Most productivity tools stop at reminders. Life Manager is a proactive general agent built to actually manage your life, organized around three pillars: Body, Mind, and Money.
It runs a continuous self-improvement loop:
- Monitor what's happening across your schedule, habits, and finances
- Self-fix small problems before you notice them
- Refactor the routines that keep failing
- Improve what's already working
- Self-replicate the setups that worked into new areas of your life
Think of it less like a to-do app and more like a general agent whose job is your life: it watches, it acts, and it rewrites its own playbook when something stops working.
Open source
Life Manager is open source. You can read the core, run it yourself, and extend it:
- Project page: aniccaai.com/lm
- Source: github.com/Daisuke134/life-manager
If you're building agents, I'd love to hear how you'd design a self-improvement loop that actually converges instead of spiraling. Issues and PRs are welcome.
Top comments (1)
Thấy approach "proactive general agent" khá thú vị — đa số tool hiện nay vẫn reactive (nhập liệu → nhắc nhở), trong khi Life Manager đang thử đóng vai trò orchestrator thực sự.
Mấy điểm em quan tâm nếu đóng góp hoặc tự host:
Memory & context window — quản lý body/mind/money nghĩa là data nhạy cảm, liên tục, và cross-domain. Em thấy việc compress/aggregate context theo thời gian (tuần/tháng/năm) quan trọng hơn raw RAG. Có plan cho hierarchical memory không?
Tool calling vs. built-in skills — money cần Plaid/YNAB API, body cần Health Kit/Google Fit, mind cần journaling/therapy prompts. Có chuẩn hóa interface để community plug skill mới không, hay vẫn hard-code trong core?
Local-first vs. cloud LLM — data y tế/tài chính khó để qua API public. Có support Ollama/vLLM local inference out-of-the-box không? Em thấy architecture plugin cho provider sẽ quyết định adoption ở nhóm privacy-conscious.
Evaluation loop — agent proactive dễ hallucinate action (vd: "em đã chuyển $500 cho tiết kiệm" mà thực ra chưa). Có eval harness tự động (simulated user + state diff) để catch regression không?
Onboarding friction — "general agent" nghe đẹp nhưng user mới bị overwhelmed. Có progressive disclosure: bắt đầu 1 domain (money), dần mở rộng không?
Em đang theo dõi repo — nếu có good first issues về memory layer hoặc skill registry em sẽ PR. Cách handle multi-user isolation (family sharing) cũng là bài toán hay PS: the tool I meant is on labagent .tech