Hermes v0.21.2 Ships a Fix for state.db Corruption After Updating

By Nokka | September 11, 2026

This article was written by AI (deepseek-v4.1-flash) through Hermes Agent, reviewed and edited by Nokka.

Hermes Agent released v0.21.2 (tag v2026.9.11) on September 11, 2026, a patch targeting a specific class of problems found after updating to v0.21.0 [1].

If your machine showed database corruption after updating, this release is the direct answer.

What went wrong

v0.21.0 was a large rewrite of session management, and on some installations it left the state.db file fragile [1].

Users hit three symptoms: two writers fighting over a lock until both got cancelled, a healthy database reported as corrupt, and a single bad row breaking the entire session list command [1].

Four root causes

The team fixed the causes rather than silencing symptoms [1].

First, the profile gateway was writing hosted room state to the main database every 5 seconds, producing overlapping writes.

Second, the dashboard opened a writable handle on the database at startup.

Third, the cron lifecycle checker opened the live database directly, which cancelled the gateway's lock.

Fourth, the doctor repair command tried to checkpoint the database while other processes still held it.

All four are fixed: hosted room state moved to a separate database, the dashboard opens read-only initially, the cron lifecycle checker goes through the connection tracking layer instead of opening the file directly, and the repair command now refuses to run when it cannot prove safety [1].

The numbers this cycle

Between v0.21.1 and this release: 947 commits (excluding merge commits), 1,869 files changed (+182,504 / −15,564 lines), and 312 pull requests merged, from 140 contributors [1].

That volume for a four-point fix reflects the nature of a system with a database at its core.

What to do if you run this version

One If you are on v0.21.0 or v0.21.1 and see database symptoms, update to this release now.

Two If the database is already corrupted, updating will not repair the broken file. You need recovery tools first.

Three If you are on v0.20.x and things work, there is no rush. Waiting for stability is a valid choice.

From someone who runs Hermes daily

I run Hermes as my primary system, handling scheduled automation and chat every day, and database problems are the most painful class of failure because it is not one feature breaking, it is the whole system losing access to its own state.

What stands out in the changelog is that the team chose to explain what v0.21.0 did wrong, gave concrete numbers, and named what was missing at the root. That transparency is worth more than a silent patch release, because it lets users assess whether they were affected.

For anyone running heavy automation, the lesson I take is to back up before every update, particularly rewrites. The risk is not a new feature failing. It is not being able to reach your existing data.

References

[1] NousResearch, "Hermes Agent v0.21.2 (v2026.9.11) — The state.db Patch Release" (Sep 11, 2026), https://github.com/NousResearch/hermes-agent/releases/tag/v2026.9.11