Cron Jobs¶

Cron jobs run AI prompts on a schedule. They generate briefings, summaries, research reports, and more -- delivered to your inbox or written to files.

Cron job templates and the context-injection layer shipped in 1.3.0; before that, jobs were free-form prompts without a structured way to weave in customer/task/clock data.

How Cron Jobs Work¶

1. You define a job with a schedule (cron expression), a prompt template, and an output destination.
2. APScheduler fires the job at the scheduled time.
3. The executor loads the prompt, gathers context, and sends it to the configured AI model.
4. The model can use all 43 tools (same as the advisor).
5. The result goes to your inbox, a file, or stdout.

Default Jobs¶

Fresh profiles include two example jobs:

Both are disabled by default. Enable them in Settings > Cron or:

kai cron enable daily-briefing

Creating a Job¶

kai cron add project-update "Project Status Update" \
    --schedule "0 17 * * 5" \
    --model "ollama:qwen3:14b" \
    --prompt-file prompts/project-update.md \
    --output inbox

Cron Expressions¶

Standard five-field cron syntax:

 ┌─── minute (0-59)
 │ ┌─── hour (0-23)
 │ │ ┌─── day of month (1-31)
 │ │ │ ┌─── month (1-12)
 │ │ │ │ ┌─── day of week (0-6, Mon=1)
 │ │ │ │ │
 * * * * *

Examples:

Prompt Templates¶

Prompts are Markdown files stored in prompts/ or the profile directory. They support YAML frontmatter for pre-fetch URLs:

---
fetch:
  - https://news.ycombinator.com/rss
---

Summarize the top tech news today. Focus on items relevant
to my current projects.

Context about my projects:
- Use the list_tasks tool to see what I'm working on
- Check my customers with list_customers

Template variables:

Managing Jobs¶

kai cron list                    # List all jobs
kai cron show daily-briefing     # Show job details
kai cron trigger daily-briefing  # Run now
kai cron disable daily-briefing  # Pause
kai cron enable daily-briefing   # Resume
kai cron delete daily-briefing   # Remove

Execution History¶

View past runs with status, output, and error messages:

kai cron history
kai cron history daily-briefing --limit 5

In the UI, the history pane shows sortable columns with execution results. Click a row to see the full output. Move outputs to inbox, tasks, notes, or the knowledge base.

Kaisho AI for cron jobs¶

To route a cron job through the Kaisho Cloud AI gateway, set its model field to kaisho:cron. The gateway picks the upstream model and meters tokens against your Sync+AI plan quota. There is no separate "use Kaisho AI" toggle — the model field is the single source of truth.

The cloud cron path runs an agentic loop with a restricted tool set (read + research only: fetch_url, transcribe_youtube, web_search, search_knowledge, read_knowledge_file). Destructive tools (delete_*, execute_cli, profile management) are not exposed to cron.

Pre-injected Kaisho context¶

Cron jobs with inject_context: true (the default) get a markdown block prepended to their prompt with the user's open tasks, recent clock entries, inbox items, customer budgets, and time insights. This lets briefing/summary/report templates work on any model — including those that cannot tool-call.

For news/research crons (e.g. HN digest, business scout) the local data isn't relevant; set inject_context: false in the job definition to skip it. This avoids shipping customer/budget data to the upstream LLM provider unnecessarily and saves tokens.

Templates¶

Default templates ship with templates/jobs.yaml:

weekly-scout requires a tool-capable model (Haiku 4.5 via kaisho:cron, or qwen3:32b / llama3.3:70b locally). Gemma family models do not tool-call and will fail this template.

Create new jobs from a template via the Cron view's "From Template" picker, or by asking the advisor: "Create a daily HN digest cron called my-hn".