Your AI subscriptions, all in one place, sharing the same context

Q: Do I need both Claude Code and Codex?

No. Winskel works with either one, signed in on your Mac. Connecting both gives Winskel more models to choose from, and lets Orchestrate split work across both.

Q: Do I pay for AI usage through Winskel?

No. Winskel runs Claude Code and Codex on your own subscription sign-ins, and it refuses a run that would bill per use unless you allow that on your Mac. Your Anthropic and OpenAI plans, their charges and their usage limits still apply.

Q: Is Winskel another code editor?

No. Winskel is a Mac app for using and coordinating the coding agents you already have. Keep your editor; Winskel works with the Git repository on your Mac.

Q: Can I choose which model works on my task?

Yes, in the Model menu beside Send. Auto lets Winskel choose one model for each message, Claude Code or Codex lets it choose within that agent, or pick an exact model and the work stays on it. Send always uses one model.

Q: Does Winskel orchestrate every task?

No. Send always uses one model. Orchestrate, a small button below the message box, splits an objective across several agents only when you press it and confirm. It may use more of your plans' usage than a single model.

Q: What happens when a provider reaches its limit?

Winskel says the provider's limit was reached instead of reporting a code failure, and the work that needs it waits. You can wait for the reset, or exclude that provider in Preferences and run the objective again on your other agent. Continuing automatically is planned, not in the current Beta.

Q: Does Winskel share memory between Claude Code and Codex?

Yes. Winskel keeps one memory per project and gives Claude Code and Codex the same brief before each run, so a task can move between them without explaining it again. Your words, what Winskel observed and what an agent only claimed stay labeled. You add facts, approve what agents propose, and can correct or forget any fact in Settings, Context and memory. Secrets are never stored in it.

Q: Is Winskel free?

Yes. Winskel is free, and there is no paid plan today. Your Claude Code and Codex plans stay with Anthropic and OpenAI.

Winskel keeps project context between your agents. Use the subscriptions you already have, with orchestration and review when you need them.

Download for MacBook a demo

Works with the Claude Code or Codex subscription you already have.Current Beta · Free · Apple Silicon, macOS 12 or later

kickoffBilling settings

Add an invoices table to the billing settings

Done 24m ago.2 files changed

1 mission ran on

Claude Code Sonnet 5.5

Tests passed on the result (9d4e2a1).

Sample project.

Built by a team from

Logos show where our team studied and worked. They don't imply endorsement.

Latest beta · Beta 32 · October 24, 2026

See your agents work, not just wait for the result.

Claude Code and Codex now show short progress updates alongside the files they read, commands they run, edits they make and checks they perform. Changes update while the task is running, and completed work folds into a cleaner result.

Beta 32 release notes

Winskel supports Codex and Claude Code

Codex
Claude Code

Keep project decisions when you switch agents

Winskel gives Claude Code and Codex the project decisions you saved before they start.

How shared memory works

Context and memory

What Winskel keeps about a project and gives to Claude Code and Codex. It works on its own: say a standing decision in chat, like “Remember that this project uses pnpm” or “Actually, use npm instead”, and Winskel keeps or updates it. Here you can check, correct, forget or limit what each agent gets.

Projectkickoff

Add a fact, like: Tests run with npm testKeep

What agents get in this project (1)

New tests go in the tests folder.
You said this. You in chat, Oct 22, 2:14 PM. Version 1. architecture. Sent 0 times.CorrectForgetClaude CodeCodex

Diagnostics

Remember that new tests go in the tests folder

Answered here · no objective created

Kept in Project Memory: "New tests go in the tests folder." Claude Code and Codex get it in this project's work.

Sample project.

Let another agent find what you missed

Have Codex review Claude Code's work, request fixes, and check the result without switching apps.

How agent reviews work

Areset.tssrc/signup/+7 -0

lines 10-16 · resetPassword

10+export async function resetPassword(email: string, token: string, password: string) {

11+ const stored = await db.resetTokens.first({ email });

12+ if (!stored || stored.expiresAt < Date.now()) return { ok: false };

13+ if (token !== stored.token) return { ok: false };

14+ await db.players.update({ email }, { password: await hash(password) });

15+ return { ok: true };

16+}

Mreset.tssrc/signup/+1 -0

lines 12-17 · resetPassword

1212 if (!stored || stored.expiresAt < Date.now()) return { ok: false };

1313 if (token !== stored.token) return { ok: false };

1414 await db.players.update({ email }, { password: await hash(password) });

15+ await db.resetTokens.delete({ email });

1516 return { ok: true };

1617}

Claude Code · Worked 1m · 2 reads, 2 edits, 1 check

Tests passed on the result (9b2f71c).

