Opportunities › Cost optimization for LLM-powered development tools › Proposed app
Spillway
Send routine coding-agent work to cheap models; keep premium ones for hard problems
- Score
- 61
- confidence 66
- Evidence
- 9
- items on the page
- Rivals
- 2
- named in sources
- First version
- 30 d
- estimate
- Type
- Desktop app
- PWA: runs in the browser, installs on Mac and Windows
- Market since
- 2025
- New market · AI estimate
ScreensConcept
Desktop app: Developers configure routing rules, review dense cost dashboards, and audit request logs—work done at a computer over long sessions. Sample screens on a computer, in the proposed brand. Their numbers are sample data.
Today
Routing on68%
About this appAI proposal
Spillway is a model router for people who use coding agents such as Claude Code or Codex. It sits between your agent and the models. Each request is scored for difficulty. Routine work, like renames, boilerplate, test stubs and small edits, goes to a local or open-source model. Hard problems still go to your premium model.
You set the rules in plain terms: which folders, task types or prompt sizes stay on the premium model, and which may go cheaper. A dashboard shows where each request went, what it cost, and how many were sent back up because the cheap answer failed.
It is built for developers and small teams who already pay for a premium subscription and suspect much of that use is routine. Unlike sending everything to one model, it gives you a routing decision you can see, change and undo.
- Tagline
- Pay premium prices only for premium problems.
- Voice
- PlainPracticalCandid
- Why this name
- A spillway sends overflow along a safe channel so the main reservoir is not overloaded. Routine requests overflow to cheaper models; the premium model keeps its capacity for hard work.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
Difficulty-based routing
Every agent request is scored and sent to a local, open-source or premium model according to your rules.
No member yet
Escalate on failure
If a cheap model's answer fails your tests or a confidence check, Spillway retries on the premium model.
No member yet
Rules you can read
Write routing rules by path, task type, language or prompt size. Each decision shows which rule fired.
No member yet
Local model support
Point Spillway at models running on your own machine or at any compatible endpoint you already use.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#1F5C4ASpruce · primary#F6F3ECPaper · surface#1F2A24Ink · text#E0862BAmber · accent#3C6E9ESlate Blue · data
Typography
Aa
Spillway
ABCDEFGHIJKLMNOPQRSTUVWXYZ 0123456789
Pay premium prices only for premium problems.
PricingAI proposal
The material gives no competitor prices. These tiers are a starting point to test, not a market finding. Any hosted inference credits would be a separate, later test.
Free
Free
- One machine
- Up to 3 routing rules
- Dry-run on recent logs
- Local model endpoints
Pro
$12 a month
- Unlimited rules
- Escalate-on-failure checks
- Savings history and export
- Several model endpoints
Team
$10 a month
- Price per seat
- Shared policies and overrides
- Monthly savings report
- Shared team dashboard
What it could be worthEstimate
What a software business making this much sells for today, from real deals. Pick a monthly revenue.
If it makes
Worth about
$190,000
Sale range $132,000 to $263,000
$120,000 a year in revenue
From 87 software businesses listed and sold on Empire Flippers in the last year, revenue vetted by the broker: they sell for about 19× their monthly revenue, most between 13× and 26×.
Measured on profit, the middle deal is 32× monthly net profit: about 2.7 years of profit.
An estimate, not an offer or financial advice. A real price depends on profit, growth, churn, how much of the work is the owner's, and the buyer.
How it would growAI proposal
How the product would reach new customers through its own use, and how the first ones would be found. A plan to test, not a result.
- Built-in loop
- Shared outputs. Developers share a savings report link or a rule-set page with a team lead, or post a rule set to teammates. Viewers see the numbers and the Spillway name on the page.
- Do they become customers?
- Viewers are often developers or leads on the same subscriptions, so they could become users or team buyers. Test whether report views lead to installs and whether leads approve Team seats.
- First customers
- Solo developers and teams of 2 to 10 engineers who pay for Claude Code or Codex subscriptions and run agents daily on routine work such as tests, refactors and docs.
- Where to reach them
- GitHub and open-source routers: Answer issues and discussions in related repositories with a dry-run report. Offer the free tier to people asking about cost.
- Developer communities: Post worked examples of real savings from dry-runs on public repos in forums and agent-focused communities.
- Direct outreach to small teams: Offer a free one-week dry-run to engineering leads of small teams already paying for agent seats.
- First milestone
- Fifty dry-runs completed and ten paying users or teams who keep routing on for 30 days with no rise in escalations.
Why people already pay for this
- VerifiedGitHub repository gained 363 stars in 46 days
- Public claimDevelopers with Claude Code and Codex subscriptions spend on premium model APIs for both routine and complex coding agent work
- Public claimOpen-source and local models like DeepSeek are viable for routine coding tasks
- Public claimPremium models are underutilized on low-complexity tasks in agent workflows
- Estimate
Tech stack
Every proposal builds on the same PWA base, installable from the browser on any phone or computer. Dashed boxes are what this app adds.
- AppNext.js and TypeScript
- PWAManifest and service worker
- UITailwind CSS
- Data and authSupabase Postgres
- PaymentsStripe
- EmailResend
- HostingVercel
- This app · Local routing proxySmall cross-platform CLI that runs beside the agentA PWA cannot intercept a coding agent's requests; a local proxy applies rules and keeps code on the machine.
- This app · Difficulty classifierRule-based scorer, then a small local modelScores prompts cheaply before any model is chosen; rules first keep it predictable.
RisksAI inference
- TechnicalMisrouting a hard task to a weak model gives bad code. Escalation checks need tests or signals that many repos lack, so routing may quietly lower quality.
- PlatformCoding agents and their subscriptions may limit custom endpoints or change terms, which could block routing through a proxy.
- CommercialEvidence shows interest (a repository with 363 stars in 46 days) but not willingness to pay. Open-source routers may make a paid tier hard to justify.
- DataDevelopers may refuse to route proprietary code through any tool. The proxy must run locally and log metadata only.
- AIOpen-source model quality changes fast, so routing rules and difficulty scores need ongoing re-testing against new releases.
Everything marked AI proposal or AI inference was written by a model from the evidence on the opportunity page. The proposed name must be checked against trademark registers before it is used.