Skip to content
WispAgentsPublic

About

Run Claude Code, Codex and other coding agents in parallel, in the background, and read a one-screen report instead of the transcript.

Topics

Resources

Contributing

Security policy

Stars

13 stars

Watchers

0 watching

Forks

Latest commit

 

History

2,156 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

wisp

wisp

Manage the code your agents write.

Declare your coding agents as code. Run them anywhere, many at once. Track every line they write, right next to your code. From the desktop app, the terminal UI, the command line or your phone.

npm License: MIT Desktop Harnesses

Website · Docs · Guides · Get the app · wisp Cloud · Changelog

The wisp desktop app: the task list, and a finished task's report with the steps, the diff and the tests

What it is

wisp runs the coding agents you already use, Claude Code, Codex, Cursor CLI, opencode, GitHub Copilot CLI, Grok Build and Antigravity, signed in with your own plans and keys. It adds what you need once agents write much of the code:

  • Declare. Your agents, crew, machines, skills, MCP servers, rules and workflows as files in your repo's .wisp/ folder. wisp plan shows what changes, wisp apply makes it so. Review your team in a pull request.
  • Run. Any agent, many at once, each task in its own git worktree, on this computer, your other computers or in the cloud. Follow up, set a goal with judges, compare two agents, ship the pull request.
  • Track. Every task is kept next to the code it wrote. wisp blame tells you which task wrote a line and what it was asked; wisp resume picks last month's task up on today's code.

Everything works four ways, the desktop app, the phone app, the TUI and the CLI, and from your own agent through wisp's MCP tools.

Quick start

You need at least one coding agent installed and signed in (claude, codex, cursor-agent, opencode, copilot, grok or agy). Then:

npx wisp-agents                            # try it: opens the desktop app (or the browser)
npm install -g wisp-agents                 # or install the wisp command (Node.js 18+)
wisp "fix the flaky auth test"             # a task in this folder, in its own worktree
wisp tui                                   # the terminal UI

Other ways in:

curl -fsSL https://trywisp.dev/install.sh | bash   # macOS or Linux
brew install wispagents/tap/wisp-agents            # Homebrew: the command line
brew install --cask wispagents/tap/wisp            # Homebrew: the desktop app (macOS 13.3+)

wisp is in alpha: expect rough edges, and tell us what breaks.

Next: your first task · give your repo a .wisp/ folder.

Apps

Desktop app for macOS, Windows and Linux. No Node needed, and it updates itself.

macOS Windows Linux
Apple silicon · Intel x64 · Arm .deb: x64 · Arm — AppImage: x64 · Arm

Every build is on the latest release, with SHA256SUMS.txt. The builds aren't signed yet: on macOS, curl -fsSL https://trywisp.dev/install.sh | bash -s -- --app installs the app without the warning; from a .dmg, use System Settings → Privacy & Security → Open Anyway. On Windows, SmartScreen → More info → Run anyway.

Phone apps for iPhone and Android are next. Join the waitlist and we email you once, when yours is out. Today, open your cloud wisp, or a computer you put on the Grid, in your phone's browser and add it to your Home Screen.

Release channels: stable (wisp-agents), beta (wisp-agents@next) and nightly (wisp-agents@nightly, a build of main every night its full suite passes). Pick one in Settings → General → Updates.

Four ways in

Desktop app Phone app TUI (wisp tui) CLI
Start a task New task (N) New task n wisp "…" --agent codex
Follow up the reply box (/) the reply bar i wisp reply <id> "…"
Answer a question its buttons its buttons 1–9, y a n wisp reply <id> "…"
Set a goal /goal …, or ⋯ → Task options › Set goal… /goal …, or the reply bar's options → Goal /goal set --text "…" wisp goal set <id> "…"
Ship the PR Ship (S) ⋯ → Ship it… /ship ship wisp ship <id>
The changes side panel → Changes (D) the Changes pill d wisp show <id>
Your crew the rail, and Org the Crew tab g c wisp crew
Why is this line here? ⌘K → Repo history g h wisp blame file:line
Every command ⌘K Ctrl-K wisp help

The browser (wisp --browser, or your cloud wisp) is the desktop app's window, menus and keys.

Your own agent drives wisp too: wisp skill install --agent claude gives Claude Code wisp's MCP server and skill in this repo (--agent codex, --agent all). See MCP from your harness.

Guides

wisp Cloud

grid.trywisp.dev: sign in with GitHub or Google and get your own wisp on a server. Your agents keep working when your laptop sleeps, and wisp computer on puts your own computers on it too, reachable from any browser with no SSH or open ports. Run the same service for your team on your own server with wisp grid serve. More.

Speed

One Rust binary runs the server, the CLI, the TUI and the built-in plugins; a small Solid app sits on top. Measured on 19 November 2026 on an Apple M2 with spikes/server-budgets (wisp serve in its release build, 5,000 seeded tasks, fake agents; medians of 5 rounds):

measured budget
cold start → first answer 54.6 ms 150 ms
first page of the task list 0.95 ms 5 ms
an agent's step reaching 4 clients, 10 tasks running (p99) 103.3 ms 150 ms
memory at rest 101.7 MB 120 MB

The app's first screen loads 172.6 KB of JavaScript on desktop and 166.1 KB as the phone app (brotli, node tools/budgets.ts on a fresh build). CI fails a pull request that goes over 173 KB and 167 KB; the goal is 120 KB (ADR-0011, ADR-0105). The server budgets above are gated in CI too, on Linux.

Privacy and telemetry

Your prompts, code, file paths and agents' output never leave your machines; they go only to the agents you run, under your own accounts. wisp sends anonymous usage counts (versions, which agents, task durations and token counts) to PostHog. Turn them off in Settings → Anonymous usage stats, or with WISP_TELEMETRY=0 or DO_NOT_TRACK=1; they're always off in CI. The full policy: trywisp.dev/privacy.html.

Develop

docs/dev.md takes you from clone to a running, debuggable wisp, with one test per area.

npm install            # the TypeScript (UI) workspaces
npm run doctor         # is this checkout ready?
npm run dev            # wisp serve from this checkout on :4800, with the web app hot-reloading
npm run dev:desktop    # the same in the desktop window

Read AGENTS.md for the rules every change follows, ARCHITECTURE.md for the map of the code and docs/adr/ for why each decision was made. Rust for everything but the UI; TypeScript and Solid for the UI; everything is a plugin. To add a coding agent, see Build plugins.

The website (trywisp.dev) is generated by scripts/site.mjs; the docs (trywisp.dev/docs) are apps/docs over docs/site.

Contributing

Issues and pull requests are welcome. Work on a branch, keep to AGENTS.md, and add a CHANGELOG.md entry under Unreleased for anything users see.

License

MIT © M0xoo

About

Run Claude Code, Codex and other coding agents in parallel, in the background, and read a one-screen report instead of the transcript.

Topics

Resources

Contributing

Security policy

Stars

13 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages