Get Started in 2 Minutes

Install curie-agent, configure your provider, and start coding—all from your terminal.

npm install -g @curie-agent/cli

Or try without installing: npx @curie-agent/cli

✓ Prerequisites

Node.js ≥ 20

curie-agent runs on Node.js 20 or later. Check with node -v.

npm / pnpm / yarn

Any package manager works. npm is bundled with Node.js.

API Key

An API key for your chosen provider (Anthropic, OpenAI, Google Gemini, OpenRouter). Not needed for local/Ollama.

1 Install

Install the CLI globally:

npm install -g @curie-agent/cli

Verify the installation:

> curie-agent --version
0.3.0

2 Choose Your Mode

curie-agent offers three ways to work. Pick the one that fits your workflow:

Default

Web Dashboard

curie-agent

Starts a local daemon on port 3457 and opens your browser. Full-featured dashboard with chat, subagents, stats, and sessions — works from any device on your network.

Terminal UI

curie-agent tui

Interactive Ink-based TUI with 6 tabs, 32+ slash commands, and thinking streaming (Ctrl+O). Runs in your terminal, connected to the daemon.

Headless

curie-agent -p "your prompt"

One-shot mode: sends a prompt and prints the answer to stdout. Great for scripts and CI pipelines.

3 Web Dashboard First Run

When you run curie-agent, the browser opens to the dashboard. Here is what happens next:

Authorization Screen

The dashboard asks for a daemon access token. If you launched via curie-agent, the token is passed automatically in the URL. If you open the URL manually, find it in ~/.curie-agent/daemon.token or pass it as ?token=....

Setup Wizard (First Time Only)

If no provider is configured, a 12-step wizard appears automatically. It walks through: provider selection (Anthropic, OpenAI, Google Gemini, Ollama, Local, OpenRouter), API key (skipped for local providers), model name, assistant identity (name and personality), your profile (name, timezone, languages), and a final review screen.

Identity Files Created

The wizard calls the daemon's identity.setup RPC method, which creates SOUL.md, USER.md, AGENTS.md, MEMORY.md, TOOLS.md, and HEARTBEAT.md in ~/.curie-agent/. Settings are saved to ~/.curie-settings.json.

Chat View Ready

The dashboard shows a greeting message. Type your first prompt in the input area at the bottom, or press Ctrl+K to open the command palette with 30+ pre-defined commands.

4 Terminal UI First Run

If you use curie-agent tui, the init wizard starts automatically on first launch:

> curie-agent tui

Welcome to curie-agent! Let's get you set up. This takes about 2 minutes.

Which LLM provider do you want to use?

  1) Anthropic

  2) OpenAI

  3) Local (OpenAI-compatible)

  4) OpenRouter

  5) Ollama (Local)

Type the provider name or number:

The wizard guides you through provider selection, API key, model, assistant identity, and your profile. On completion, identity files and skills are created in ~/.curie-agent/.

5 Configure your identity

Edit the identity files created by /init to define who your agent is and what it knows about you. These are plain Markdown — open them in any editor.

~/.curie-agent/SOUL.md

# SOUL.md
Name: Curie
Archetype: Efficient Assistant
Voice: Concise, direct, no filler
Red lines: Never exfiltrate data

~/.curie-agent/USER.md

# USER.md
Name: Alex
Timezone: Europe/Warsaw
Role: Software Engineer
Goals: Ship curie-agent 1.0
Preferences: TypeScript, no Java

The agent reads both files at every session start. The more context you add to USER.md, the less you have to repeat yourself across sessions.

6 Enable heartbeats

Heartbeats make your agent proactive — it wakes at scheduled times, reads your memory and TODOs, and sends you a briefing. Run these in the TUI:

Morning briefing /heartbeat daily 07:15

Nightly synthesis /heartbeat dreaming 23:01

Activate /heartbeat enable

To receive briefings on Telegram, connect your bot first:

Set bot token /channels set-bot-token <your-bot-token>

Set your user ID /channels set-user-id <your-telegram-id>

Get a bot token at @BotFather on Telegram. Your Telegram user ID can be found at @userinfobot. Once configured, briefings arrive at your scheduled time.

7 Start using every day

You're set up. Here's how to get productive quickly:

Slash command reference

Full reference for all 32+ commands — /todo, /wiki, /skill, /context, /heartbeat and more

Daily usage guide

How to use deep research, planning, memory, and Telegram together every day

Try these now

/skill — list built-in skills
/todo list — your task board
/deep-research <topic> — start researching

☆ What Gets Created

The init wizard creates the following files in ~/.curie-agent/:

~/.curie-agent/
├── SOUL.md # Your AI's personality, voice, operational stance
├── USER.md # Your profile, timezone, preferences
├── AGENTS.md # System directives, workspace rules, capabilities
├── MEMORY.md # Long-term memory: lessons, projects, milestones
├── HEARTBEAT.md # Proactive routine definitions
├── TOOLS.md # Local environment registry
├── settings.json # Provider config, API keys, heartbeat schedule
├── tasks.json # Unified task store
├── wiki/ # Knowledge base (created by wiki init)
│   ├── WIKI.md # Schema and workflow procedures
│   ├── index.md # Content catalog
│   ├── raw/ # Immutable source files
│   └── pages/ # Entity, concept, and summary pages
└── skills/ # Built-in skills
    ├── deep-research/
    │    ├── SKILL.md
    │    └── references/ # Research methodology guides
    ├── planning/
    │    ├── SKILL.md
    │    └── references/ # Planning templates and frameworks
    └── wiki/
        └── SKILL.md # Ingest/Query/Lint workflow skill

Identity Files

Each file is a rich markdown template—not a minimal stub. AGENTS.md defines the agent's full system directives (memory operations, task management, safety red lines). HEARTBEAT.md outlines detailed Intraday/Daily/Weekly/Monthly/Dreaming routines.

Settings

settings.json stores your provider API keys, model choice, heartbeat schedule, safety guard toggles, and theme preferences. Update it directly or use slash commands (/model, /theme, /heartbeat).

☆ Built-in Skills

curie-agent ships with three pre-installed skills. The agent automatically loads the right skill when your request matches:

deep-research

Structured multi-source research methodology. 4-phase execution process, source evaluation criteria, 5 output formats (summary, comparison table, deep report, annotated sources, fact-check), and domain-specific guidance for medical, legal, financial, scientific, and technical topics.

research reports fact-checking

planning

Universal planning across any domain: project roadmaps, event timelines, goal decomposition, habit building, and content calendars. Includes 7 planning modes, prioritization frameworks (Impact × Urgency, RICE), and output templates for every scenario.

projects events goals

wiki

Persistent, compounding knowledge base. Three workflows: Ingest (source → structured pages), Query (natural language → cited answer), Lint (health check: orphans, broken links, stale claims). Use via TUI natural language, /wiki slash command, or curie-agent wiki CLI verbs.

ingest query lint

Adding Custom Skills

Create your own skills by adding SKILL.md files to ~/.curie-agent/skills/ (global) or .curie-agent/skills/ (per-project). Each skill needs YAML frontmatter with name and description, and the agent auto-discovers them at startup.

# ~/.curie-agent/skills/my-tool/SKILL.md

---

name: my-tool

description: >-

 Use this skill when the user needs to do X...

---

# Your skill instructions here...