[Feature] Interactive CLI Wizard for Zero-Config Onboarding

[Zepan]

🚀 The Problem

Currently, setting up picoclaw requires users to manually create or edit a configuration file (YAML/JSON) and set environment variables. For non-technical users or those deploying on a fresh embedded device (like a Raspberry Pi Zero), this friction is too high.

We want to achieve a "Zero-Config Start": if a user runs the picoclaw binary without any arguments or existing config, it should not crash or exit. Instead, it should enter an Interactive Wizard Mode.

✨ Proposed Solution

Implement a TUI (Terminal User Interface) wizard that guides the user through the essential setup steps when no configuration is detected.

The Wizard should:

1. Auto-Detect Environment: Check for existing tools (e.g., ollama), network status, and available hardware (e.g., NPU presence).
2. Step-by-Step Configuration:

• Provider Setup: "Do you want to use a local model (Ollama) or a cloud API (OpenAI/Anthropic)?"
• Token Input: Securely prompt for API keys if needed.
• Channel Selection: "Where should this agent live? (CLI, Telegram, Discord...)"

3. Generate Config: Save the inputs to a valid config.yaml and immediately start the agent.

🛠 Technical Implementation Details

1. Detection Logic

• On startup, check for config.yaml in default paths (./, ~/.picoclaw/, /etc/picoclaw/).
• If missing -> Trigger Wizard::run().

2. Wizard Flow (Draft)

Welcome to PicoClaw! 🦐
It looks like this is your first time running. Let's get you set up.

[?] Select your AI Provider:
 > Local (Ollama)
   Cloud (OpenAI)
   Cloud (Anthropic)

[If Cloud]:
 [?] Enter your API Key: ********************

[?] Select your primary Channel:
 > Terminal (Chat right here)
   Telegram Bot
   Discord Bot

[?] Configuration saved to config.yaml. Starting Agent... 🚀

✅ Acceptance Criteria

• Running ./picoclaw without a config file launches the interactive wizard.
• Users can select at least one Provider and one Channel via arrow keys.
• API Keys are masked during input.
• A valid config.yaml is generated upon completion.
• The agent starts automatically after the wizard finishes.

[mrbeandev]

Hi! I’m already working on this issue and I’m planning to open a PR soon. Could you assign it to me so others don’t duplicate effort? I’ll post progress updates here.

[Zepan]

ok, did you join the discord dev group?

[mrbeandev]

not yet , please share a invite

[Zepan]

please send an email to support@sipeed.com with the subject line: [Join PicoClaw Dev Group] + Your GitHub account. We will send the Discord invite link to your inbox.

[mrbeandev]

sent

[mrbeandev]

I've opened PR #360 with the full implementation for this feature.

Here's what I built:

Wizard Flow:

1. Environment scan — auto-detects Ollama, internet connectivity, and NPU hardware to suggest smart defaults
2. Provider selection — arrow-key menu for 7 providers (Ollama/vLLM, OpenRouter, OpenAI, Anthropic, Gemini, Zhipu, Groq)
3. Dynamic model fetching — after entering an API key, I fetch the available models directly from the provider's API and show an interactive picker. For providers with large catalogs (e.g., OpenRouter has 340+ models), there's a type-to-search filter
4. Channel selection — CLI, Telegram, Discord, Slack with token prompts
5. Config saved + auto-start — generates config.json and offers to launch the agent immediately

Key technical choices:

• Raw terminal I/O for arrow-key menus (with number-based fallback for non-interactive terminals)
• Masked input for API keys via readline
• All provider API endpoints tested against live services before integration
• 10 unit tests, all passing

Happy to address any feedback!

[PixelTux]

Hi, I’m working on the picoclaw agent chat and trying to keep the project lightweight. I’ve looked at

• fatih/color (minimal, but limited styling capabilities + 2 direct deps)
• charmbracelet/lipgloss (more expressive, but potentially heavier + 11 direct deps)

Is there a preferred package for the CLI/TUI? I’d like to stay aligned with the official direction.
Thanks!