Hanji · An experimental collaboration lab

A thin surface
between people and agents.

Hanji is a research lab exploring how people work with agents. We build interfaces that make these interactions clearer, so people can understand what is happening, exercise judgment and guide what comes next.

Bring existing content or create a fresh workspace. The Mac desktop preview is currently shared by invitation.

WHY HANJI EXISTS

Clarity for a new way of working.

Agents are becoming our virtual companions and colleagues. Their capabilities are evolving faster than our ways of working with them. Hanji exists as a thin layer between people and agents: making context, contributions and choices easier to understand, so people can make sound judgments.

Our current workspace is one expression of this research. Think, Plan and Document explore different ways to interact with agents while keeping people able to understand, question and direct the work.

Our mission and principles

The research, in practice.

One workspace, three surfaces. Think, Plan and Document are our current experiments in making work with agents clearer and more useful.

The same surfaces, for people and agents.

The research, in practice.
Think — A canvas for notes, diagrams and conversations. Shape a thought before you decide what it needs to become.
Think · Give an idea some room.An example workspace in the desktop preview.
Plan — Issues, projects, dependencies and objectives. Keep the decision and its conversation close to the work.
Plan · Make the next step clear.An example workspace in the desktop preview.
Document — Read and edit real Markdown, connect your repositories, and review changes with the people and agents you trust.
Document · Keep what matters.An example workspace in the desktop preview.

01 · Think

Give an idea some room.

A canvas for notes, diagrams and conversations. Shape a thought before you decide what it needs to become.

Explore Think

02 · Plan

Make the next step clear.

Issues, projects, dependencies and objectives. Keep the decision and its conversation close to the work.

Explore Plan

03 · Document

Keep what matters.

Read and edit real Markdown, connect your repositories, and review changes with the people and agents you trust.

Explore Document
Follow an idea through the workspace

One workspace. Two ways to work with your agent.

People and connected agents can read, create and edit across Think, Plan and Document, within the access you give them. Work directly, ask for help across the workspace, or hand off a focused task.

Bring your preferred agent. Connect Claude or Codex; other compatible agents can use Hanji’s scoped MCP tools.

Across the workspace · Conductor
What changed in this project, and what should we do next?
On a specific task · Plan
@claude Clarify the acceptance criteria and suggest the next step.

01

A conductor with the wider view.

One standing agent conversation follows you across all three surfaces. Ask questions, create or edit content, and organize the workspace with the relevant context close at hand.

Meet the conductor

02

A mention starts a focused task.

In Plan, mention @claude or @gpt in an issue comment or workspace thread to ask your configured agent for help. The task gets its own session and relevant context, helping limit unnecessary context.

Explore working with agents

03

Agents learn how Hanji works.

Hanji gives agents instructions about its tools and workflows: managing tasks, structuring a brainstorm, and creating or improving documents. Their brief connects those tools to the work you are asking for.

How agents connect

04

Changes you can trace and revisit.

File-backed work has Git history, so you can review changes and return to earlier versions. Live canvases also save through a journal and Git checkpoints. The guides explain recovery and what to back up.

History and recovery

A personal need. An ongoing practice.

Hanji began with an individual need to work better with agents. It grew through countless iterations in everyday use: trying an interaction, noticing what remained unclear, and refining the interface.

We now share the results publicly and as open source, while continuing to experiment. As AI tools and methods evolve, so will Hanji. The products are working explorations of that question, with room to change as we learn.

Hanji

The question stays open.

How can we work better with our new virtual companions and colleagues? Try the current experiments, and help shape what we explore next.