An agent runtime that treats your job as a state machine, not a chat

What it does

LazyMind is a Go-based “AI Skill Runtime”: a workspace where knowledge bases, reusable Skills, and executable Workflows come together to run knowledge-intensive tasks end to end. Instead of re-uploading context and babysitting every agent step, you pick the knowledge and workflow once, and it plans, executes, and produces editable artifacts — reports, documents, slide decks — with citations and version history. It runs as a local-first desktop app (macOS/Windows) or as a shared team/enterprise service.

The interesting bit

Workflows are defined as state machines with visible steps, approval checkpoints, retries, and rewinds — so a failed step means rerunning that step, not restarting the task. It also works as a plugin for external agents: Codex and DeepSeek Harness sessions can call LazyMind’s knowledge, Skills, and Workflows natively, including progress tracking and artifacts.

Key highlights

• Local-first desktop mode aimed at non-programmers: architects, PMs, testers, content creators
• Knowledge foundation with multi-source ingestion (local files, Feishu, Notion, object storage), OCR via PDFReader/MinerU/PaddleOCR-VL, hybrid retrieval, reranking, and inline citations
• Skills convert into versioned, publishable, rollback-able Workflows — expert practice becomes a managed asset
• PPT workflow goes from research through outline approval to editable PPTX export with speaker notes
• Feedback loop: bad cases get captured, evaluated, A/B tested, and deployed as verified improvements

Caveats

• All demos, videos, and articles are in Chinese; English documentation exists but the community material skews one language
• The README is long on product framing and short on technical internals — how the state machine engine actually executes isn’t detailed here

Verdict

Worth a look if you do long, multi-step knowledge work and are tired of chat-window archaeology. If you just want a quick RAG chatbot, this is considerably more machinery than you need.