runshift / docs
docs
runshift is the agent operating system for builder-operators. these docs cover how to run it.
what is runshift
runshift is a control tower for AI agents. not a dev tool, not an automation platform — an operator interface for people running real work with AI. you direct agents in plain language, monitor them in one view, and approve consequential actions before they happen.
it's model-agnostic. your agents can run on Claude, GPT-4o, or Gemini. runshift sits above all of them — a single control plane with a full audit trail of every decision.
how it works
relay
the orchestration agent. lives in the right panel. you direct agents in plain language through relay. relay routes commands, surfaces gate decisions, and maintains the audit trail. every action flows through relay.
Trust gates
human approval interrupts. before any agent takes a consequential action — sending a message, publishing content, logging a decision — runshift fires a gate. you approve, deny, or ask for more context. gates fire to Slack with the draft content and approve/deny buttons.
AMP
the Agent Message Protocol. the reporting contract between agents and runshift. defines what data agents pass back on every run: status, tokens used, cost, output. full spec at runshift.ai/amp.
in these docs
getting started
how to set up runshift, run your first agent, and read the deck.
the deck
outcome cards, agent status, cost tracking, and how to read the execution view.
gates
what trust gates are, how they fire, and how to act on them from Slack or the dashboard.
AMP spec
the Agent Message Protocol. the reporting contract between agents and runshift.