AI Workroot – a local-first workspace protocol for durable AI work

[aiworkroot-admin]

Hi HN,

I built AI Workroot, an open-source local-first workspace protocol for durable AI work:

https://github.com/ai-workroot/ai-workroot

The starting point is simple: most AI work disappears into chat history.

People talk with AI, solve problems, make decisions, learn lessons, and create useful knowledge, but much of that
context is scattered, forgotten, or locked inside a specific tool. I wanted to explore a different direction: an AI
workspace that can grow with a person, a team, or a role over time.

AI Workroot is not another agent runtime. It is closer to an AI workspace operating protocol.

Its goal is to give AI work a durable home:

• identity: who this workspace serves
• tasks: what is being worked on
• memory: what happened
• knowledge: what can be reused
• decisions: what was chosen and why
• handoff: where to continue next time
• forgetting: how painful or outdated context can become quiet, archived, tombstoned, or deleted by user choice

The philosophy behind the project is that AI should not only help people finish more tasks. It should also help
people remember, understand, grow, and continue their work with more clarity.

The project is designed to be:

• local-first
• agent-neutral
• simple for ordinary users
• rigorous for AI agents
• portable across models, tools, and operating systems
• useful for both personal and team workspaces

The current release is v0.9.527.

It already includes a user-space / kernel-space architecture, a first-use guide, an operating protocol for AI
agents, local runtime/index support, validation scripts, and documentation for product experience, architecture,
scaling, extensions, and long-term knowledge preservation.

I would really appreciate feedback from this community, especially on:

• whether the first-use experience is simple enough for non-engineers
• whether the workspace/kernel abstraction makes sense
• how local AI workspaces should handle memory, knowledge, forgetting, and handoff
• how this could interoperate with existing agent runtimes, MCP/ACP, coding agents, and personal AI assistants
• what would make this genuinely useful in daily work

This is an early public release, but the intention is serious: to explore a durable, human-centered foundation for
working with AI over the long term.

Repo:

https://github.com/ai-workroot/ai-workroot

Feedback, criticism, and suggestions are very welcome.