- Agent workflow
- A repeatable business process where one or more AI agents do part of the work and a human owns the decision. Unlike a one-off chat, an agent workflow has a clear input (the brief), an agent run, a human approval boundary, and an output that becomes reusable team memory.
- Workflow room
- Raltic's unit of work: a shared space for a single repeatable process where the brief, agent updates, approvals, tasks, artifacts, and final decision all stay together. It looks familiar like a channel, but it is organized around getting work done, not just messages passing by.
- Human-in-the-loop (HITL)
- A design where an AI agent can draft and propose work, but a human reviews and approves before it ships — to a customer, a repository, or a teammate. Raltic keeps this approval boundary visible inside every workflow room instead of hiding it behind an automated action.
- Approval gate
- The explicit checkpoint where a human accepts, edits, or blocks an agent's output before it has any external effect. In Raltic the gate is part of the workflow, so agents can move fast while a person stays accountable for what actually goes out.
- Cloud agent
- An agent that runs in Raltic's managed sandbox with zero local install, so a workflow can start immediately. Cloud agents suit low-risk workflows where speed matters and there is no need to keep private code or keys on a local machine.
- Bridge runtime
- A local AI runtime — such as Claude Code or OpenAI Codex — connected to Raltic through a lightweight bridge so the agent executes on your own machine. Source code and provider keys stay local; only the messages, artifacts, and run status the agent chooses to share cross into the workflow room.
See the terms in action on the
workflow templates or the
runtimes overview.