What changed
CompozyOS positions itself as an operating system for AI agents, aiming to simplify the management of autonomous agent sessions. The core problem it addresses is the bottleneck that arises not from the AI models themselves, but from the difficulty of managing concurrent agent sessions. When running tools like Claude Code, Codex, and Cursor in parallel, users often face challenges such as multiple terminals, accidental prompts to the wrong agent, and a lack of clear oversight on which agent requires attention. CompozyOS replaces the need for manual gluing of loops, cron jobs, webhooks, and approval systems by providing a built-in framework for these functionalities.
Installation and Setup
Installation can be done via a curl command that fetches a verified installer script, which pins the latest release and checks Sigstore provenance. Alternatively, CompozyOS can be installed via Homebrew, npm, Go, or built from source. The CLI is compatible with macOS, Linux, and Windows. After installation, users can start a session by picking a provider, starting the local daemon, and creating a new agent session. The current experience is described as technical, involving a terminal, a local daemon, and a config.toml file.
Session Management and Visibility
CompozyOS organizes agent sessions into distinct states: Needs you, Working, Waiting, In review, or Done. The 'Needs you' state is highlighted to draw immediate attention. Key features for session management include:
- Worktrees: Enable parallel agents to work on their own checkouts of the same repository.
- Session Context: Branches, pull requests, and their associated checks are directly linked to the session.
- Handoff: Allows a session to be continued by another agent CLI when the current one concludes.
- Approvals: Pauses tool calls until human approval is granted, bringing waiting work to the forefront.
Autonomous Agents and Workflows
Agents in CompozyOS are designed to operate autonomously, checking in on their own schedules. An always-on agent has a defined name, a visual representation, and a primary session per project. These agents can review work on a schedule without incurring model costs if no action is needed. They can be pointed to sources like new GitHub issues or approved scripts. Reports are consolidated into a main session card, and proposals for actions can be automatically executed up to a daily limit set by the user. Agents maintain notes on their purpose, and nothing is extracted without user awareness. Check-ins require models like Claude, Codex, or OpenCode.
Inter-Agent Communication and Loops
Agents can hand off work to subagents, monitoring their progress live. An agent can complete its turn and be scheduled to wake up when specific events occur, such as the completion of a pull request's checks or the exit of a command, eliminating the need for constant terminal monitoring. Loops are defined as named contracts with a goal, a verification method, and a termination condition, designed to run unattended and end honestly. Loops are bounded by design with iteration caps and can include human gates for approval. They support seven named endings: done, no-op, blocked, failed, exhausted, stalled, and canceled. Loops can be run and managed via the web, CLI, and HTTP.
Scheduling and Event Triggers
CompozyOS integrates scheduling, runtime triggers, and signed webhooks into a single management point. Each run is recorded, and completed runs are automatically removed from the session list. This allows for scheduled briefings, such as a weekday morning summary, to be prepared without manual oversight.
Supported Integrations
CompozyOS supports running existing agent CLIs as durable sessions. It integrates with a range of agent CLIs including Claude Code, Codex, Cursor Agent, Gemini CLI, GitHub Copilot CLI, and others. It also connects with model providers and gateways such as OpenRouter, Vercel AI Gateway, and xAI. User subscriptions and API keys remain private, as CompozyOS does not resell model access.
Built-in Features
The system includes several built-in components to enhance agent operations:
- Sessions: Agent work that persists beyond terminal sessions, with capabilities for forking, continuing with other agents, rewinding to checkpoints, and mid-turn steering.
- Files: Tools for browsing and searching projects, and reviewing session diffs.
- Terminal: Runtime-owned shells that can be opened, watched, and handed off to agents.
- Desktops: Persistent environments with tabbed interfaces, side-by-side splits, a command palette, and global hotkeys.
- Workspaces: A view of all projects, enabling work without a specific project context, and providing parallel agents with their own worktrees.
- Tasks: Durable work units that agents claim, heartbeat, and close with evidence.
- Profiles: Allow for separate work organization without acting as security boundaries.
- Gateway: A daemon that is initially reachable only on the local machine, with remote access configurable.
Use Cases
CompozyOS supports various use cases, such as triaging new GitHub issues, delivering specs while away, reviewing code changes until clean, preparing daily briefings, managing release checks as claimed work, and requiring human approval before sensitive actions.
Extensibility
The platform is designed to be built upon, with a Marketplace for extensions, plugins, and MCP servers. Extensions can add skills, agents, tools, hooks, and check-in sources. Development for extensions can be done in Go or TypeScript, with React for views, using provided SDKs.
Why it matters for builders
CompozyOS offers a structured approach to managing the growing complexity of AI agent workflows. By abstracting away the manual orchestration of multiple agent sessions and providing built-in features for scheduling, approvals, and inter-agent communication, it allows builders to focus more on agent logic and less on infrastructure. This can significantly accelerate development cycles and improve the reliability of autonomous agent systems.
Practical impact
Builders can begin by installing CompozyOS using the provided curl command or package managers. Experiment with running existing agent CLIs like Claude Code or GitHub Copilot CLI within CompozyOS sessions to experience improved visibility and control. Explore the session states and try setting up simple loops or scheduled check-ins for an agent to observe its autonomous behavior. For those developing agent-based tools, investigating the extension SDKs in Go or TypeScript can enable integration with the CompozyOS ecosystem.
Caveats and source limits
The provided source material describes CompozyOS as a GitHub project with a recent release (v0.7.7) and significant community traction (2900 stars, 189 forks). However, specific details regarding pricing, independent performance benchmarks, or a definitive release date for a stable, production-ready version are not present. The current user experience is noted as technical, implying that further refinement may be needed for broader adoption. The source also mentions experimental surfaces for connecting external apps.
Sources
Claim check: 12/12 supported claims - 12 evidence links - 100% avg confidence
- CompozyOS is an operating system for AI agents and an autonomous control plane.supported - github.com
- CompozyOS allows users to see and steer every Claude Code, Codex, and Cursor session.supported - github.com
- CompozyOS provides built-in loops, schedules, permissions, and approvals for agent orchestration.supported - github.com
- CompozyOS can be installed via a verified installer script, Homebrew, npm, Go, or from source.supported - github.com
- The CompozyOS CLI runs on macOS, Linux, and Windows.supported - github.com
- CompozyOS organizes agent sessions into states: Needs you, Working, Waiting, In review, or Done.supported - github.com
- CompozyOS supports agents that check in on their own schedules.supported - github.com
- Agents in CompozyOS can hand work to subagents and monitor their progress live.supported - github.com
- CompozyOS integrates scheduling, runtime triggers, and signed webhooks.supported - github.com
- CompozyOS supports agent CLIs including Claude Code, Codex, Cursor Agent, Gemini CLI, GitHub Copilot CLI, and others.supported - github.com
- CompozyOS integrates with model providers and gateways like OpenRouter, Vercel AI Gateway, and xAI.supported - github.com
- CompozyOS includes built-in features such as durable sessions, file browsing, integrated terminals, persistent desktops, and workspaces.supported - github.com
Caveats
- Single-source caution: verify critical details at the linked source.
Radar score 88/100 - how it was calculated
- Reliability 82: GitHub metadata supports source trust
- Freshness 95: Fresh GitHub release date
- Novelty 77: Fresh GitHub release
- Technical 89: Repository technical metadata
- Developer 96: Developer tooling signals
- Ecosystem 72: Fresh GitHub release
- Confidence 100: Claims have reliable evidence
Discussion
Loading comments...