Branch ideas. Review reasoning. Materialize software.
NodalMerge Studio is a persistent collaborative runtime where humans and AI agents share goals, reasoning, execution, and artifacts as one content-addressed graph. Any point in that graph can be reviewed, branched, or materialized into a working tree — replicated live across human and headless peers.
Every step an agent takes produces a durable artifact.
AI agents are fast, but their goals, reasoning, and execution usually disappear into isolated sessions, logs, and pull requests. Here, every artifact can be inspected, branched, replayed, merged, and audited — the durable graph is the product; agents are features of it.
Goal
A human describes what needs to be accomplished.
Plan
An orchestrator agent breaks the goal into work units and tasks — inspectable, reviewable, branchable. Big goals decompose recursively: a subsystem becomes its own sub-plan that reconciles back up.
Work units
Worker agents execute tasks, each on its own branch. Every unit has lineage and scope.
Branch changes
Auditable, attributable, comparable changes accumulate on the branch.
Merge proposal
Workers submit structured proposals with full lineage — reviewable, approvable, supersedable, replayable.
Approved state
By default a human approves; approved changes land in the authoritative branch, committed and traceable back to the goal.
Structured autonomy, not blind trust.
Durable artifact graph
Every plan, task, proposal, and decision is a persistent DAG node. Trace any approved change back through its entire lineage to the original goal.
Human-governed by default
Agents propose; humans decide. The merge review workflow enforces human approval before speculative work lands in the authoritative branch — unless a goal explicitly opts out.
Branch and replay
Branch from any artifact checkpoint to spawn alternate executions. Replay a proposal under a different model to compare outcomes. Roll back to any known good state.
MCP-native
A frozen MCP v1 tool surface with 117 tools across 21 namespaces. Integrate any MCP-compatible client with the Studio host.
Reconciliation, not collisions
Run agents in parallel on a shared candidate branch without babysitting merge conflicts. When two proposals collide, reconcile with a dedicated agent, restart the losing goal from a clean snapshot, or resolve it yourself — the choice, and the audit trail, are always yours.
Studio is v1 - deliberately scoped.
Every constraint in NodalMerge Studio is a purposeful decision. Planning runs at bounded depth with capped turns and retries, so goals decompose predictably instead of sprawling. Reconciliation is deliberate and reviewable: parallel agents converge on a shared branch under rules you set, never as silent auto-merges. Model profiles are configurable per role, so the right model does the right work. Human approval stays the default gate before anything lands. We scoped Studio to the decisions we can stand behind and audit. What ships next is chosen with the same care.
Run your first goal
Install the VS Code extension and see the artifact graph build itself.