| name | jobs |
| description | Work with scheduled jobs and container services. Use when creating, updating, pausing, debugging, or inspecting jobs; checking runs/events; or changing runit services like webui and jobs. |
Jobs and Services
The container uses two related systems:
- Services are long-running runit processes. Definitions live in
/home/agent/.config/term-llm/services/<name>/run and are reinstalled into /etc/sv on container start by /home/agent/.config/term-llm/init.sh.
- Jobs are scheduled or manual units managed by
term-llm serve jobs, exposed on http://127.0.0.1:8080, and controlled through term-llm jobs.
Default services:
| Service | Purpose |
|---|
webui | Runs term-llm serve web on port 8081. |
jobs | Runs term-llm serve jobs on port 8080. |
Default jobs:
| Job | Schedule | Purpose |
|---|
mine-sessions | every 30 min | Mine session transcripts into memory fragments. |
update-recent | every 10 min | Promote fragments into recent.md. |
system-upgrade | Daily 05:00 UTC | Run the distro package upgrade (pacman -Syu on Arch, dnf upgrade on Fedora). |
Inspecting jobs
term-llm jobs list
term-llm jobs get <job-id-or-name>
term-llm jobs runs <job-id-or-name> --limit 20
term-llm jobs active
term-llm jobs run get <run-id>
term-llm jobs run events <run-id> --limit 200
Use --json when scripting or when you need exact IDs and payloads:
term-llm jobs --json list
term-llm jobs --json runs mine-sessions --limit 5
Creating a program job
Program jobs execute a command directly from the jobs worker:
term-llm jobs create --data '{
"name": "example-program",
"enabled": true,
"runner_type": "program",
"runner_config": {
"command": "/usr/local/bin/term-llm",
"args": ["memory", "update-recent", "--agent", "'$AGENT_NAME'"]
},
"trigger_type": "cron",
"trigger_config": {"expression": "*/30 * * * *", "timezone": "UTC"},
"concurrency_policy": "forbid",
"timeout_seconds": 300,
"misfire_policy": "skip"
}'
Creating an LLM job
LLM jobs run agent instructions in the background. Always set runner_config.agent_name explicitly. LLM jobs also require runner_config.cwd; it roots the run's file/shell tools without changing the jobs server process directory.
term-llm jobs create --data '{
"name": "daily-review",
"enabled": true,
"runner_type": "llm",
"runner_config": {
"agent_name": "'$AGENT_NAME'",
"instructions": "Review recent memory and produce a short status note.",
"cwd": "'$PWD'"
},
"trigger_type": "cron",
"trigger_config": {"expression": "0 9 * * *", "timezone": "UTC"},
"concurrency_policy": "forbid",
"timeout_seconds": 1800,
"misfire_policy": "skip"
}'
Updating and operating jobs
term-llm jobs update <job-id-or-name> --data '{"enabled": false}'
term-llm jobs pause <job-id-or-name>
term-llm jobs resume <job-id-or-name>
term-llm jobs trigger <job-id-or-name>
term-llm jobs delete <job-id-or-name> --cancel-active
term-llm jobs run cancel <run-id>
Prefer update over delete/recreate when preserving run history matters.
Service changes
To add or change a long-running service, edit files under /home/agent/.config/term-llm/services. Each service directory needs an executable run script. Then restart the container or run the init hook:
bash /home/agent/.config/term-llm/init.sh
Debug service logs through Docker output for the workspace, or enter the container and inspect /etc/sv, /etc/runit/runsvdir, and the service run script.
