The Claude Code desktop workspace

From ‘chat’ that scrolls by,
to a ‘workspace’ that remembers.

Rondo is a macOS workspace built around Claude Code. Open a folder and its structure, its instructions, and everything you’ve made in it become the agent’s context. Research, planning, writing, organizing — work that isn’t code, done with the most powerful agent there is. No terminal required.

Early beta Free · No uploads · Your real Claude Code

macOS blocks the first launch — here’s the 15-second fix →

Rondo running on a real vault: a task rail with status dots, working folders with the harness dock, a live markdown document, and a running Claude Code agent
  1. 01 tasks
  2. 02 folders + harness
  3. 03 documents
  4. 04 agent

The real app on a real vault. One agent playing, one waiting for you — and the document it just wrote, already open.

Real Claude Code underneath — your ~/.claude, your skills, your MCP servers, your sessions. Resumable with --resume. One click to the raw terminal.

Why a workspace

Three reasons a folder beats a chat window.

No terminal, nothing lost

Ask in chat, watch every tool call as a card, read the result as a document. The real CLI runs underneath the whole time.

Context accumulates

What you make today is background knowledge tomorrow. The folder is the memory — you stop re-explaining every session.

Files stay yours

Everything is a plain local file on your disk. Nothing is uploaded. Delete the app and you lose an app, not your work.

The difference

This is how it changes

Pasting the same context into every new session

Open the folder — it already knows

A black screen and a wall of scrolling text

Chat cards, with the document opening beside them

A developer IDE where you use a tenth of the buttons

Four columns made for work: tasks, folders, documents, agent

Folder organization stays a New Year’s resolution

PARA scaffolded on day one — and the agent knows what it means

Switching windows to check if it’s done

When it’s your turn, the menu bar calls you with ⏸

Skills, MCP, hooks — no idea what’s in force

The harness in eight lines, zeros said out loud

In practice

Ask in words. Keep the files.

Three ways people actually use it — drawn from real folders.

  1. “Draft a plan for my October ‘Using AI’ lecture.”
    + plan.md
  2. “Research cases and references to match the plan.”
    + references.md
  3. “Turn this into a two-hour curriculum with a slide outline.”
    + curriculum.md

Plan, research and draft in one folder

From the second ask on, you never re-explain the plan. It’s in the folder.

The loop

Every task leaves the next one better off.

You already work this way. What you made last week is the brief for this week — nobody re-explains it to you. Agents don’t work that way. They start from zero unless you paste. Rondo removes the pasting by making the folder the memory.

A · B

“Read the launch plan and draft a go/no-go checklist for the Aug 12 release.”

Fifteen words. You had to name the date and the file.

+ go-no-go.md

A · C

“Draft the beta banner copy.”

You didn’t mention Aug 12. You didn’t mention the gates. It already knows — they’re in the folder now.

+ banner-copy.md

A · D

“Ship it.”

Two words.

+ release-notes.md

context launch-plan.md go-no-go.md banner-copy.md CLAUDE.md

Prompt one: fifteen words. Prompt three: two. Same folder — it just knows more now.

Nothing was pasted. The artifacts are the context.

Structure

PARA by default — and the agent knows what the folders mean.

A fresh vault is scaffolded as PARA on day one: Projects, Areas, Resources, Archives. Then Rondo turns that structure into a role map the agent can query — not a naming convention it has to guess at.

  • Roles, not namesstructure.yml maps roles to paths. Rename 2. Projects to anything; the role survives.
  • The agent asks instead of you telling — a built-in MCP server, rondo-vault, rides into every session. The agent queries your tree, roles, and recent changes itself.
  • Search knows what archive means — last year’s dead project stops outranking this week’s work.
  • Already organized your own way? Switch the preset to custom and map the roles onto the structure you have. Nothing is required.
.aiworkspace/structure.ymlprofile: para
roles:
  projects: "2. Projects"
  archive:  "5. Archive"

› vault_roles{ "profile": "para",
  "roles": { "projects": "2. Projects",
             "archive": "5. Archive" } }

The harness

One hundred percent of Claude Code, visible.

