Example page made by Ghostlaunch. This product doesn't exist yet.
Back to the exampleFor Claude Code, Codex and Cursor users
Shuntly is a CLI proxy in front of your coding agent. It sends each task to the cheapest model that can handle it and saves your frontier quota for the hard work.
$15/mo after trial. Uses the plans you already pay for.
Your coding agent spends frontier tokens on everything: renames, test runs, grep-and-summarize, boilerplate. Then the cap hits halfway through real work.
You hit the 5-hour limit with half the files changed. Now you wait for the reset and rebuild the context from scratch.
You're doing what you did last month, but your quota disappears by lunch and you can't tell what's eating it.
Your weekly quota goes to renaming variables and reading test output, work a cheaper or local model could do.
You run two agents and a local Qwen, but picking a model by hand every time is so tedious you default back to the expensive one.
Shuntly plugs in through your agent's API base URL. You keep your editor, your agent and your habits.
Install the CLI and set the API base URL in Claude Code, Codex or Cursor. Add your existing subscriptions and any Ollama models.
Shuntly classifies each task (refactor, test fix, search, architecture) and sends it to the cheapest model that can handle it.
If a cheaper model's output fails your tests or checks, Shuntly reruns the task on the frontier model automatically.
Shuntly classifies every request by task type and routes searches, renames and boilerplate away from your most expensive model.
Cheaper output that fails tests or checks is escalated to the frontier model, so the shortcut never ships broken code.
Routes across Claude Pro/Max, ChatGPT/Codex and local models via Ollama. You don't need a new provider or a new API bill.
That Qwen you set up in Ollama takes the routine work automatically, so you never have to switch by hand.
A per-task log shows which model handled what, how many tokens it used and when an escalation happened.
It works as a CLI proxy via the API base URL. Your agent, your prompts and your editor stay the same.
You regularly hit the 5-hour or weekly limit before your workday ends.
You use Claude Code and Codex side by side and want a clear rule for who gets which task.
You run Ollama but rarely use it because switching mid-flow is a hassle.
Probably not for: Occasional users who never come close to their plan limits; Teams that require every request to go to a single approved model.
The usual fixes push the limit back or turn it into busywork. Shuntly changes what spends your quota.
| Criteria | Shuntly | $200/mo plan upgrade | Switching via /model |
|---|---|---|---|
| Routine tasks skip the frontier model | Yes | No | Partly |
| No judgment calls mid-flow | Yes | Yes | No |
| Uses your local models | Yes | No | Partly |
| Falls back if a cheap model fails | Auto, on failed checks | No | Only if you notice |
| Monthly cost | $15 | $100–$200 | Free |
Every feature is included. Try it free for 14 days on your real workload.
Full routing on the plans you already pay for. Cancel anytime.
$15/ month
Everything in Monthly, with two months free.
$144/ year
Model usage runs on your existing subscriptions and local models. Shuntly doesn't resell tokens.
Shuntly only hands off tasks it classifies as routine, like searches, renames and boilerplate. If a cheaper model's output fails your tests or checks, it escalates the task to the frontier model automatically. You can also see every routing decision in the task log.
Claude Code, Codex and Cursor, connected through the API base URL. It routes across Claude Pro/Max, ChatGPT/Codex subscriptions and local models served by Ollama.
Install the CLI, connect your subscriptions, optionally add Ollama models, then point your agent's API base URL at Shuntly. You don't need to rewrite prompts or change your editor.
A bigger plan pushes the cap back, but you still pay frontier prices for trivial work, and heavy users still hit weekly limits. Shuntly makes the plan you already have last longer, for $15 a month.
Shuntly runs as a CLI proxy on your machine. Requests go to the providers you already use, and tasks routed to Ollama stay local. We'll share full data-handling details with early access users before the trial starts.
It's $15/month, or $144/year if you pay annually. If it doesn't stretch your quota, cancel before the trial ends and you won't be charged.
Keep Opus for the hard reasoning. Let Shuntly route everything else and catch failures before they cost you.