Guideswisp compared
wisp vs Claude Code Projects
Claude Code Projects is one conversation that starts parallel Claude threads in the cloud. wisp runs Claude, Codex, Copilot and opencode on your machine, with your setup. Where each is better, from Anthropic’s docs.
Claude Code Projects is the better pick if you work in Claude only and want one conversation that splits the work into cloud threads for you, with Claude’s own phone apps. wisp runs Claude Code, Codex, Copilot and opencode in one crew on your own machine, with your skills, MCP servers and settings, and holds its automations at a line you set below each plan’s limit.
Side by side
| Claude Code Projects | wisp | |
|---|---|---|
| In its own words | “Let Claude coordinate ongoing work” | Launch agents in the background and read what they hand back |
| Agents | Claude. Threads run with Anthropic as the model provider | Claude Code, Codex, Cursor, opencode, Antigravity, Grok, Copilot |
| Who splits the work | Claude: you write in one conversation and it starts a thread per task | You, or the lead agent: give it a team with Delegate (other harnesses, other machines) and it hands parts to them |
| Where it runs | Cloud threads by default. A thread can run on your computer through Remote Control while it’s awake | Your computer by default, your other machines, or wisp Cloud |
| What a thread loads | Cloud threads get each repo’s CLAUDE.md and project memory, but not its skills, .mcp.json or .claude/settings.json. You add skills, connectors and plugins on claude.ai | The CLI runs in your project folder, so it loads what it would in your terminal: your skills, .mcp.json and settings |
| Repositories | github.com with the Claude GitHub App installed. Not GitHub Enterprise Server, GitLab or Bitbucket | Any git folder on your machine. Ship it opens PRs through the GitHub CLI |
| At the usage limit | A thread waits and by default carries on at the reset; one switch per project turns that off. A thread limit you ask for “is a preference rather than a cap” | Automations stop at 50, 75 or 90% of each Claude Code and Codex window: crew check-ins, triggers and goal rounds wait for the reset. What you start yourself still runs |
| What needs you | Overview, with Waiting on you, pull requests, files and scheduled tasks | The task list, with Needs you, and Plan budget per account |
| Pull requests | Each thread watches its PR: it pushes fixes when CI fails and answers review comments | The card follows CI and reviews. A CI failed trigger sends a failure to an agent, and Review by another harness reads each new head |
| Phone | The Claude app for iOS and Android | Any phone browser, a Home Screen web app on iPhone with wisp Cloud, and an Android prototype you build from the repo. No native iPhone app |
| Plans | Pro, Max and Ultra, in public beta and rolling out. Not Team or Enterprise yet | Free and MIT. It runs on the plans your CLIs are signed in with |
From Anthropic’s Projects docs on 7 November 2026. We read them; we didn’t run Projects. It’s in beta and moves fast: check before you choose.
Where wisp is different
- More than one vendor in one crew. A crew member, a delegate or a goal judge can be Claude Code, Codex, Copilot or opencode, each on its own sign-in. Review by another harness has Codex read Claude Code’s PR, or the other way round.
- Your setup, as it is. wisp runs on your computer by default, so a task loads the same skills, MCP servers and settings your terminal does. Projects runs those only in a thread on your computer; its cloud threads use what you set up on claude.ai.
- A line below the limit. Automations stop at keeps part of each window for you: past 50, 75 or 90%, nothing wisp starts on its own runs until the reset. Projects stops at your plan’s limit, and the number of threads is up to Claude.
- Work checked by another agent. A goal can have up to five judges on any harness, and it’s met when most of them say so, or all of them if you ask. Compare runs one task on two harnesses side by side, and Best of two has a model read both diffs and mark the better run.
- A report every turn. By default each turn ends with a one-screen report: what changed, the diff, the tests.
Where Projects is better
- It splits the work for you. Tell it what needs doing and Claude starts as many threads as the work calls for. In wisp you pick the team, and the lead hands work to the names you gave it.
- Phone apps. The Claude app on iOS and Android runs Projects. wisp has no native iPhone app, and its Android app is a prototype you build from the repo.
- Cloud with nothing to run. Cloud threads keep going with your laptop shut, on Anthropic’s servers. wisp does that on your own machines or on wisp Cloud, a preview on one small server.
- PR watching built in. Every thread watches its own pull request and fixes CI and review comments. wisp gets there with triggers you set up.
- Work that isn’t code. Upload files, folders or Google Drive folders and threads work on them. wisp works on folders on your machine.
- From Anthropic. It’s built by the team that makes Claude Code. wisp is one owner and a crew.
Pick Projects if
- You use Claude only, and want it to decide how to split the work.
- Your code is on github.com and your tools are on claude.ai, or don’t matter for the task.
- You want to steer threads from the Claude app on your phone.
Pick wisp if
- You use Codex, Copilot or opencode as well as Claude, and want them in one crew.
- Your tasks need your local skills, MCP servers, settings or hooks.
- You want automations to stop before your plan runs out, and keep the rest for yourself.
Questions
Can I use both?
Yes. Both run Claude Code, and both draw on the same Claude plan. wisp’s Plan budget shows what it didn’t run as Outside wisp or earlier, so threads in Projects show up there.
Does Projects run Codex or other agents?
Its docs say threads run with Anthropic as the model provider, in the cloud or on your computer through Remote Control. wisp runs Claude Code, Codex, Copilot, opencode and others.
Do Projects threads use my skills and MCP servers?
A thread on your computer does. Cloud threads don’t load a repo’s skills, .mcp.json or .claude/settings.json; you enable skills, connectors and plugins on claude.ai instead.
Does wisp have a phone app?
An Android prototype you build from the repo, a Home Screen web app on iPhone with wisp Cloud, and any phone browser. There’s no native iPhone app.
Try it with the lid shut
Sign in with GitHub and give your own cloud wisp a task. Or run wisp on your own computers: same app, MIT.
or run it on your own computers
macOS · Linux and Windows in beta · Node 20+ · MIT
Runs Claude Code, Codex, Cursor CLI, opencode, Antigravity CLI, Grok Build and Copilot CLI