Pricing
One meter. Developers and Workers alike.
You pay for the tokens your people and your Workers use, on one rate card with four lanes, at rates 60% or more below comparable frontier API pricing. No seats. No platform fee. No per-Team or per-Worker charge. Every run prints what it cost while it runs.
The meter runs when the work runs
An engineer in YAGNI Code, Proctor reviewing a pull request, Fletcher walking a test in a browser: each step is metered on the lane it ran on and printed while it runs, and the total is on the case file when the work item is done.
Workers and developers pay the same
There is no separate price for a Worker. Teams are not a priced unit. The same per-token rates apply whether a person or a Worker spent the tokens, so adding a Worker to a Team is a usage decision, not a procurement one.
No per-seat charges, ever
Your whole team is inside the workspace at no charge: reading the record, answering gates, managing Teams, reviewing pull requests. We never bill you for a reader.
The rate card
Four lanes. A router picks.
Every step is sized before it runs and placed on the cheapest lane that holds quality. Routine work never pays top rates. The per-token rate for each lane is shared on the call.
Efficient
The fast lane
Small, well-bounded steps: mapping a repository, reading a ticket, summarizing a thread. Fast open weight models that cost the least per token.
repo maps · tool calls · routine reads
Standard
The everyday lane
Most of the work: proposals, standups, ordinary implementation, follow-up review rounds. Strong open weight models that hold quality on routine engineering.
proposals · the Brief · everyday code
bailey
harperAdvanced
The build-and-review lane
Plans, builds, and pull request review that read the whole repository. The strongest open weight models, evaled on our harness before they serve a request.
plans · builds · PR review · browser tests
wright
proctor
fletcherPeak
The judgment lane
A few tokens of the strongest reasoning model on the card, still open weight and US-hosted, spent where a wrong call is expensive: the critique on a consequential plan, the hardest refactor.
plan critiques · hard refactors
reeveCaps keep it predictable: a spend cap per Worker and per day, and a cap on items in flight and review rounds per Team. Usage shows what ran where, by developer, by Worker, by Team, by day.
For finance, security, IT
The boring part, handled.
Monthly by default
Month-to-month, cancel any time. Annual contracts with invoicing are available; we set those up on the call.
SOC 2, DPA, BAA
Our SOC 2 Type II monitoring window is in progress. A DPA and a Business Associate Agreement are available on request, and the trust portal is public.
Encrypted and scoped
Connector credentials are customer-scoped, AES-256-GCM encrypted, and resolved only at execution time, never placed in model prompts. Repositories are reached only through the workspace GitHub App.
Sanitized action trail
Tool calls record the triggering user or Worker, tool, sanitized parameters, status, duration, and timestamp. Result payloads and message bodies are not copied into the audit row. A sample export is available on request.
Custom security review
We support custom security questionnaires, share current control evidence, and will join your security review call.
The standard ones
MSAs, NDAs, MNDAs. We sign yours or you sign ours. Procurement mailbox at procurement@yagni.app. AP at ap@yagni.app.
We've been through procurement at companies of thirty, fifty, two hundred, five hundred. We can move at your pace.
Questions on pricing
The small print, large.
Why isn't the rate card on the page?
Because the rates move with the models behind them, and because the useful number is your bill, not a per-token figure. Thirty minutes with the founder ends with the per-token rate for each lane and an estimate built from your own usage export: which lanes your work lands on, what a month costs, and what the same month costs at frontier API rates.
Do Workers cost more than developers?
No. One rate card. Proctor reviewing a pull request and an engineer working in YAGNI Code pay the same per-token rates on the same lanes. There is no per-Worker fee and no per-Team fee. Workers tend to cost less per task than a person doing the same work by hand, because the router puts each step on the cheapest lane that holds quality.
Who picks the lane?
A router sizes each step before it runs and places it: a repository map on Efficient, an ordinary implementation on Standard, a whole-repo review on Advanced, a consequential critique on Peak. Nobody hand-picks models. Each Worker has a default lane, and the Usage report shows what actually ran where.
Do you charge per seat?
No. Humans are free, always. Your whole team works inside the workspace: reading the record, reviewing pull requests, answering gates, managing Teams. Per-seat math punishes exactly the behavior we want more of.
How do we keep the bill predictable?
Caps. Set a spend cap per Worker and per day, and a cap on items in flight and review rounds per Team. Usage shows spend by developer, by Worker, by Team, and by day, with the frontier benchmark alongside, so the number an admin forwards upward is one report, not a spreadsheet.
Can we start small?
Yes. A pilot is one repository, one backlog, and the first Workers at Training. Run it beside whatever you use today and read the comparison from your own Usage report. Month-to-month by default, cancel any time; annual contracts with invoicing, MSAs, NDAs, and security review are all available and set up on the call.
The rates come from the call.
Thirty minutes with the founder ends with the per-token rate for each lane and an estimate built from your own usage export. Bring one repository and a backlog, and the first Workers start at Training that day.
or read how agent teams work