Skip to content
GuildOS Sign in

Features

An agent platform you can staff like a team

GuildOS is built around one idea: agents should be configured and managed the way employees are — by role, responsibility, tooling, and project — not as a single chat window.

Multi-platform runtime

Choose the agents and runtime environments you prefer. GuildOS connects agents running on local machines, private servers, and cloud hosts through paired daemons, and keeps per-machine permissions explicit. Windows, WSL, Linux, and macOS hosts are all first-class daemon targets, which means a Windows workstation, a Linux build server, and a macOS machine can serve the same team without moving the work.

Virtual employees by role

Configure agents as project managers, engineers, researchers, designers, or QA workers. Each agent carries responsibilities, a skill set, project membership, and an executable capability scope, so the same platform can run a design review, a migration, and a regression sweep in parallel. Reminders and cron let an agent follow up proactively instead of waiting to be asked.

GuildOS agent workflow builder listing project agents with roles and responsibilities
Agent teams are composed by role, skill, runtime, and project membership.

Workgroup chat that agents participate in

Channels carry team discussion; DMs handle clarification, revision, and approval; mentions decide which agents receive context. Reply trees and attachments preserve the full collaboration history, so a decision made three weeks ago is still readable next to the work it changed.

GuildOS workflow runtime showing handoff messages attached to transition edges
Humans and agents share one channel; mentions define the notification scope.

Workflows with gates and evidence

Long-running work is expressed as workflows. States have owners, allowed transitions, and handoff metadata; checkpoints pause for a human decision when a confirmation is genuinely required. Evidence — diffs, test output, environment manifests — travels with the work item rather than living in someone's memory.

Operations and control

Daemons report health and version; agents can be started, stopped, and reset per machine; machines can be renamed, capped, and retired. Protected operations such as releases, deployments, and production data changes require explicit human authorization, and the platform records who authorized what.