How to Schedule Tasks with DeepSeek in Hermes Agent (November 2026)

By Cameron Voss and West Mao · Updated

TL;DRUse Hermes cron to turn a folder of project notes into a Markdown brief every weekday at 9 a.m. Set the job’s time zone and model, save the result locally, and run it once by hand before relying on the schedule.

What do you need before you schedule a task?

You need a working DeepSeek connection, a running Hermes Gateway and a folder the gateway can read. Make a small test folder on the gateway host and put two Markdown files in it. If Hermes runs in WSL, Docker or on a remote host, use a path visible in that environment.

Sample input for a first run
FileExample contentsExpected result
handoff.mdThe onboarding checklist is ready for review.Shows up under Progress with the filename.
release.mdQA sign-off is still pending. No owner is recorded.Shows up under Blockers. The brief leaves the owner unspecified.

Test the connection with a short chat first. The direct DeepSeek example below uses deepseek-v4-pro through the built-in deepseek provider. Start with the Hermes Agent setup guide if you still need to connect the model.

How do you create a weekday brief at 9 a.m.?

Use 0 9 * * 1-5 for 09:00 on Monday through Friday and set an explicit time zone. In this Linux, macOS or WSL shell example, replace /absolute/path/to/brief-test with the real folder path on the gateway host. Check hermes cron create --help if your installed version differs from the options shown here.

hermes cron create '0 9 * * 1-5' \
  'Read handoff.md and release.md from the job folder. Return a Markdown brief
with headings Progress, Blockers, and Next Steps. Cite the filename for each
item. If an owner or date is missing, say so. Leave the input files untouched.' \
  --name 'Weekday project brief' \
  --timezone Asia/Shanghai \
  --workdir /absolute/path/to/brief-test \
  --provider deepseek \
  --model deepseek-v4-pro \
  --deliver local

The prompt names the files and output format because a scheduled run starts in a fresh session. With local delivery, Hermes stores the final response under the active profile’s cron/output/ directory. You can inspect the Markdown there without setting up email or a messaging channel.

Which model will the job actually use?

Hermes checks the job’s own model first, then the cron default, then the main agent model. The example pins DeepSeek for this job so a change to your everyday chat model leaves the brief on the intended model.

  1. Job pinUse the model and provider chosen for this job, when set.
  2. Cron defaultOtherwise, check cron.model and cron.model_provider.
  3. Main agentIf both are unset, use the current main agent model.
Figure 1. Model selection for an agent-backed scheduled run.

If you use a different provider or a Coding Plan, use its configured provider and model ID. If you already verified that connection as the main model in your current profile, check the job editor’s model choices or use the documented --pin option to pin the current model when creating the job.

How do you test the brief before tomorrow morning?

Run the job manually while the gateway is active, then inspect the newest run and its output. If the gateway is down on a local setup, start it with your normal service command or run hermes gateway run in a separate terminal for the test.

hermes cron status
hermes cron list
hermes cron run JOB_ID
hermes cron runs JOB_ID

Replace JOB_ID with the ID from the list and allow time for the run to finish. Check that the brief mentions the checklist and QA blocker, cites both files and leaves the missing owner unspecified. Look at the next run time too. A manual run before the next scheduled slot tests the workflow and leaves that scheduled slot in place.

Why did the task miss its slot or produce the wrong brief?

Diagnose the first failing step
SymptomFirst checkWhat to do next
Nothing ranGateway status, whether the job is paused, and the next scheduled time.Run hermes cron status. Confirm the gateway remains active at the scheduled hour and check the job’s time zone.
The job used another modelJob-level model and provider.Check the pin first. For an unpinned job, check the cron default before changing the chat model.
The brief lacks details from a fileThe gateway’s view of the job folder and access to the file tools.Check the actual path and filenames on the gateway host, then repeat the two-file test.
The run finished, but there is no messageDelivery target.With --deliver local, look for the result in the profile’s cron output folder.

Common questions

Does the scheduled job need a chat window open?

No. The gateway handles scheduling and execution. Keep that process running on the host that can reach the input files.

Does this workflow need a search subscription?

No. It reads local files and uses model inference. If you add web search later, check that tool’s setup and charges separately.

Back to the DeepSeek + Hermes Agent hub