Skip to main content

First Launch

What happens when you launch Sparky for the first time.

Sparky keeps everything inside the ~/.sparky/ folder, all data, models, and dependencies live there. Nothing is installed system-wide, and removing the folder cleanly uninstalls everything. On first launch, a few dependencies need to be downloaded into this workspace.

Splash Screen

On first launch, Sparky downloads the required dependencies:

  • Embedding model (~78 MB) — powers the local knowledge base via llama.cpp

The splash screen shows download progress for each dependency. This only happens once.

Onboarding

After the splash screen, Sparky guides you through two steps:

  1. Profile — Set some basic profile information like display name, preferred language, and timezone.
  2. LLM Connection — Connect your first AI provider, at least one provider has to be connected.

Sparky supports the following providers out of the box:

ProviderAuth Method
Anthropic ClaudeAPI key (OAuth Pro/Max discontinued since 04.04.2026)
OpenAIOAuth (ChatGPT) or API key
GitHub CopilotDevice login
Google GeminiOAuth
Google AI StudioAPI key
MistralAPI key
OllamaLocal — no auth needed

The connection flow is simplified, the entire authentication process is managed by Sparky. Credential storage, token refresh, and session management are all handled automatically. No need to worry about expiring tokens or manual re-authentication, once connected, it just works. Connect your agent and enjoy.

Local Storage

All data is stored locally on your machine within the ~/.sparky/ folder:

  • Workspaces — Each workspace, ~/.sparky/workspaces/, has its own folder with chats, knowledge base, and settings
  • Conversations — SQLite database, folders for attachments
  • Secrets — OS keychain (Keychain on macOS, DPAPI on Windows)
  • Models~/.sparky/models/
  • No telemetry, no cloud sync, no account required