Getting Started

This guide focuses on the daily usage flow. For command details, see commands.md.

1) Launch Spettro

./bin/spettro

On first launch in a folder, Spettro shows a trust dialog:

• Yes, trust this session
• Yes, and remember this folder
• No, exit

2) Configure model access

Use:

• /connect to add provider API key or a local endpoint
• /models to select provider/model

Spettro supports:

• Native Anthropic API.
• OpenAI-compatible providers through provider-specific base URLs.
• Local OpenAI-compatible endpoints (for example LM Studio/Ollama).

Model metadata is loaded from https://models.dev/api.json and cached locally.

3) Work with the agent system

Spettro starts with default_agent from manifest (default: plan).

• Explore first with glob/grep/file-read.
• Spawn sub-agents in parallel via the native agent tool (multiple calls in one response run concurrently).
• Switch mode/agent with Shift+Tab or /mode.

4) Approval flow

If permission is ask-first:

1. Generate a plan.
2. Review and refine if needed.
3. Run /approve to execute through the coding agent.

5) Persist and resume conversations

• /compact summarizes long threads (optionally focused).
• /clear saves current conversation and clears the active thread.
• /resume loads a prior saved conversation for the same project.

6) Customize

Edit spettro.agents.toml and agents/*.md (for example agents/docs-writer.md), then restart Spettro.