AI Ops

You cannot operate what you cannot see

A single agent in a terminal is easy to watch. A fleet of agents across repositories and machines is not. Operating agents at scale needs the same things operating services needs: a live view of who is doing what, budgets and the ability to enforce them, metrics that distinguish productive work from loops and waste, and a way to reconstruct what happened when something went wrong. Chat transcripts do not give you this. A ledger that every agent writes to, with identities, leases, costs and evidence attached, does.

Live: leases, inbox and status

nool status shows the state of the ledger, active leases and unstaged work. nool announce status lists every lease: who holds what, in which mode, with its fencing number, claim state and origin, and can follow grants and releases as they happen. nool inbox shows notifications and unassigned open tasks with their acceptance-criteria attention state. nool task board lays tasks out by lifecycle column, and nool task telemetry attributes token usage and cost to a task. When two agents want the same scope, nool announce wait blocks one until the other releases rather than making an operator arbitrate.

Cost, waste and loops

nool usage reports token consumption per provider, model and agent. budget-set and budget-status enforce and show per-agent token budgets; personas carry a daily USD budget on top. nool usage analytics reports cost per line, waste ratio and efficiency; nool usage bounce reports context an agent compressed and then re-read in full; nool usage agent reports bug rate, severity and patterns per agent; nool usage dashboard combines tokens per bug across threads. nool insights loops detects repeated failures in agent sessions, and nool insights conflicts shows the semantic merge conflicts Nool resolved. These are the numbers that tell an operator which agent to cap, retrain or retire.

Reconstruction and reporting

When something breaks, nool debug replay steps through a Git ref or an agent run, with a terminal UI if you want one; nool debug blame walks from the failure to its cause; nool debug bisect narrows a regression with a test command. nool insights produces project reports with blast-radius statistics and time-saved estimates, formatted for executives, managers or engineers. Nool's own internal case study used exactly these reports: across three repositories in July 2026, 1,307 knots, 299 findings and 407 resolved semantic conflicts, with per-repository drift velocity in knots per day, published as dogfooding metrics rather than customer proof.

Health and release readiness

Operating a ledger has its own hygiene. nool doctor runs release-readiness and repository health checks; --strict makes it fail on anything it would otherwise warn about, and --fix repairs what it safely can. nool status warns when parallel timelines have diverged and says what to do: merge before release, or nool pluck a thread out of the active timeline while keeping its history. nool checkpoint marks releases. nool prune clears temporary and cached files, and nool migrate moves ledger files from older layouts into the current one after an upgrade. These are the commands an operator runs on a schedule, and they are the difference between a ledger that stays trustworthy and one that slowly stops being read.

Frequently asked questions

Do I need to add tracing to my agents?

No. Agents that work through Nool write to the ledger as a side effect of announcing, proposing and solidifying. Costs, leases, tasks and evidence are recorded there and read back by the commands above.

Can I cap what an agent spends?

Yes. nool usage budget-set caps tokens per agent, personas carry a daily USD budget, and persona runs are bounded by a maximum number of turns. nool fleet start can also run a self-check before any reviewer is charged.

What does Mission Control add?

A desktop control plane for the same data across repositories: projects, missions, governance state and agent activity, for people who operate fleets rather than run commands. It is part of an organisation licence.

How do I know which agent is causing regressions?

nool usage agent reports bug rate, severity and patterns per agent, nool debug bisect finds the knot that introduced a regression, and every knot names the agent id it was proposed under.

What does nool doctor check?

Release readiness and repository health: the state of the DAG, diverged timelines, integrity of the ledger and its Git mirror, and the checks a release should pass. --strict fails on warnings, --fix applies safe repairs, and nool fleet start can run it as a self-check before any reviewer is charged.