augur

AI feasibility, with receipts

Can AI do this? Find out before you build.

Drop in a website URL or describe a task. augur asks the model you choose for a confidence-rated verdict and three concrete implementation plans — and tells you what the analysis itself cost, when the provider reports it.

Run it yourself Changelog
The layout of a result, for an illustrative task: “Sort incoming support email by topic and urgency.” Plans, prices and numbers come from the model you pick.

What it does

A concrete answer to a hand-wavy question.

Useful for scoping automation work and getting a first plan on paper — not for skipping the build.

01

Three ways in

Paste a live website URL, pick a suggested common task, or describe anything in plain words. Pages are fetched with a guard that refuses private and loopback addresses.

02

A verdict with a confidence

“Yes, AI can do this” or “Partially automatable”, with a 0–100 confidence score instead of a shrug.

03

Three ranked plans

Each names the tools and services involved and an estimated monthly cost range. Expand any plan — or all three — for the full setup steps.

04

Honest token economics

Shows the tokens the provider reported and a cost estimated from list prices. When usage or a price is missing, it says so rather than guessing.

05

Your model, your choice

Anthropic, OpenAI, open models through OpenRouter, local models via Ollama or LM Studio, or any OpenAI-compatible endpoint — picked per analysis.

06

Feedback and 12 languages

Rate any plan with a thumbs up or down and a note; it's stored with the plan. The interface speaks twelve languages, in light or dark.

Bring your own key

Set up where your AI runs, in a minute.

A first-run wizard asks whether you'll use hosted models, local ones, or both. Server-side keys are optional.

  • Encrypted in your browser. Keys are sealed with a non-extractable WebCrypto key and can sit behind a passphrase.
  • Sent per request, never kept. The server uses a key for the one analysis and doesn't store or log it.
  • Test before you trust. Check a key or endpoint from Settings without saving anything.
augur's first-run setup: choose hosted models (Anthropic, OpenAI, OpenRouter with your API key), local models (Ollama, LM Studio, llama.cpp) or both. The same setup step in dark mode.

Current limits

What augur doesn't claim.

An initial answer from one model, written down so you can check it — not a guarantee that the workflow will work.

One model at a time

Each analysis runs on the single model you select. It doesn't compare providers side by side or pick the best one for you.

Plans aren't tested

Generated plans and their monthly cost ranges are estimates. Nothing validates a plan against a deployed workflow.

Built for one person

Analyses are kept in a single local store. Optional sign-in doesn't partition data by user, so don't put sensitive tasks into a shared deployment.

Run it

One port, your machine.

There's no hosted augur yet. It runs locally with Node — the server and the app share a single port.

From a checkout
npm install
cp .env.example .env   # every provider key is optional
npm run dev            # open http://localhost:5000

Needs Node.js 22 or newer. Add a model in the setup wizard, or put a key in .env. docker compose up --build works too.