Agents, skills, hooks, MCP servers, permission rules, a model override — your power tools live scattered across .claude/, .mcp.json, settings.json and your home directory, and all of it silently applies the moment you start. Rondo reads the whole stack for the folder you’re about to work in, and shows it before you type.

The harness dock: eight rows under the folder tree — agents, skills, hooks, MCP, commands, CLAUDE.md, permissions, model, each with a count
Always there, under the folder tree. Eight lines, counts included — read it before you type.
The harness drawer: the full inventory of agents, commands, skills, hooks and MCP servers, each row labeled with its source
The full inventory, one click away. Every row says where it came from — project, global, plugin, user — and opens its source.
  • Zeros are stated out loud — a folder with no hooks says 0. “Nothing will run” and “I don’t know what will run” are different sentences.
  • MCP servers show a pulse — connected, connecting, or failed, read live from the session, with a per-server tool count.
  • Install by asking — “add the Notion MCP” goes to a built-in helper that runs claude mcp add for you, behind a permission card you approve.

It’s a reader, not a launcher. Rondo shows you the harness; Claude Code runs it. We don’t hide the CLI behind a form.

Comparison

Different from the bare terminal.

Same Claude Code, same sessions, same files. What changes is what you can see.

Terminal Rondo
Re-explain the context every session The folder is the memory
Results buried in scrollback Documents open in a tab beside the agent
The harness hidden behind config files Eight lines, always visible, sourced
Subagents are one spinner A live card per worker — pop one out on top
Waiting looks exactly like thinking Fermata: a native alert and ⏸ in the menu bar
And when you do want the terminal… One click — the same session, handed to a real PTY

And that’s half of it

Four more things worth a look.

Every screen below is the real app, explained with the reason it exists — on the product page.

See the whole product, with real screens →

FAQ

Frequently asked questions

I’m not a developer. Is this for me?

Yes — that’s the point. Rondo exists for research, planning, writing, and organizing: work that isn’t code. You ask in plain language and read the results as documents. The terminal is there if you ever want it, never because you need it.

I’ve never used Claude Code. Can I start here?

Rondo runs on top of the real Claude Code CLI, so you need a Claude account and the CLI installed once. After that, Rondo is the only window you open — and it’s a gentler place to learn Claude Code than the terminal ever was.

Do I have to reorganize my existing folders?

No. PARA is the default for new vaults, not a requirement. Point Rondo at any folder you already have, switch the structure preset to custom, and map the roles onto what you run today. Folders you already have are never touched.

Where do my files go?

Nowhere. Rondo uploads nothing; your files stay plain files on your disk, and Rondo’s own state lives in one .aiworkspace/ folder. The conversations themselves are Claude Code talking to Anthropic — exactly as they would from your terminal.

Can I use it on my Obsidian vault?

Yes. Wikilinks, callouts, canvases and frontmatter render the way you expect, because they’re the same files Obsidian reads. Rondo adds one folder of its own metadata and touches nothing else.

How much does it cost?

Rondo is free during the beta. Your Claude usage is your own — whatever plan or API key you already use with Claude Code, Rondo uses the same one.

Windows? Intel Mac?

Not yet — Rondo is macOS on Apple Silicon today. If you want to hear when that changes, write to us and you’ll be the first to know.

Download

Don’t let today’s work scroll away.
Hand over the folder.

Free during beta. macOS on Apple Silicon.

macOS

Apple Silicon · v0.0.304 · 195 MB

Download Rondo

Rondo-0.0.304-arm64.dmg

  1. Open the DMG and drag Rondo into Applications.
  2. First launch: macOS blocks it once, because the beta isn’t notarized yet. Approve it in System Settings → Privacy & Security → Open Anywaystep-by-step with screenshots.
  3. Open a folder as a Vault. Your agents are home.