Hermes Agent v0.23.3: What DeepSeek Users Should Check After Updating
By Cameron Voss and West Mao
TL;DRHermes Agent v0.23.3 reached the CLI, Microsoft Store, Docker and Hermes Cloud release channels. Standalone Windows and macOS installers are on a separate schedule. After updating, check the running version, then verify your DeepSeek connection and one real task.
What changed in Hermes Agent v0.23.3?
Nous Research dates the v0.23.3 patch to November 19, 2026. It brings changes merged after v0.23.2 into the stable release. The notes mention changes to prompt caching and conversation compaction, among other areas, and say the curated changelog will come with v0.24.0. For a DeepSeek setup, the release notes alone give you no basis to expect a particular cache saving or a change to your provider settings. If you want to see the patch itself, it is tagged in the Hermes Agent repository.
The browser workspace arrived earlier, with v0.23.0 on November 5. If you are upgrading from before that release, the workspace is part of what you are catching up to. For v0.23.3 specifically, start by identifying how the copy of Hermes you actually use was installed.
How do you update your copy of Hermes?
Update through the channel that supplied the running copy. This matters if you installed Hermes on more than one machine, use WSL, or connect from a desktop app to a separate gateway.
| Your installation | What to do | What to check |
|---|---|---|
| CLI install | Run hermes update in its environment. | Run hermes --version there after the update. |
| Microsoft Store app | Check for the app update in the Store. | Reopen the Store-installed copy. |
Desktop installed from .exe or .dmg | Wait for the next bundled release. The standalone installers were unchanged in this patch. | Identify the gateway separately if your app is connected to another host. |
| Docker / Hermes Cloud | The release lists nousresearch/hermes-agent:stable and :latest. Follow your deployment’s update process. | Inspect the running deployment. Pulling a new image alone leaves an existing container running its old image. |
What should you verify after updating?
Start with a short, tool-free prompt in a fresh session. If it works, move on to a small task that exercises the feature you rely on. The two checks help you tell whether a problem starts with the model connection or a particular tool.
- Running versionCheck the environment hosting your agent.
- Plain DeepSeek replyConfirm provider and model in a fresh chat.
- One real workflowCheck an expected output or recent run.
- Record the version and profile. On a CLI install,
hermes --versionidentifies the executable you are calling. If the UI talks to a gateway on another host, check the backend you are testing as well. - Verify a plain reply. In a fresh chat, confirm the provider and model shown for that session, then ask for a short response. For the direct DeepSeek API, the provider’s setup example uses
deepseek-v4-pro. For a Coding Plan connection, check the values delivered with your plan. - Try one familiar workflow. Ask the agent to read a known test file, or inspect the next run of a scheduled job you already rely on. For a file test, confirm that the result cites the expected filename and contents. For a scheduled job, check its run record and output.
What if the updated install still shows the old version or stops replying?
If the version is still old, check the install channel, shell and gateway before rerunning the updater. On Windows, the native shell and a WSL distribution can point to different Hermes installations. If the version is right but the chat fails, start with the selected provider, model, key and endpoint. If plain chat works and only one tool fails, troubleshoot that tool separately. The Hermes and DeepSeek troubleshooting guide has checks by symptom.
Common questions
Did v0.23.3 update the standalone Hermes Desktop installer?
No. The release notes say the standalone Windows and macOS installers will update in the next bundled release. The Microsoft Store app has its own update path.
Should I change my DeepSeek model after updating?
Keep the model ID that already works with your provider and verify it in a fresh chat. The patch notes give no DeepSeek-specific model migration to follow.