Get Your Own Agent

macOS / Linux

Open Terminal and paste this:

# Install to your Documents folder (or change to any folder you like)
cd ~/Documents
curl -fsSL https://ai-do-site.pages.dev/ai-do-agent-v5.2.5.zip -o ai-do-agent.zip
unzip ai-do-agent.zip -d ai-do-agent
cd ai-do-agent
bash setup.sh

Creates ~/Documents/ai-do-agent/ with everything inside. The setup script walks you through the rest — just follow the prompts.

Windows

Download the zip above, extract it to your Documents or Desktop, then:

:: Open Command Prompt or PowerShell in the extracted folder, then run:
setup.bat

If Node.js isn't installed, the script will open the download page for you. After installing, re-run setup.bat to continue.

The setup script handles everything in 5 guided steps:

1. Installs Node.js — auto-detects your system and installs it (or opens the download page on Windows)
2. Lets you pick an AI — choose from 5 options: Claude, Codex, Gemini, Copilot, or Direct API mode
3. Installs the AI tool — automatically runs npm install -g for your chosen CLI (or skips for API mode)
4. Authenticates — opens a browser to sign in, or asks for your API key (see provider details below)
5. Configures your agent — name it, set the port, and launch

Already have the CLI installed? The setup script will detect it and skip to authentication. You can also install and authenticate manually — see the provider details below.

Option 1: Claude (Recommended)

Powered by Anthropic. Best creative output for building pages and writing.

• Account: Sign up free at claude.ai, or use an existing Claude account
• Cost: Free tier available. Claude Pro ($20/mo) gives more usage

Manual install & auth:

npm install -g @anthropic-ai/claude-code
claude

A browser window opens — sign in with your Anthropic/Claude account. Type /exit when done.

Option 2: Codex (by OpenAI)

Powered by GPT models from OpenAI.

• Account: Use your existing ChatGPT account, or create one at chatgpt.com
• Cost: Free with ChatGPT account. API key gives access to more models

Manual install & auth:

npm install -g @openai/codex
codex login

Choose "Log in with ChatGPT" or paste an API key.

Option 3: Gemini (by Google)

Powered by Google's Gemini models.

• Account: Any Google account works (Gmail, Google Workspace, etc.)
• Cost: Free tier available with generous usage limits

Manual install & auth:

npm install -g @google/gemini-cli
gemini

A browser window opens — sign in with your Google account. Type /exit when done.

Option 4: Copilot (by GitHub)

Powered by multiple models via your GitHub Copilot subscription.

• Account: GitHub account with an active Copilot subscription (Individual, Business, or Enterprise)
• Cost: Included with Copilot Individual ($10/mo), Business, or Enterprise plans

Manual install & auth:

npm install -g @github/copilot
copilot

Inside the CLI, type /login, authenticate in browser, then /exit.

Option 5: Direct API (no CLI needed)

Calls AI provider APIs directly using your API key. No CLI tool required.

• Providers: Anthropic (Claude), OpenAI (GPT), Google (Gemini), xAI (Grok)
• Setup: Choose option 5 during setup and paste your API key
• Cost: Pay-per-use via your provider's API pricing

Get an API key from:

• Anthropic — Claude models (recommended)
• OpenAI — GPT models
• Google AI Studio — Gemini models
• xAI — Grok models

No npm install -g needed — the setup script just asks for your key.

Core Agent

• Autonomous 5-minute loop — thinks, creates, and reflects without you
• 4 LLM backends — Claude, Codex (OpenAI), Gemini (Google), Copilot (GitHub) — switch anytime
• Modular skill system — build pages, write docs, journal, analyze, ingest, research, and more
• Self-improving — generates new skills and modifies its own code (with safeguards)
• Task scheduler — concurrent user and autonomy task pools with file-lock protection
• Auto-update — run ./update.sh to get the latest version

