GENOS AGENT RUNTIME / OPEN SOURCE

An agent runtime built for
what happens next.

GenOS versions workspace state, runs supervised agents within boundaries, and keeps evidence attached to decisions. Fork approaches, inspect the results, and promote a change only when its checks support it.

Explore the runtime Try the zero-token demo

Alpha software · capabilities and evidence vary by execution path

A GENOS WORKFLOW / ILLUSTRATIVEGENOS / 01
01
Versioned starting statesnapshot · workspace · context
02
Counterfactual branchesisolated candidates · budgets
approach Aapproach B
03
Evidence gatediff · checks · decision
Promote, hold, or revise

A successful run alone does not establish a valid outcome.

8 orchestration topologiesRust runtime + Node.js control planeCLI + leased MCP tools

REVIEWED 18 OCTOBER 2026

The runtime moved forward.

Topology runtimes, native workers and durable workspaces now carry stronger bounded evidence. The release review keeps every advance beside the limit that still applies.

Review all six evolutions
01

Topology runtimes moved beyond composition

Rhizome now has a provider-backed growth and routing path marked implemented. Holobiont and Metapopulation move to partial with verified host-veto and regional-runtime slices.

02

All 19 native worker types have bounded qualification

The executable inventory now covers 19 WorkerKind values, 25 methods and 24 benchmark cases, with C2/C3 evidence for one bounded task per type.

03

Morphogenesis gained an adaptive, evidence-gated control path

The runtime can characterize a mission, select admitted topology variants, execute bounded graph changes and retain verified outcomes for later decisions. All 19 organization policies now have executable step coverage.

01 / THE RUNTIME

Work can branch.
The reasoning stays visible.

GenOS provides concrete operations for state, execution, and review. Coverage depends on the adapter and workflow; the evidence ledger records those limits.

01 / 04

Version the state

Snapshot supported workspace state and compare later changes with a concrete starting point.

Runtime model
02 / 04

Explore alternatives

Fork isolated candidates, inspect diffs, and replay supported trajectories when the inputs permit it.

Explore the sandbox
03 / 04

Supervise execution

Launch agents through configured runtimes with workspace boundaries, budgets, events, and handoffs.

Orchestration
04 / 04

Gate the decision

Keep receipts and checks with the proposal. A gate can hold or reject a change when evidence is missing.

Evidence ledger

02 / ORCHESTRATION

Eight ways to
organize the agents.

Topologies define different coordination patterns. Each has its own contract, implementation status, and validation gaps.

01 / 03

Compare & specialize

Trinity · A-Team

Parallel candidate worlds and domain specialists.

02 / 03

Coordinate & share

Biocénose · Syncytium · Rhizome

Quorums, shared state, and capability-based routing.

03 / 03

Adapt & recover

Holobionte · Biome · Métapopulation

Host controls, resource cycles, and lineage-aware recovery.

A registered topology does not imply every declared capability is active or validated end to end.

Compare all eight topologies

03 / START WITH EVIDENCE

Inspect the system.
See its limits.

The site separates implemented paths from partial, experimental, and proposed work. Recorded runs retain incomplete outcomes; the source and capability contracts explain what each path can actually do.

Read the source and product contract ↗