# wisp

> Open-source launcher that runs AI coding agents (Claude Code, OpenAI Codex, opencode, Cursor CLI, Antigravity CLI, Grok Build) in parallel and in the background, each in its own git worktree, on this computer or other machines over SSH, and hands back a 30-second animated presentation instead of a 4,000-line transcript.

- Website: https://trywisp.dev/
- GitHub: https://github.com/M0xoo/wisp
- npm: https://www.npmjs.com/package/wisp-agents
- License: MIT
- Requirements: Node.js >= 20 and at least one supported coding agent CLI installed and signed in.

## Quick start

```sh
npx wisp-agents
npx wisp-agents "fix the flaky auth test"
npx wisp-agents --browser
```

Global install:

```sh
npm install -g wisp-agents
wisp "fix the flaky auth test"
wisp --mode plan "audit dependencies"
```

## Supported agent harnesses

- **Claude Code** (`claude`)
- **OpenAI Codex** (`codex`)
- **opencode** (`opencode`)
- **Cursor CLI** (`cursor-agent`)
- **Antigravity CLI** (`agy`)
- **Grok Build** (`grok`)

## Key features

- **Queue, not a chat log**: Sessions are grouped into `Needs you`, `Working` (live one-line status), and `Done`.
- **21 presentation blocks**: Agents finish with structured JSON blocks (`hero`, `diff`, `tests`, `stats`, `steps`, `files`, `commits`, `graph`, `tree`, `code`, `compare`, `flow`, `chart`, `terminal`, `checklist`, `callout`, `image`, `question`, `actions`, `markdown`, `next`).
- **Real diffs**: Captured directly from the agent's own file-edit tool calls.
- **Mid-run questions (`ask_user`)**: Every agent gets an `ask_user` MCP tool so it can ask clarifying questions mid-run; user answers resume the running session.
- **Terminal takeover**: Press `T` to open the session in Terminal, iTerm, Ghostty, Warp, cmux, kitty, WezTerm, or Windows Terminal with the agent's native `--resume` command.
- **Permission modes**: `Plan` (read-only), `Edits` (auto-accept file edits), `Full auto` (unattended commands, edits, and tests).
- **Learns as you go**: After a session goes quiet, a background retro records your preferences and the agent's lessons, globally and per project; later runs get them in their prompt. Review and edit them in the Memory view.
- **Rerun & updates**: Rerun any task with another harness/model (`⇧R`); an in-app Update button installs new `wisp-agents` releases.
- **Worktree per task & Ship it**: Each task runs in its own git worktree and branch; one click (`S`) commits, pushes and opens the PR, then follows CI and reviews; **Merge** squash-merges.
- **Subagents**: Give the lead agent a team on other harnesses (e.g. Claude Code delegating to Codex ×3) through a `delegate` tool.
- **Checkpoints & rollback**: Every turn is snapshotted; restore to before any turn or revert a single file.
- **Machines**: Run agents on your other computers over SSH (hosts found in `~/.ssh/config`, Tailscale and `known_hosts`); **Run on** sends one prompt to this computer and several machines in parallel, as one task with a row per machine.
- **Compare, recipes, triggers**: Run one task on two harnesses side by side; saved prompts with `{params}`; triggers on a schedule, failed CI or labelled issues.
- **Usage & phone**: Spend and success rate per project and model; questions and finished tasks on ntfy or Telegram.
- **Zero runtime dependencies**: The server uses Node.js built-ins only and stores plain JSON state in `~/.wisp`.

## Integrations & programmatic surfaces

- **MCP server**: `wisp mcp` (stdio MCP server exposing `wisp_launch`, `wisp_wait`, `wisp_get`, `wisp_reply`, `wisp_list`, `wisp_stop`, `wisp_agents`).
- **Agent skill**: `wisp skill install --agent all`
- **Scripting CLI**: `wisp "…" --json`, `wisp ls`, `wisp show <id>`, `wisp wait <id>`, `wisp reply <id> "…"`, `wisp stop <id>`
- **JavaScript SDK**: `import { WispClient } from 'wisp-agents'`
- **HTTP + SSE API**: `GET /api/openapi.json` (OpenAPI 3.1), `POST /api/sessions`, `GET /api/events`
- **Webhooks**: `PATCH /api/config` with `session.waiting` and `session.finished` events.

## Documentation

- [Landing Page](https://trywisp.dev/)
- [Integrations Guide (MCP, CLI, SDK, HTTP API, Webhooks)](https://trywisp.dev/integrations.html)
- [Supported agents: how wisp runs each CLI](https://trywisp.dev/agents/)
- [Claude Code](https://trywisp.dev/agents/claude-code.html) · [Codex](https://trywisp.dev/agents/codex.html) · [Cursor CLI](https://trywisp.dev/agents/cursor.html) · [opencode](https://trywisp.dev/agents/opencode.html) · [Antigravity CLI](https://trywisp.dev/agents/antigravity.html) · [Grok Build](https://trywisp.dev/agents/grok.html)
- [Guides](https://trywisp.dev/guides/)
  - [Run Claude Code and Codex in parallel](https://trywisp.dev/guides/run-claude-code-and-codex-in-parallel.html)
  - [Run multiple Claude Code agents in parallel](https://trywisp.dev/guides/run-multiple-claude-code-agents-in-parallel.html)
  - [Run coding agents on multiple machines (SSH)](https://trywisp.dev/guides/run-coding-agents-on-multiple-machines.html)
  - [Move a task to another machine or agent mid-way](https://trywisp.dev/guides/move-a-task-to-another-machine-or-agent.html)
  - [Multi-agent coding: a lead delegating to other harnesses](https://trywisp.dev/guides/multi-agent-coding.html)
  - [Claude Code vs Codex, on your own repo](https://trywisp.dev/guides/claude-code-vs-codex.html)
  - [Run coding agents in the background](https://trywisp.dev/guides/run-coding-agents-in-the-background.html)
- [Changelog](https://trywisp.dev/changelog.html)
- [README](https://github.com/M0xoo/wisp/blob/main/README.md)