FAQ #

Models and usage #

Which model should my roles use? #

Split by how hard the work is, not by one favorite model. On a ChatGPT plan, for example:

nomarmy army assign sr-dev codex gpt-6-astra    # the core and the harder implementation
nomarmy army assign jr-dev codex gpt-5.6-sol    # simple, fully specified pieces
nomarmy army assign pm codex auto               # the General picks per job

A frontier model like gpt-6-astra is worth it for subtle work and review; a lighter one like gpt-5.6-sol handles well-specified coding for a fraction of the usage. With a role on auto, the General picks per job, and /feature tells it to use the lighter model for routine work and the frontier one for subtle work. nomarmy stats --model gpt-6-astra shows what each model has actually been doing.

Can I switch a running job to a cheaper model? #

Not in place: a job's model is fixed when it's dispatched. But you can stop it and finish the work on another model. Ask the General to stop it (the local_worker_stop tool), or run nomarmy jobs --stop <job id> yourself: the worker ends within about 15 seconds, without a report-recovery call or a verification run, and its worktree is kept. Then a new job with continue_from: <job id> and model: gpt-5.6-sol picks the work up where it stopped.

Messaging the General mid-job is safe either way: nothing about the running job changes unless it's stopped, and the General applies what you say to the jobs it sends next. To move work to a cheaper model from now on:

For every job on a role: nomarmy army assign sr-dev codex gpt-5.6-sol. It applies to the next job, with no restart.
For one job: ask the General to send it with model: gpt-5.6-sol. A job's model overrides the role's.
To finish a job on a cheaper model: if it comes back partial, failing verification or stopped, the General can continue it with continue_from: <job id> and model: gpt-5.6-sol. The new job starts from the old one's unfinished work, so nothing is redone.

Does nomArmy switch models when I'm near my usage limit? #

No, the General does. nomArmy reads how much of a subscription's limit is used (Codex from every job's own session log, Claude from what Claude Code reports), shows it in army, local_worker_capacity and the status line, warns from 80%, and holds new jobs on that agent at the limit until you say go. Which model to use with the headroom left is the General's call, or yours with army assign. A reading is only as fresh as the last job on that agent: after you reset your usage, the next job's reading shows it.

Why is my usage going so fast? #

A worker re-reads its whole conversation on every step, so most of a job's tokens are cached context, and a frontier model spends its allowance fastest. Put the routine roles on a lighter model (above), keep briefs focused on one outcome, and prefer mode: verify (no model at all) for checks that only run existing tests. nomarmy stats shows tokens by model.