for claude code · macOS

Six live sessions.
One window.

A native Mac workshop for Claude Code - every session, chat, MCP and skill in a real interface instead of hand-edited JSON.

Download for Mac
.dmg · macOS 15+
free · win & linux soon
atelier - workshop 6 sessions
01 ▸ auth-svc
02 ▸ web
03 ▸ infra
04 ▸ docs
05 ▸ tests
06 ▸ scratch
same claude binary · same files · one window
on the workbench

The whole
Claude Code stack.

Every primitive - sessions, agents, loops, harnesses, hooks, skills, MCP, plugins - in one real interface. No hidden state, no config you can't see.

01 · sessions

Six parallel sessions

Run six side by side in one window - same binary, same files.

02 · agents

Subagents, visual

Build and route to custom agents - isolated context, per-agent model and tools.

03 · loops

Goal-driven loops

Set a goal, a verifier and a turn budget; watch it run to a finish line.

04 · harnesses

Plan → work → review

Wrap long tasks in a disciplined cycle with a progress file and checkpoints.

05 · hooks

Lifecycle hooks

Deterministic control at every event - toggle PreToolUse, Stop and more, no JSON.

06 · skills

Skills, per project

Install, pin per project, or fork a skill into its own context.

07 · mcp

MCP manager

Add, edit and toggle servers in a real UI instead of parsing JSON.

08 · plugins

Plugins & marketplaces

Browse marketplaces; bundle your setup to share with the team.

09 · history

Searchable history

Every past chat indexed - find where you solved it the first time.

10 · checkpoints

Checkpoints & control

Rewind a session to any point; set model, dir and permissions per task.

loops & harnesses · the new way to work

Stop prompting.
Start running loops.

Give Claude a goal it can check, a verifier to grade the work, and a turn budget so it can't run away. Atelier builds the loop, watches every turn, and stops at the finish line.

For longer work, wrap it in a harness - a plan → work → review cycle with a progress file and checkpoints, so a fresh context picks up exactly where the last left off.

/goal verifier model --max-turns progress file checkpoints
Run your first loop
the loop running
goaldefine done
workdoer · 1 turn
checkverifier
fail ↻ repeat · within budget
↓ pass
done✓ verified
delegation
main sessionyour context
reviewer
isolated
explorer
isolated
tests
isolated
↑ only the summary returns ↑
agents

A bench of
specialists.

Build custom agents - a reviewer, an explorer, a test runner - each with its own prompt, model and tool access. Atelier gives /agents a real editor.

Each runs in its own context and reports back a summary - the exploration noise never lands in your main session. Delegate the heavy work, keep your window clean.

a tool, not a wrapper

Same claude. Just a workshop.

Atelier drives the claude binary already on your machine, reads and writes the same files, and uses the auth you already set up.

No lock-in, no hidden state, no cloud middleman. Your config stays yours - you can still edit it by hand whenever you want.

uses your installed claude
reads / writes your files
your existing auth
config still hand-editable
no account required
the file system you care most about - treated with care.

Open the workshop.

Free for macOS. Windows and Linux are on the bench.

Download for Mac · .dmg
macOS 15+ · free · no account