Have AI agents produce software without losing control.
Norn turns your intent into a living plan, then that plan into local, bounded, traceable delivery.
You can continue with your usual agent or let Norn orchestrate execution. Either way, the repository, the decisions and the evidence stay under your control.
Your agents code fast.
The real problem starts after.
When several agents step in, it becomes hard to know:
- ?who decided what
- ?what was actually validated
- ?why the code changed
- ?what can be applied without risk
Norn takes the project out of chat. The plan becomes the source of resumption and governance.
Frame.
Plan.
Deliver.
You start from the expected outcome. Norn builds a living plan, stabilises it, then turns it into delivery — by you or by a team of bounded agents.
Externals outside the SSO scopesettled
Norn prepares the context and prompts. You execute with your agent. The lead agent orchestrates execution agents: one lot per agent, each bounded to its scope and worktree.
Resume the project at the right place, even with another agent.
Claude
- L1write scope
- L2access migration
- L3account review
- Problemcontractor access not covered
- Decisionsscope frozen · phased rollout
- Evidenceauth/scope.rs:142 · 42 tickets
- Constraintsno API break
- Openorphan account status
- Resume atL2 · access migration
Codex
- L1write scope
- L2access migrationresumes here
- L3account review
Nothing important is applied because a model decided so.
“The agent is done.”
+184 −37 · 6 files- the plan used
- each task's scope
- the execution profile
- dependencies
- the evidence produced
- the repository state
- the risk before applying
Apply to the repository?
Norn does not ask agents to behave.
It limits what they can do.
An agent does not freely obtain shell, Git, network or commit access because its reasoning calls for it.
Start light. Automate when the framework is ready.
Norn prepares everything. Nothing leaves without your action.
Norn sends within the budget and rules you have set.
Autonomy is never a condition for using the product.
Norn confronts intent with what actually exists.
When a repository is already implemented, Norn starts by observing it before confronting that reality with the request.
- →Important technical claims stay linked to file:line evidence and to the observed repository state.
- →If the repository changes between stabilisation and planning, Norn refuses to plan as if nothing had happened.
SSO covers contractor access.
You can explain how the delivery was produced.
Norn keeps fingerprint-signed documents and append-only governance. Decisions, fixes, risks and debts are not rewritten afterwards.
You keep an auditable trajectory from the need to the application.
Control stays local.
Norn is designed as a local, auditable control plane.
norndurable state
Web, terminal or automation.
Cockpit to frame, follow, authorise and audit without living in commands.
Terminal cockpit for expert users.
Complete surface for automation and advanced workflows.
All three use the same capabilities and the same use cases.
When the problem is no longer generating code, but mastering its delivery.
Norn is for people already using Claude Code, Codex or other agents to work on real repositories.
- →delegate more without losing the evidence
- →frame before launching
- →parallelise without letting agents step on each other
- →resume with another agent
- →keep human authority over what actually enters the repository
Norn does not try to replace your favourite model.
It sits around their work.
It is the layer that turns agentic work into governed delivery.
Give your agents more work.
Not more freedom than needed.
Install Norn, open your repository and start from the outcome you want.