Opportunities › LLM routing optimization for coding agents › Proposed app
Switchyard
Send each coding-agent task to the right model and cut your LLM bill
- Score
- 58
- confidence 79
- Evidence
- 37
- items on the page
- Rivals
- 12
- named in sources
- First version
- 40 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: Developer tool for coding agents requires monitoring dashboards, reviewing routing decisions, and analyzing spend patterns during long work sessions. Sample screens on a computer, in the proposed brand. Their numbers are sample data.
This week
Within budget$412
About this appAI proposal
Switchyard is a hosted endpoint you point Claude Code, Codex or any compatible coding agent at. During a session it picks a model for each step: a cheaper or faster one for routine edits, a frontier one for hard reasoning. You see what each choice cost and why it was made.
It is built for engineering teams that now ration frontier model spend by hand, picking models per task and watching budgets. Switchyard replaces that habit with rules you set and routing that learns from your own sessions, including whether a model's cache is still warm.
You bring your own provider keys, so there is no markup on tokens. Frontier APIs, open-source models and self-hosted models can sit in one pool. You pay a flat fee for the routing.
- Tagline
- The right model for every step of the job.
- Voice
- PlainPreciseFrugal
- Why this name
- A switchyard is where rail cars are sorted onto the right track. The name says what the product does with agent requests, in one plain word.
In the first version
Members say which features they would use. It tells PWAS what people want; it changes no score.
Drop-in agent endpoint
Change one base URL in your agent's settings. Requests, streaming and tool calls pass through in the format the agent already expects.
No member yet
Per-step model choice
Each request is scored on task type and context size, then sent to the cheapest model that clears your quality bar.
No member yet
Cache-aware routing
Tracks which model holds a warm prompt cache and avoids a switch when the cache saving outweighs the cheaper rate.
No member yet
Bring your own keys
Use your own accounts with every provider. No token markup; Switchyard only charges for routing.
No member yet
Brand identityAI proposal
Check the name against trademark registers before using it.
Palette
#1E5B4FSignal Green · primary#F6F1E7Paper · surface#1F2A24Ink · text#E8892BAmber · accent#3C6E9ESlate Blue · data
Typography
Aa
Switchyard
ABCDEFGHIJKLMNOPQRSTUVWXYZ 0123456789
The right model for every step of the job.
PricingAI proposal
Cursor's pricing page lists paid plans from $20 to $40 a month, so Team sits in that range. OpenRouter lists prices on request. The material gives no other prices; these tiers are a starting point to test.
Free
Free
- One developer
- Your own provider keys
- Basic routing rules
- 7-day session history
Team
$30 a month
- Per developer seat
- Cache-aware routing
- Budgets per team and repo
- Savings report
- 90-day session history
Business
Contact us
- Self-hosted model pools
- SSO and audit log
- Custom routing models trained on your traffic
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
$194,000
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
- No loop in the product. Switchyard sits between the agent and the model provider, so developers' colleagues do not see it. Growth comes from content, open discussion of cost savings, and word of mouth among engineering leads.
- Do they become customers?
- Savings reports can be shared with finance and other team leads, but that is not a built-in loop. To test: whether leads who see a report forward it and start a trial of their own.
- First customers
- Engineering leads at companies of 10 to 100 developers who already run Claude Code or Codex and have a monthly model budget they ration by hand.
- Where to reach them
- Hacker News and dev forums: Publish cost and speed comparisons from real sessions, and join threads on agent spend and routing.
- Direct outreach to eng leads: Offer a free one-week savings audit of their agent traffic, then convert to Team.
- Open-source community: Offer a compatible config for the existing open-source router so users can try the hosted version.
- First milestone
- Ten teams routing live traffic, and five of them paying for Team after a one-month trial.
Why people already pay for this
- AI inferenceFrontier LLM APIs are expensive and optimization of cost is a primary concern for users of coding agents.
- VerifiedHacker News post received 122 points and 42 comments
- AI inferenceCoding agent systems incur high costs from routing all tasks to frontier models when simpler models could handle many tasks.
- Public claimOrganizations need faster task execution from their coding agents, not just correct results.
- Public claim
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 · Streaming proxyEdge or long-running proxy serviceAgent requests stream for minutes and must add very little delay; serverless functions may time out.
- This app · Routing classifierSmall hosted model or rules engineScores each request by task type and context to choose a model in milliseconds.
RisksAI inference
- TechnicalSwitching models mid-session can break tool calls, lose cache savings or lower answer quality. Each provider differs in streaming and error behavior, and the proxy must add almost no delay.
- CommercialAn open-source router already exists and is free to adopt. Teams may self-host it, so Switchyard must win on hosting, reports and budgets.
- PlatformAgent vendors and model providers can add their own routing or change API formats and terms, which would undercut or break the endpoint.
- AIThe router may send a hard task to a weak model without noticing. Quality checks and a fast fallback to a frontier model are needed, and savings claims must be measured per customer.
- DataTeams must trust a third party with prompts containing source code and with their provider keys. Retention limits and a no-logging option may be needed to sell to larger companies.
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.