Signal routing

Not every prompt needs a powerful cloud model.

Many prompts, like writing an email, summarizing a document, answering a question, or helping with code, can run locally. PromptRouter automatically chooses the right path, local or cloud, based on what your request needs.

Try PromptRouter → See how it works
Local trace Cloud trace Routed automatically
PromptRouter
＋
Ask anything… (Enter to send, Shift+Enter for newline)
↑
◔ Temporary Mode Agent ⌄ Approval Ask each time ⌄ AutoLocalOnline ⚙ qwen3.5-…
LOCALOn this machine
Foundry LocalMicrosoft
Ollama
LM Studio

Plus every model in the Foundry catalog — DeepSeek R1, Qwen, Phi, Mistral, gpt-oss.

CLOUDOnly when you say so
GitHub Copilot
OpenAI
ClaudeAnthropic
DeepSeek
KimiMoonshot
Groq
Mistral
OpenRouter

Or any OpenAI-compatible endpoint. Keys stay in your OS credential store, never in a config file.

Local model setup

Picking a local model shouldn't be a struggle.

PromptRouter automatically detects your hardware and recommends a local AI model that fits your machine.

Your computer already has AI potential.

1
Detect your hardwareMemory, processor, graphics and available storage.
2
Find compatible modelsIdentify local models your machine can realistically run.
3
Recommend the best fitBalance model capability, speed and your workload.
Computer check complete

We found a good local fit

Based on this computer and a preference for balanced speed and quality.

16 GBMEMORY
8 coresPROCESSOR
Apple GPUGRAPHICS
Qwen 3.5 · 9BRecommended

A strong fit for chat, writing, coding and private documents. Runs completely on this computer.

Local AI

Powerful AI doesn't have to mean cloud AI.

Your local model can do more than simple chat. Use Ask, Plan and Agent modes to keep getting work done, even when you're offline.

OFFLINE No internet connection
⌘
Local model Running on your computer
→
AskGet answers from your local AI.
PlanBreak a task into clear steps.
AgentLet it work with approved files and tools.
✓ You can still get work done without the cloud.
01

Ask

Ask questions and get answers directly from your local model.

02

Plan

Turn a bigger task into steps before you start executing it.

03

Agent

Let your local model perform useful work with files and approved tools on your machine.

PowerShell
# route and answer
$ prompt-router "explain what a WeakMap is"

# force a destination
$ prompt-router --local "refactor this function"
$ prompt-router --online "compare Rust and Go GC"

# understand and measure
$ prompt-router doctor
$ prompt-router stats
$ prompt-router models recommend -q 5

Command line

Same engine. Same guarantees.

CLI and app share one engine, so safety can't drift. --json to script it, --explain to see why.

  • setup — scan hardware, pick a model
  • doctor — check every engine, model and key
  • stats & history — split, privacy, latency, redacted
  • Needs Node 22+ and one engine — local or cloud
PromptRouter

Ask. Let PromptRouter decide.

Use local AI when your computer can handle it. Use cloud AI when the task needs it.

Private beta for Windows x64, macOS Apple Silicon and Linux x64. Installers are currently unsigned.

Try PromptRouter → Other platforms and release notes