Help
Straight answers about Winskel, the guides people need most, and how to reach a person. The current version is Current Beta.
Get started
When something is wrong
A few things to know
What is Winskel?
Winskel is a Mac app that runs Claude Code and Codex together on the Claude and ChatGPT plans you already pay for. You describe the task. Winskel picks a model for each part, runs independent parts at the same time in separate Git worktrees, asks you when a decision is yours, and brings every change back for your review with the checks it observed. See how Winskel picks a model.
Who is Winskel for?
Developers who already use Claude Code, Codex, or both, and want one workspace for them: one objective, the work assigned across their models, and one review of what changed. It runs on Apple Silicon Macs with macOS 12 or later and needs Claude Code or Codex installed and signed in. One is enough.
Which coding agents does Winskel work with?
Claude Code and Codex, through their command line tools on your Mac and under your own accounts. Winskel routes only to models your accounts can run, and their subscriptions, usage limits and charges stay separate from Winskel. See supported models.
Is Winskel partnered with Anthropic or OpenAI?
No. Winskel is an independent product. It is not made by, partnered with, or endorsed by Anthropic or OpenAI.
Does my code stay entirely on my Mac?
No. Claude Code and Codex send relevant code and context to Anthropic and OpenAI as part of their normal operation. Winskel's own storage is separate: Winskel does not store your repository's files or patches, which the app reads from your Mac when you open them. Winskel does keep the records it needs to coordinate work, including objective text, file paths, and short excerpts of agent activity that can include code. The Privacy page explains what Winskel collects.
Does routing get smarter as more people use Winskel?
Not today. Winskel records which model it chose and why, but learning from outcomes is still in development. Routing follows a fixed, versioned policy, and outcomes from other developers do not currently influence your routing.
Is Winskel faster or more efficient with tokens?
Winskel does not currently claim a measured speedup or a token savings percentage. Any performance claim needs published benchmark results and a clear explanation of what was measured.
Do agents edit the same working copy?
No. Independent missions run in separate Git worktrees, each with its own working files and branch. Their changes still need to be integrated and reviewed, and separate working copies do not guarantee that changes never conflict. A worktree is not a security sandbox. See parallel work.
Does Winskel always use multiple models?
No. Most objectives stay with one model. Winskel divides suitable work into separate missions when it helps, and several missions can run on the same model, so more agents working does not necessarily mean more models. See model per task.
Launch film
Winskel at its launch, in 57 seconds. The film is music with interface sounds and no narration; its words are below it.
Read the film's text
No narration: the film is music with interface sounds. Its words, in order:
- 0:00“coding with multiple AI tools is powerful”, then “but the workflow is broken”.
- 0:06Title: “the old way”. A Claude Code terminal and a Codex chat side by side, with context pasted from one to the other: “switch tools”, “copy context”, “repeat yourself”, “lose flow”.
- 0:13Title: “one workspace”. The Winskel app: “Claude Code + Codex”, “Claude Code and Codex in one unified platform”.
- 0:21Title: “winskel orchestrates the work”. A task is typed in the project acme-dashboard and planned into three missions: two run at the same time on Codex (GPT-5.6 Sol) and Claude Code (Sonnet 5.5), and the third waits for both, then runs on Claude Code (Opus 5.5). “pick the right model”, “split the work”, “run in parallel”, “bring it back together”.
- 0:32Title: “why it matters”. Before: “manual switching between tools”. Now: “one streamlined workflow”. Then: “one place to work”, “less manual routing”, “clearer review”, “assign tasks to your agents”, “less manual coordination”, “stay in flow”.
- 0:39Title: “review what happened”. The review shows the changes and four checks passing (typecheck, tests, lint, build): “see what changed”, “see what ran”, “review before applying”.
- 0:46Title: “bring your own subscriptions”. Winskel, the orchestration layer, over Claude Code and Codex, each on your own subscription: “works with your existing subscriptions”, “use your own Claude Code and Codex access”.
- 0:52Winskel: “One workspace for Claude Code and Codex.” “Model orchestration for coding.” “Use the right model for the right work.”