Most agent tooling today is a hosted, opaque service that trades control for convenience. As teams experiment with dozens of coding agents and mixed providers, the hard problems become orchestration, isolation, and reproducible state — not model quality. mngr flips that tradeoff: it treats agents as first-class, Git-like resources you run on compute you control.
What Sets It Apart
- Git-like agent lifecycle semantics — create, clone, snapshot, push/pull, and fork agent state so agent environments and histories are reproducible across hosts. This makes experimentation auditable and portable (so what: reduces drift when iterating on agent-driven changes).
- Provider-agnostic, SSH-native architecture — agents run in tmux sessions on any SSH-accessible host (local, Docker, Modal, etc.). Built on familiar primitives (SSH, git, tmux), so teams avoid a locked-in managed control plane (so what: lowers onboarding friction and improves security posture).
- Cost- and security-minded defaults — agents auto-shutdown when idle, support network allowlists, and use per-host SSH isolation (so what: easier to reason about inference spend and safer to run untrusted agents).
- Developer-first UX — a terse CLI surface (create/list/connect/message/exec) with tab completion and programmatic messaging makes it easy to script and integrate agents into CI or local workflows (so what: faster iteration and straightforward debugging via direct SSH into agent sessions).
Who It's For and Tradeoffs
Great fit if you are a developer or small team who: wants to run many coding agents without a managed SaaS; needs reproducible agent state or the ability to snapshot/fork agent work; prefers simple, scriptable tooling built on SSH/git/tmux.
Look elsewhere if you need a polished multi-tenant web UI, enterprise-grade role-based access controls out of the box, or a fully managed hosted orchestration platform — mngr intentionally avoids a central managed control plane and relies on host-level tooling and provider integrations (tradeoff: more operational responsibility in exchange for control and transparency).
Where It Fits
Use mngr as the orchestration layer when you value provable isolation, reproducible agent state, and local-first experimentation. It pairs well with containerized providers (Docker, Modal) and can slot into CI pipelines or developer workstations where SSH access and simple lifecycle scripts are preferred.
