FrontierAgent

FrontierAgent is an open-source agent runtime, terminal product and evaluation suite for long-horizon research and file-based work, shipped as the frontier-agent TUI. Its ReAct mode runs one stateful agent that researches, reads files, writes deliverables and runs commands inside a task-scoped sandbox, while its Agent Team mode has a coordinator break the request into independent assignments, dispatch them to parallel sub-agents, and synthesize their reports.

Shell and file tools share one task-scoped filesystem with a read-only inputs directory, a working directory and a persistent outputs directory, and mutating operations show a diff and require approval unless explicitly bypassed. Sessions are checkpointed and traced locally, and the same workflow engine that powers the TUI also runs the benchmark suite used to evaluate the developer's own models.