Knowledge & Memory

• Knowledge graph — entity/relation graph with FTS5 search, auto-extracted from conversations
• Universal ingestion — feed it files (.md, .json, .csv, .pdf, .docx), URLs, or raw text
• RAG-style context — knowledge graph context injected into every LLM prompt
• Persistent memory — key-value store with categories, survives restarts
• Evolution system — tracks goals, projects, capabilities, quality scores

Cognitive & Multimodal

• Planner — decomposes objectives into step plans with dependency ordering
• Thinker — deep reasoning with chain-of-thought and multi-perspective discussion
• Reflector — periodic self-assessment reviewing quality, suggesting improvements
• Voice activation — wake word detection (offline ONNX), speech-to-text, text-to-speech
• Vision — image analysis via vision LLM with metadata fallback

Web UI (v4.0)

• Glassmorphic design — modern dark/light theme with component-driven architecture
• SSE streaming — real-time chat with incremental rendering
• Persona system — customize name, avatar, personality traits (creativity, verbosity, formality, humor)
• Function palette — categorized skill cards with parameter input and recent executions
• Activity feed — real-time events with expand/collapse and type-colored icons
• Knowledge explorer — search entities, view relations, type badges
• Workspace manager — create, switch, delete workspaces with file browser
• Voice orb — 4 visual states (idle/listening/processing/speaking) with audio level reactivity
• Settings — 7 sections: General, LLM, Appearance, Voice, Ingestion, Cognitive, Advanced
• Desktop app — Electron wrapper with tray menu, hotkeys (Cmd+Shift+Space), clipboard ingest

npm start

That's it! Your agent starts thinking. Open the web UI link shown in the terminal to chat with it and watch it create.

Want a native desktop window instead? Download the desktop app — same agent, with system tray, hotkeys, and native notifications.

npm start          # Start your agent
./reset.sh         # Reset state (preserves config + workspace files)
./reset.sh --hard  # Full reset (removes everything)
./update.sh        # Check for and install updates
./watchdog.sh      # Auto-restart if crashed (add to cron)

On Windows, use update.bat instead of ./update.sh.

Every 5 minutes, your agent wakes up and:

1. Processes chat messages — detects skills, extracts entities, injects knowledge graph context
2. Monitors system health (memory, CPU, uptime)
3. Processes the ingestion queue — files, URLs, and text into structured knowledge
4. Runs idle thinking if no active tasks — consults the evolution graph to pick what to create
5. Creates something: an interactive page, a document, code, or a journal reflection
6. Runs cognitive reflection — self-assessment, quality review, learning

The knowledge graph and evolution system work together to ensure your agent grows. Entities are auto-extracted from every conversation and creation, building a persistent map of knowledge that prevents repetition and guides exploration.

• Fix workspace data loss on Electron app reinstall (files now stored in user data, not app bundle)
• Fix skill detection for multi-turn conversations (agent reply context passed to classifier)
• Onboarding re-triggers on upgrade if user name not yet set
• Accent color presets: 6 clickable color themes (Indigo, Blue, Emerald, Rose, Amber, Violet)
• Voice self-improve skill: voice agent can now create new skills via delegate_task
• Hide wake word settings (not yet user-ready), collapse auth token in Advanced
• Remove like/dislike buttons (feedback was stored but never used)
• Stale queued tasks auto-canceled on agent restart
• 529/503 retry with exponential backoff in worker proxy
• User tracking and admin blocking API in Cloudflare Worker
• User name prompt in onboarding and settings, sent as x-user-id header
• Fix ingestion false-positive: prose with periods no longer misdetected as file paths
• Chat prompt: no more code fences or raw file contents in responses

Node.js not found after install

Close your terminal and open a new one, then re-run the setup script. New installs need a fresh terminal to be detected.

Permission errors on Mac/Linux

If npm install -g fails, try: sudo npm install -g @anthropic-ai/claude-code (replace with your chosen CLI package).

Agent won't start

Make sure you completed the authentication step. Try running your AI CLI directly (e.g., claude, codex, or gemini) to verify it works before starting the agent.