AI agents and models
Runstead can use locally installed AI command-line tools and API-based models. It does not supply AI credits or sign in to those tools for you.
Command-line agents
Section titled “Command-line agents”Install and sign in to the command-line agent you want to use, such as Codex, Claude Code, or another supported harness. Configure a named agent in the project, select its harness and model, and check that Runstead can find the executable.
Add an Ask an AI agent task to a workflow, choose the agent, and enter its prompt and context. Review the working folder and permissions. An agent can act on files and tools available to its process.
API models
Section titled “API models”Configure a named model with its provider and model ID. Put the provider API key in a project secret and select that secret as the model’s credential. For a local model server, provide its reachable base URL.
Inspect a run
Section titled “Inspect a run”Use Tools → Review AI prompts to inspect the authored prompt before running. After a run, open Runs & logs to review the recorded prompt and task output.
AI prompts and context are sent to the provider or tool you select. Provider billing, retention, and privacy rules apply. A local Runstead workspace does not mean every task stays offline.