kage-ai 0.1.0

AI-native cron task runner for per-project scheduled prompts and commands.

kage 影 - AI Native Cron Task Runner

English | 日本語

kage is a tool for running scheduled tasks using AI CLIs (codex, claude, gemini, etc.) or standard shell commands, managed on a per-project basis.

Features

• AI Native: Run AI prompts directly from a cron schedule.
• Flexible AI Providers: Built-in support for codex, claude, gemini, and copilot with easy customization.
• Inline Overrides: Customize commands, AI models, or parsers (like jq) for each specific task.
• 3-Layer Configuration: Configuration is merged from library defaults, user overrides (~/.kage), and workspace-specific settings (.kage).
• Web UI: Monitor task execution and logs through a sleek browser dashboard.

Installation

The easiest way to install kage is via the interactive installer:

curl -sSL https://raw.githubusercontent.com/igtm/kage/main/install.sh | bash

Or install from PyPI:

pip install kage-ai

Alternatively, install with uv:

uv tool install git+https://github.com/igtm/kage.git
kage onboard

Getting Started

1. Global Setup (First time only):

   kage onboard
   

   This initializes ~/.kage/, the database, and the crontab entries.

2. Configure AI Engine: Create ~/.kage/config.toml and specify your default engine.

   default_ai_engine = "codex"
   

3. Initialize Project: Run this in your project directory.

   kage init
   

   This creates .kage/tasks/sample.toml.

Task Definition Samples

Define tasks in .toml or .md files under .kage/tasks/.

• *.toml: existing format (single or multiple tasks per file)
• *.md: front matter + markdown body, one file = one prompt task only

# Auto-refactor using AI
[task_refactor]
name = "Daily Refactor"
cron = "0 3 * * *"
prompt = "Please clean up the code in src/"
provider = "claude"

# Classification with JSON/JQ parsing
[task_labels]
name = "Ticket Labeling"
cron = "*/30 * * * *"
prompt = "Classify this issue as JSON '{\"label\":\"...\"}': 'Cannot login'"
provider = "codex_json"
parser_args = ".label"

# Standard Shell Command
[task_cleanup]
name = "Log Cleanup"
cron = "0 0 * * 0"
command = "rm -rf ./logs/*.log"
shell = "bash"

---
name: Nightly Research
cron: "0 2 * * *"
provider: codex
---

Collect benchmark updates and summarize differences.
Add comparison points for quality, speed, and cost.

In markdown tasks, the entire body after front matter is treated as the prompt.

Commands

• kage onboard: Initialize global settings and OS-level daemon.
• kage init: Initialize current directory as a kage project.
• kage daemon install: Register kage to system scheduler (cron/launchd).
• kage daemon remove: Unregister kage from system scheduler.
• kage daemon status: Check daemon registration status.
• kage config <key> <value> [--global]: Update configuration via CLI.
• kage config-show [--workspace <path>]: Show resolved config (merged defaults/user/workspace), including loaded providers and commands.
• kage doctor: Check setup health and validate config/task files (unknown keys, type errors, invalid cron, missing front matter, etc).
• kage ui: Launch web dashboard (default: http://localhost:8484).
• kage logs: View execution history.
• kage run: Force run all scheduled tasks (normally executed by cron/launchd).
• kage task list: List all tasks across all registered projects.
• kage task show <name>: Show details for one task.
• kage task run <name>: Run one task immediately.
• kage project list: List registered projects.
• kage project remove [path]: Unregister a project.

Release / Publish

# 1) Build package
uv build

# 2) Create release (example: v0.0.1)
gh release create v0.0.1 --title "kage v0.0.1" --generate-notes

# 3) Publish to PyPI (token auth)
TWINE_USERNAME=__token__ \
TWINE_PASSWORD='<pypi-token>' \
uvx twine upload dist/*

Codex Provider Note (Headless / launchd)

When defining a custom codex command template, place global flags before exec:

[commands.codex]
template = ["codex", "--ask-for-approval", "never", "--sandbox", "workspace-write", "exec", "{prompt}"]

codex exec --ask-for-approval ... may fail depending on CLI version.

License

MIT