Get Started

Welcome to Wispy

Wispy is autonomous AI agent infrastructure powered by Gemini. It lives in your terminal, your messaging apps, and your APIs with 120+ built-in tools spanning file system, web, browser automation, vision, sessions, process management, skills.sh registry, code execution, memory, communication, media, blockchain, cross-model delegation, and 42+ integrations.

Whether you're a developer automating workflows, a team integrating AI into Slack or Telegram, or an explorer building on-chain identity with ERC-8004, Wispy adapts to your context and gets work done.

Tip

New here? Jump to Quickstart to have Wispy running in under two minutes.

Quickstart

1
Install Wispy
bash
```
npm install -g wispy-ai
```
Or use Homebrew: brew tap brn-mwai/wispy && brew install wispy
2
Run the setup wizard
bash
```
wispy onboard
```
3
Start chatting
bash
```
wispy chat
```

Try Marathon Mode

bash

wispy marathon "Build a full-stack SaaS dashboard"

Info

Requires Node.js 20+. Wispy uses Gemini 2.5 Pro by default. You'll need a Google AI API key or Vertex AI credentials.

Your First Chat

Once installed, start an interactive REPL session with wispy chat. Wispy will greet you and await your instructions.

bash

$ wispy chat
☁️👀 Wispy v1.0.0 — ready.

You: Summarize the files in this directory
Wispy: I see 12 files. Here's a summary...
  - package.json — Node.js project config
  - src/index.ts — Entry point with Express server
  ...

You can also send a single prompt without entering the REPL:

bash

wispy agent "What's the weather in Lagos?"
wispy agent -s my-project -t high "Refactor this codebase to TypeScript"

Info

Use Ctrl+C to exit the chat session. Use -s to name a session and -t to set thinking level.

Core Concepts

Wispy is built around a few key ideas:

Sessions

Every conversation is a session. Sessions persist as JSONL files in ~/.wispy/sessions/, preserving message history and context. Sessions can be main, cron, group, sub, or heartbeat type, with role-based access control.

Tools

Wispy ships with 120+ built-in tools across file system, web, browser engine (39 automation tools), vision, webcam, sessions, process management, skills.sh registry (67K+ community skills), code execution, memory, communication, media, cross-model delegation, blockchain, and identity. Tools are automatically selected based on your prompt, with dangerous commands blocked by default.

Channels

Channels are the interfaces through which you interact with Wispy: CLI, Telegram, WhatsApp, Discord, Slack, Signal, Matrix, REST API, WebSocket, Web dashboard, or Agent-to-Agent (A2A).

Skills

Dynamically loadable skill packs that extend Wispy with focused capabilities: research, codegen, documents, web3, browser, cron, voice, images, a2a, content, twitter, and chainlink. With v1.6, Wispy also connects to the skills.sh registry with 67,000+ community skills you can install with a single command.

Marathon Mode

For multi-step, long-running tasks. Wispy uses ultra thinking (24K tokens) to plan, then executes milestones with checkpointing, loop detection, and auto-recovery. Think of it as giving Wispy a project rather than a prompt.