skill files that encode logic for repeatable tasks
The longer you run it, the more it knows about your specific projects. It doesn’t solve the same problem twice.
Supported gateways: Telegram, Discord, Slack, WhatsApp, Signal, CLI — and with v0.13.0, Google Chat becomes the 20th platform.
What’s new in v0.13.0
Multi-Agent Orchestration with Kanban
The headline feature. Hermes now includes a Kanban board as a durable coordination layer for fleets of agents. Each agent reports a heartbeat at regular intervals. If an agent misses the heartbeat — it’s marked as zombie and the system initiates recovery. Recovery from hallucinations is also built in: when an agent produces output that doesn’t match expected constraints, the orchestrator flags it and reassigns the task.
This means you can run multiple Hermes instances working toward a shared goal, coordinating autonomously for hours without manual oversight.
/goal Command
Lock the agent to a specific objective across turns. Once defined, Hermes won’t deviate from the goal — not even with context resets or gateway interruptions.
Checkpoints v2
Real state pruning, not just snapshots. The agent can now surgically trim its working memory to what’s relevant, reducing bloat in long-running sessions.
Auto-resume for Gateways
After a restart — whether intentional or not — gateways reconnect and resume without losing session context.
The numbers behind it
The project sits at 135K stars on GitHub. On OpenRouter, Hermes is currently the #1 app/agent by daily inference volume: 224 billion tokens/day (vs. 186 billion for OpenClaw). It’s not a benchmark — it’s production use across real developer workflows. MIT license across the entire project.
Who this is for
If you run workflows that span more than a single context window — database migrations, test suite expansion, iterative refactors, documentation generation at scale — Hermes v0.13.0 gives you the infrastructure to coordinate multiple agents without glue code. The skill file system ensures that the second run of any repeatable task is always faster than the first.
Connection speed matters here. Hermes is self-hosted, so your server’s bandwidth and latency affect gateway reliability. The new auto-resume for gateways is a direct acknowledgment of that reality.
GitHub: NousResearch/hermes-agent · Release: v0.13.0