YOUR MODEL
Bring the intelligence.
Connect Anthropic, Gemini, OpenAI-compatible endpoints, local model servers, or local GGUF and MLX files. Change models without abandoning the session.
01
INSTALL
Install Shadow.
curl -fsSL https://blackfrostai.com/shadow/install.sh | shFirst run
shadowReconfigureshadow onboardOne self-contained binary; no Node runtime required. The installer verifies its signed manifest and checksum. Current signed binary: v10.0.3. Windows supports built-in PowerShell 5.1; PowerShell 7 is optional. Existing users can run shadow update to get the latest signed release. Review remote scripts before piping them into your shell. macOS / Linux entrypoint ↗ Windows entrypoint ↗ Reviewed source ↗
02
INSIDE
THE AGENT WITHOUT THE RENTED IDENTITY
Your model.
Your keys.
Your machine.
YOUR MACHINE
Keep control local.
Credentials and readable configuration stay under ~/.shadow. Workspace and permission controls shape how the agent works. When enabled and available, OS-level shell sandboxing adds another boundary; Linux requires bubblewrap.
YOUR INTERFACE
Stay close to the work.
Work in the Snowfall terminal: a full-width conversation, pinned composer, compact status bar and classic block-letter SHADOW banner. Guided setup uses compact, searchable provider menus and keeps your entries when a connection fails. Named sessions appear in resume and terminal titles. No Shadow account is required.
YOUR SOURCE
Inspect every layer.
Shadow is MIT licensed and developed in public. Read the source, trace the installer, change the runtime, or build it yourself.
03
BOUNDARY
LOCAL-FIRST, NOT NETWORK-BLIND
Know what leaves the machine.
Shadow does not send analytics or crash reports to Blackfrost. Network traffic can go to model providers, MCP servers, web tools, optional update checks, and release downloads you configure or invoke.
You decide the provider, workspace, permissions, and operating boundary. Shadow makes those choices visible instead of hiding them behind an account.