Postgres over Dynamo for the job store
yagni · /go run · liveSESSION $0.03
- you/go move PR review from a modal to a full screen
- maprepo mapped · fast model · $0.03
- planplan staged for approval · strongest model · $0.11
- youplan approved · one edit
- buildbuilt behind flag · standard model · $0.19
- reviewadversarial review · 2 findings fixed · $0.09
- teststests green · 14 added
- decidedecision recorded · full screen over modal, and why · team ledger
- prPR #482 opened · session total $0.42
- ratesame run at frontier API rates · $1.12
- nextnext task queued · starts already knowing this repo
The drop-in replacement for frontier coding agents
Coding agents that know your company.
YAGNI is a CLI and desktop app, drop-in with your skills, plugins, and MCPs, backed by an agent platform that learns how your team builds software. Every task routes across vetted, US-hosted open weight models, 60% or more under frontier API rates, with the cost printed as you go.
30 minutes with the founder · bring your usage export
01 · The bill is changing
Seats are becoming API rates. The math breaks.
Frontier vendors are moving growing teams off flat subscriptions and onto API usage pricing. At those rates a power user clears $9,000 a month, and a typical developer runs near $4,000. Across a team, the bill adds up fast, for the work your team already does today.
Adoption is not the problem. When most of your PRs are agent-touched, the spend scales with your best habits. The bill is the problem.
The open weight models are cheap and competitive. There is no question in my mind, this is where we are all going to land.
02 · How 60% is possible
Open weight models, routed intelligently.
The best open weight models are now genuinely good, at a fraction of frontier API prices. YAGNI serves them from US hosting with zero data retention, and every new release is evaled and vetted before it serves a single request. Model churn lands on us, not on your team.
Then an advisor model sizes up each task and routes it. Mapping the codebase runs on a fast, inexpensive model. A hard refactor gets the strongest one. Routine work never pays top rates, and it happens automatically instead of developers hand-picking models.
60% or greater savings vs comparable frontier API rates. Across the board.
The router · every task, sized and placed
map the codebasefast model$
implement the planstandard model$$
hard refactor · final reviewstrongest model$$$
evaled before it ships · rollback flags on every release · cost printed per step
03 · See it run
One command. Plan, build, review, PR.
This is the /go pipeline from the CLI, as your developers see it: map, plan, implement, adversarial review, pull request. The run pauses at the checkpoints that are yours, and the meter prints every step.
~/repo · yagniLIVE COST $0.42
- you/go move the PR review from a modal to a full screen
- maprepo mapped · 212 files in scope · fast model · +$0.03
- planimplementation plan staged · strongest model · +$0.11
- youplan approved
- buildimplemented behind flag pr_review_page · standard model · +$0.19
- reviewadversarial review · 2 findings · both fixed · +$0.09
- teststests green · 14 added
- prPR #482 opened for your review · session total $0.42
- ratecomparable frontier run · $1.12
- The run waits here. Checkpoints are yours.
every /go run: map · plan · build · up to three review rounds · PR
YAGNI · desktop
- ✓repo mapped
- ✓plan approved by you
- ✓built behind flag
- ✓adversarial review passed
- ✓PR #482 opened
The desktop app is the same engine with a window around it. Full parity with the CLI: one workflow, one meter, one record.
No migration project
Your skills, plugins, and MCPs slot straight in. If your team runs a frontier CLI today, they already know how to run this: same commands, same muscle memory, different bill.
skillspluginsMCPsworktreesday one
04 · It compounds
Every run makes the next one better.
Team knowledge. Every run writes to a decision ledger: what was chosen, what was rejected, and why. Your agents read it before they plan, so they stop re-deciding what your team already settled. Your engineers read it instead of asking around. Your next hire reads it on day one.
Domain knowledge. Agents answer from your business, not just your repo: the customer who filed the ticket, the deal that depends on it, the incident behind the alert. Grounded in your connected tools, cited so you can check the work, and deeper every week you use it.
A harness built to work as a team. When one developer corrects an agent, every developer's agents learn it. Shared playbooks and review rules, per-developer usage and budgets, one record across every repo. Not thirty copies of a personal tool that happen to share a bill.
That is the difference between a discount and an asset: thirty days in, YAGNI knows how your team builds software, and everything it learned is yours.
The decision ledger · written as you work
D-0142
D-0157
webhooks retry through the queue, never cron
D-0163
PR review is a full screen, not a modal
yours in the app and at the CLI · searchable · the record stays when a session ends
05 · Quality holds
Cheaper never means sloppier.
Two mechanisms guard every run. An adversarial reviewerattacks the change before the pull request opens. And Ask YAGNI checks the work against what your business tools actually say, so the code fits the decision, not just the diff.
Adversarial review · every run
- ✓round 1 · race on the retry path · found and fixed
- ✓round 2 · missing index on the backfill · found and fixed
- ✓round 3 · clean · the PR opens
up to three fix rounds before a PR opens, on every /go run
Ask YAGNI · the business fit check
As code completes, the agent checks the change against the record: the Jira ticket, the Confluence page, the Slack thread where scope changed on Tuesday. An extra reviewer with the context the developer might not have had open.
JiraConfluenceSlackGitHub
optional · with nothing connected, the CLI still works on day one
06 · Run it like an admin
Built to pass your security review.
Usage per developer
Spend by person, by model, by day. Cost controls live in the admin panel, not in a spreadsheet you build later.
US-hosted · zero data retention
Two independent inference providers, both hosted in the US, both under zero data retention agreements.
Your identity provider
Google single sign-on today. Enterprise IdP and finer-grained permissioning are scoped with your team during rollout.
SOC 2 Type 2 underway
The monitoring window is in progress. Full security posture report and trust center access before your pilot starts.
07 · The kicker
Agent Teams that expand your bandwidth.
Once YAGNI is in your developers' daily work, hand a whole lane to an agent Team. A proposer finds the next item, a builder takes it through plan and implementation, a specialist reviews, and the lane ends at a pull request a person signs off on. Your engineers own a wider zone while the routine lane keeps moving.
Every Team starts in training, where nothing ships without a signature. Autonomy is earned from the record, never assumed, and you set the leash per Team: how many items in flight, which rules in the playbook, when it may act alone.
Training
Supervised
Autonomous
The vulnerability queue · steady state
VULN-214dependency bump · transitive CVEPR open
VULN-209input sanitization on legacy formplan approved
VULN-217response header hardeningproposed
two or three items a day, taken to draft PRs with specialist review · a person verifies and merges
08 · Pricing
One meter. At least 60% under frontier rates.
No seats to count and no platform fee to justify. You pay for the tokens your team uses, at rates that come in 60% or more below comparable frontier API pricing, and every session prints what it cost while it runs.
Metered, not per seat
Your whole team is in the workspace at no charge. The meter runs only when the work runs, and it flexes down as easily as up.
The floor is 60%
Vetted open weight models plus intelligent routing keep every task off top rates unless it needs them. The savings hold across the board, not just on the easy runs.
Prove it on your usage
Pilot free with two or three developers. Run it beside your current agent, then read the comparison from your own sessions, not our deck.
Before your renewal, not after.
Keep the workflow. Cut the bill.
Bring your usage export to the call. You leave with a savings estimate computed from your team's actual numbers, and a pilot plan that starts with two or three developers this week.
30 minutes with Jack, the founder · no migration project on the other side