Spacecraft

Spacecraft (sc) runs agent harnesses: YAML/JSON blueprints that describe a multi-agent pipeline as a directed graph. A blueprint plus its bindings plus the chosen model is a reproducible artifact — same inputs always produce the same run.

It’s LLM-adapter agnostic (Anthropic, OpenAI, Ollama local/cloud, vLLM, llama.cpp, SGLang), composable with Starflow for scheduled runs, and ships with sc mcp so other agents can launch harnesses over MCP.

Get started

• Getting started — install sc, scaffold a blueprint, launch it locally
• CLI reference — every sc subcommand
• Blueprint schema — fields, agent + node shapes, JS guard conditions

Adapters

• Adapter guide — pluggable inference + memory + gateway + vector DB adapters; how to write your own

Compose with Starflow

• Gateway + personal agent — always-on personal assistant pattern over Spacecraft + Starflow
• Local registry operator — registry use cases
• Harness design — design deep-dive
• Starflow CLI export — convert a Starflow pipeline into a Spacecraft harness
• Guided-RFP smoke — end-to-end scenario walkthrough

Concepts

Term	Meaning
Blueprint	A YAML/JSON file declaring agents, nodes, edges, optional memory config. Versioned and reproducible.
Bindings	Inputs + secrets the blueprint references at launch — resolved from a file, vault, or env vars.
Hangar	Local registry of built and pulled spacecraft. sc hangar to inspect.
Capability catalog	Unified view of inference adapters, memory adapters, tool types, gateways, vector DBs, depot models, and registered packages — sc capabilities.
Gateway	An inbound channel adapter (Telegram, HTTP webhook) that maps external messages into MissionSignals for an always-on agent.