Always-on Services

Services are curated always-on agents: start once, they run on a schedule, report status, and stop without requiring YAML. Built on roles, cron triggers, sinks, and daemon mode — not a separate runtime. Available since v2026.7.6.

Prerequisites

Linux (process supervision is Linux-only in v1)
A configured model provider (initrunner setup or env keys)
For collector: pip install "initrunner[search]"

Quick start

# See what's available
initrunner service list

# Start collector (needs initrunner[search] + a provider)
initrunner service start collector acme.com

# Or with explicit schedule / sink
initrunner service start collector acme.com \
  --every daily \
  --sink file:./collector-report.md

# Health
initrunner service status collector

# Force one cycle without waiting for cron
initrunner service run collector

# Stop (keep instance) or purge local files
initrunner service stop collector
initrunner service stop collector --purge

Commands

Command	What it does
`service list`	Catalog + local status
`service info <slug>`	Params, requires, defaults
`service start <slug> [primary]`	Materialize + start daemon
`service status <slug>`	State, process health, last run, cost
`service run <slug>`	One-shot tick (briefly stops daemon if running)
`service stop [--purge]`	Stop daemon; `--purge` deletes instance dir
`service logs <slug>`	Daemon log tail

Start options

Flag	Meaning
positional	Value for the service `primary_param` (e.g. `target`)
`--set key=value`	Any declared param (duplicate keys rejected)
`--every`	`hourly` / `daily` / `weekly` / raw 5-field cron
`--sink file:/path` or `webhook:url`	Replace entire sink set (repeatable)

Do not pass the primary both as a positional and via --set.

Shipped services

`collector`

Continuous monitoring for one target. Searches and reads public sources, diffs against memory, writes a brief when something material changes.

Param	Required	Default	Notes
`target`	yes	—	Company, person, domain, topic (primary)
`alert_severity`	no	`medium`	`low` / `medium` / `high`

Default schedule: daily (06:00 UTC). Requires: initrunner[search].

Runtime agent name is service-collector (isolates audit, memory, and budgets from any role also named collector).

How it works

Catalog — templates under initrunner/service_catalog/<slug>/ (service.yaml + role.yaml).
Start — validates params and requirements, writes a generationed instance role with {{ params.* }} filled, injects schedule/sinks, starts a supervised daemon.
Stop — signals only a verified daemon process; instance config remains unless --purge.
Run — one-shot tick; if the service was running, the daemon is stopped, the tick runs, then the daemon is restarted.

~/.initrunner/services/<slug>/
  state.json
  role.<n>.yaml
  daemon.log
  data/                 # relative file sinks

~/.initrunner/services/.locks/<slug>.lock   # never purged

Purge scope

stop --purge deletes instance configuration, logs, and file outputs under ~/.initrunner/services/<slug>/. It does not delete:

global audit history
memory store data for service-<slug>
budget counters

v1 limits

Linux only for start / stop / run supervision
One instance per service slug
Agent entry only (no Flow services)
Per-service process (not a multi-service supervisor)
Sinks: file: and webhook: only
No reboot auto-start (re-run service start after reboot)