Open ReviewSecond Opinion

CodexGPT-6.1 Sol·Working00:00

Codex has started; nothing reported yet.

Codex found 1 issue, 1 that blocks the request.

HighReset token can be reusedsrc/signup/reset.ts, line 15The token is not invalidated after a successful password reset.

Fix selected issues (1)Sent to Claude Code, which built this. No new review starts by itself.

Claude CodeSonnet 5.5·Working00:00

Claude Code has started; nothing reported yet.

Claude Code · Worked 1m · 1 read, 2 edits, 1 check

Tests passed on the result (4c1e9a2).

Open ReviewSecond Opinion

Sample project.

Use several agents for larger tasks

Choose Orchestrate when parts of the work can be split, and set who plans, implements and reviews. It uses more tokens than a single model.

How Orchestrate works

OverviewAutopilotPreferencesCustom●HistoryIn use: Custom

Define how Winskel runs your work.

DescribeBuilderWorkflow code

yamljsonFormatDownload exampleImport fileExport workflow

1234567

version: 1
planning: auto

roles:
  planner:
    harness: claude_code

Valid. This is the workflow the other views show.

Unsaved changesDiscardSave workflow

Sample project.

Startup programs

AWS Activate
Microsoft for Startups
Notion for Startups
MongoDB for Startups

Winskel is free during beta

Bring your own Claude Code or Codex subscription. Winskel does not include or replace its usage allowance.

Free

Download for Mac

macOS 12 or later, on Apple Silicon.

What Free includes

Claude Code and Codex in one app
One shared memory across Claude Code and Codex
No limit on projects, active models or routes during the beta
Model: Auto, Claude Code, Codex or an exact model
Chat, Plan or Build
Your Claude Code and Codex usage at a glance
Orchestrate, when you choose it
Receipts: what ran, the checks Winskel observed, and what was not verified
Review, Apply to project and Undo apply
Preferences and Custom rules

Questions

Do I need both Claude Code and Codex?

No. Winskel works with either one, signed in on your Mac. Connecting both gives Winskel more models to choose from, and lets Orchestrate split work across both.

Do I pay for AI usage through Winskel?

No. Winskel runs Claude Code and Codex on your own subscription sign-ins, and it refuses a run that would bill per use unless you allow that on your Mac. Your Anthropic and OpenAI plans, their charges and their usage limits still apply.

Is Winskel another code editor?

No. Winskel is a Mac app for using and coordinating the coding agents you already have. Keep your editor; Winskel works with the Git repository on your Mac.

Can I choose which model works on my task?

Yes, in the Model menu beside Send. Auto lets Winskel choose one model for each message, Claude Code or Codex lets it choose within that agent, or pick an exact model and the work stays on it. Send always uses one model.

Does Winskel orchestrate every task?

No. Send always uses one model. Orchestrate, a small button below the message box, splits an objective across several agents only when you press it and confirm. It may use more of your plans' usage than a single model.

What happens when a provider reaches its limit?

Winskel says the provider's limit was reached instead of reporting a code failure, and the work that needs it waits. You can wait for the reset, or exclude that provider in Preferences and run the objective again on your other agent. Continuing automatically is planned, not in the current Beta.

Does Winskel share memory between Claude Code and Codex?

Yes. Winskel keeps one memory per project and gives Claude Code and Codex the same brief before each run, so a task can move between them without explaining it again. Your words, what Winskel observed and what an agent only claimed stay labeled. You add facts, approve what agents propose, and can correct or forget any fact in Settings, Context and memory. Secrets are never stored in it.

Is Winskel free?

Yes. Winskel is free, and there is no paid plan today. Your Claude Code and Codex plans stay with Anthropic and OpenAI.

Start using Claude Code and Codex in Winskel

Download Winskel for Mac and keep your agents, their context and their work in one place.

Download for Mac

Works with your own Claude Code and Codex sign-in. No new AI bill.Current Beta · Free · Apple Silicon, macOS 12 or later

Questions? Join us on Discord.

Winskel is independent. It is not made by, partnered with or endorsed by Anthropic or OpenAI.