Run covers two things: keeping what we built running, and shipping a set number of improvements every month, each one named so you can see exactly what changed. No hourly rates, no vague retainer.
A retainer that only keeps the lights on stalls. A retainer that only ships features lets the foundation rot. Run holds both at once: an always-on baseline underneath, and a named set of improvements on top.
The baseline is not optional and never comes out of your improvement count. The improvements are scoped, shipped, and invoiced as named outcomes, never as an hourly rate.
Priced into the fee, running whether or not you ship anything new that month.
A fixed number of scoped improvements shipped each month, each one named so you see what changed.
The entry point. We keep what we built running: the full reliability baseline (monitoring, evaluation, incident response, model-upgrade regressions, prompt versioning) and nothing more. No build blocks; new work is quoted as you need it. The tiers below add monthly build throughput on top, and you can move up as the system earns it.
One small or medium improvement shipped each month. A steady, deliberate pace for a system that mostly works and needs to keep improving.
Quarterly review
Two to three improvements a month, or one larger enhancement. Enough room to keep momentum without stacking work into a backlog.
Quarterly review + monthly check-in
Several changes a month, or a rolling larger enhancement. For systems that are load-bearing and expected to keep expanding.
Monthly review
The reliability baseline is identical in kind across all three tiers (evaluation, monitoring, incident response, model-upgrade regressions, prompt versioning) and is never drawn from your block count. Higher tiers buy more improvement throughput, not a better safety layer. Six-month minimum term.
Not a ticket, not an hour, not a meeting. A block is a real change that lands in production and shows up by name in your review.
A new automation step, a new integration, a new report, a new set of evaluation scenarios, or a tuning pass on something already live.
Small, medium, or larger. We agree the size before we start so the month's plan is clear and nothing is a surprise.
You see the outcome, not the mechanics. The price per unit stays internal to us; you approve and track work as named results.
The price per unit stays internal. Clients see named outcomes, agreed and sized up front, so a month of Run reads as a list of shipped changes rather than a bill of hours.
Once per quarter, borrow forward against the following month to land a bigger change now. The next month starts lighter by exactly what you borrowed. Simple, and only once per quarter.
Break the enhancement into stages that each fit a month's block count. You keep the reliability baseline the whole time and the change ships in deliberate, reviewable steps.
If it is genuinely a project, convert it to a fixed-price mini-Build, value-priced at the Run preferred rate. It runs alongside Run without draining your monthly blocks.
Unused blocks roll forward a single month and expire at the quarter boundary. The cap keeps the pace honest: Run is momentum, not a savings account you draw down all at once.
Evaluation, monitoring, incident response, model-upgrade regressions, and prompt versioning are part of the monthly fee and always on. None of them come out of your block count, so a busy build month never quietly turns off the checks that catch failures.
A block is always a change you asked for and can name. Keeping the existing system healthy is our standing obligation, not a line you spend against. If reliability work grows because the system grew, that is on us to carry, not on your blocks to absorb.
Because the two budgets are separate, there is no lever that lets anyone deprioritize evals or monitoring to ship one more feature. The failure mode where a system quietly degrades under feature pressure is designed out of the pricing.
Kill switches, observability, evaluation harnesses, model-upgrade regressions. We built the habit inside our own infrastructure, and Run keeps it funded by default so the thing that works on Tuesday still works on Monday two months later.
Run Plus: $9,000 a month, 5 blocks a month. Here is how a real quarter spends, rolls, and converts.
| Month | What shipped | Blocks used |
|---|---|---|
| Month 1 | Conflict-check step added to intake (3), plus 2 new evaluation scenarios (1). 1 block rolled forward. | 4 of 5 · 1 rolled |
| Month 2 | NDA and MSA redline extension (4), plus a tuning pass on a step that was over-flagging (2). Rolled block absorbed the overage. | 6 of 5+1 |
| Month 3 | A partner wants a client portal, so it converts to a fixed-price mini-Build ($18,000, 3 weeks) at the Run preferred rate. Blocks that month went to a new intake source (3) and a monthly volume report (2). | 5 of 5 + mini-Build |
A working session with our team · 30–60 min · a plan you keep either way