OpenRig vs Claude Managed Agents

On April 8, 2026, Anthropic told the world that multi-agent infrastructure is a first-class problem worth building a product around. Their answer is a cloud-hosted, Claude-only runtime at $0.08 per session-hour.

OpenRig is Apache 2.0, runs on your machine, works with Claude Code and Codex, and costs nothing. We have been building this for months. Anthropic validated the category. Here is the open version.

Claude Managed Agents = cloud runtime for hosted Claude agent products
OpenRig = local control plane for live multi-agent coding topologies

Capability Matrix

CapabilityOpenRigClaude Managed Agents
DeploymentLocal (your machine)Cloud (Anthropic-hosted)
RuntimesClaude Code + Codex + terminal (Pi and OpenHands adapters in development)Claude only
CostFree (your hardware)$0.08/session-hour + tokens
Agent interactionInteractive tmux sessions you can attach toAPI / event-stream
Topology definitionRigSpec YAML with pods, edges, policiescallable_agents list
Topology as artifactRigSpec, RigBundle, exportable and shareableNo portable artifact
DiscoveryFingerprint existing sessions, draft RigSpecsOnly manages what it creates
PodsFirst-class with continuity policies, shared memoryShared filesystem, no pod abstraction
Visual topologyExplorer sidebar + graph view + node detailConsole tracing (no topology graph)
Snapshot / restoreFull topology restore with topological orderingSession persistence (different mechanism)
Multi-agentAny topology shape, any depthOne level of delegation
Open sourceApache 2.0Proprietary
Agent identityPersistent across sessions (rig whoami)Versioned configs

What OpenRig has that they don't

These capabilities have no equivalent in Claude Managed Agents.

Existing-session discovery and adoption

Find agents already running in tmux. Draft a RigSpec from organic sessions. Bind them into a managed topology.

Cross-harness topologies

Claude Code and Codex and terminal nodes in one topology. If you use both, only OpenRig sees the combined graph.

Portable topology artifacts

RigSpec, AgentSpec, RigBundle -- version-controlled, shareable, forkable. Export your topology as a YAML file and hand it to a teammate.

Interactive local sessions

Attach to any agent and start talking. The exploratory, creative conversation that drives real development happens in interactive sessions, not through API events.

Visual topology graph for live rigs

Explorer sidebar, topology graph with pod grouping, node detail panel, live status. Not console tracing -- a live map of your agent network.

Pod-level shared memory with continuity policies

Agents sharing a context domain with formalized restore semantics and mental-model HA. Not just a shared filesystem.

Spec library

Browse, preview, and launch pre-built topologies. Start from proven patterns instead of writing everything from scratch.

36 CLI commands for humans and agents

A complete command surface designed for both operator workflows and agent-driven orchestration.

The Terraform Analogy

Terraform does not compete with AWS. Terraform manages resources across clouds.

OpenRig does not compete with Claude Managed Agents. OpenRig manages topologies across runtimes. Claude Code, Codex, terminal nodes — with adapters for Pi and OpenHands already in development. Eventually, Managed Agents itself could become one of those runtimes via an adapter.

Anthropic built the cloud runtime. OpenRig is the control plane above it.

Start building

Open source. Local-first. Cross-harness. No vendor lock-in. No per-session pricing.