Build a private, local-first personal AI assistant with governed autonomy.
English README | Chinese README
LiAgent OS is a local-first AI agent OS for building a private personal assistant with governed autonomy across local models and hybrid cloud services.
It combines conversation, tool use, task scheduling, heartbeat execution, human approval, and auditability into one long-lived loop.
The long-term goal is bigger than a chat assistant: LiAgent OS is being built toward an all-around personal AI agent that can understand, organize, monitor, research, and follow through across your daily digital workflows while still keeping you in control.
- Long-lived execution instead of one-shot chat.
- Human approval and auditability for risky actions.
- One runtime across CLI, Web, and Discord.
- Local-first by design, with hybrid local-cloud model services as a first-class path.
- It keeps following through on tasks instead of stopping at one reply.
- It treats budget, permissions, approvals, and audit logs as runtime concerns.
- It is built for local models first and keeps improving the hybrid path across local and cloud model services.
- It is aiming toward a trusted personal AI agent, not just a smarter chat box.
This public update focuses on making LiAgent OS more observable, more consistent, and more reliable during long-running agent workflows.
- Clearer runtime visibility, status updates, and execution feedback.
- More consistent cancellation and runtime-state handling.
- Better transparency across tool execution and fallback paths.
- Stronger recovery behavior for interrupted and preempted runs.
- Unified runtime-status behavior across Web and Discord.
- Improved stability for longer multi-step task chains.
LiAgent OS is already beyond a toy demo. It has working loops for conversation, tools, task orchestration, heartbeat execution, confirmation gates, and runtime observability. The current phase is about turning those capabilities into a more reliable governed autonomy stack for a future full-spectrum personal AI agent.
The next phase continues to push on:
- stronger local-model experience and bootstrap paths
- better hybrid routing across local and cloud model services
- lower latency and tighter cost control for real workloads
- more reliable long-running execution in practical workflows
- deeper privacy, safety, and trust guarantees as first-class product direction
- English: README(EN).md
- Localized README (Chinese): README (Chinese)
- Getting started: docs/getting-started.md
- Current boundaries: docs/current-limitations.md
- Architecture: docs/architecture.md
- Roadmap: docs/roadmap.md