v0.5.0 — first public release

First public release of the KageOps engine — the open-source core behind KageOps.

What's in the box

• Sensei orchestrator + 8 specialist Autonauts — decomposes a one-line brief into routed tasks across strategy, architecture, engineering, QA, design, data, and marketing agents, from idea to a verified build.
• Local-first by default — embedded Postgres (PGlite) runs in-process; no Docker, no external services. Your code never leaves your machine.
• Hard cost guardrails — per-run USD budget kill switch, per-task AI-call caps, per-agent token ceilings, dry-run-first workflow.
• Provider-agnostic — Claude (API or CLI), OpenAI, Gemini, OpenRouter, or fully local via Ollama.
• Phase gates — build verification + acceptance gates check the output before the pipeline calls it done; failed acceptance loops back to the engineer agent with a bounded retry budget.
• Headless CLI runner — run the full pipeline without the GUI, including --dry-run (no AI spend) and --resume.

Getting started

See the README quick start · Docs: https://kageops.dev

License

AGPL-3.0 — free forever: unlimited projects, all agents, self-host.