Skip to content

Free · no card · no sign-up

Get started free.

Install Halfly in minutes. It watches your real traffic and shows what you would save, before you pay a penny. When you like the number, switch it on for £5 a month.

  1. Step 1: Install Halfly

    One command. Requires Python 3.10+.

    Terminal
    pip install halfly

  2. Step 2: Create your config

    halfly init writes halfly.yaml. Pick your strong model (the one you use today) and a cheap one, and set their prices in US dollars per million tokens: they drive every savings number Halfly reports.

    Terminal
    halfly init

  3. Step 3: Add your provider key

    Put it in a .env file next to halfly.yaml (Halfly loads it), or set it in your environment. On OpenAI or another provider? Use the key name from halfly.yaml, such as OPENAI_API_KEY. Your keys stay on your machine.

    .env
    ANTHROPIC_API_KEY=sk-ant-...

  4. Step 4: Start the proxy

    Halfly now listens on http://localhost:8000, in front of your providers.

    Terminal
    halfly serve

  5. Step 5: Point your app at Halfly

    Change the base URL to http://localhost:8000/v1 and ask for the model halfly-auto. Your SDK, streaming and tool calls keep working.

    app.py
    from openai import OpenAI client = OpenAI(base_url="http://localhost:8000/v1") reply = client.chat.completions.create(    model="halfly-auto",    messages=[{"role": "user", "content": question}],)

  6. Step 6: Watch the “would save” number grow

    Open localhost:8000/dashboard. On the free plan Halfly runs in observe mode: every request still goes to your strong model, so your users see no change, while the dashboard counts what routing would save you, and at what checked quality.

  7. Step 7: Switch the savings on

    Like the number? Buy Starter (£5 a month) or Pro (£10 a month). After checkout you get your licence key on screen: activate it, restart halfly serve, and routing switches on. If you ever stop paying, Halfly goes back to observe mode and your app keeps working.

    Terminal
    halfly licence activate <key>halfly serve
    Buy Starter · £5/moBuy Pro · £10/mo

Like the number?
Switch it on.

Start free, then pick a plan when your dashboard shows the saving is worth it.

Free

£0forever

See what you would save, on your own traffic.

  • Installs in minutes. No card, no sign‑up
  • Observe mode: every request still goes to your strong model, so your users see no change
  • Dashboard shows what routing would save, and at what checked quality
  • Budget caps on your AI spend
Start free

Switch the savings on

Starter

£5a month

Turn the savings you saw into real ones.

  • Routing with the quality dial: easy prompts to the cheap model, hard ones to your best
  • Exact cache: repeat questions are answered for free
  • Fallback to the other model on errors and rate limits
  • Budget caps, savings dashboard and halfly report
Buy Starter

Pro

£10a month

Savings that keep growing as Halfly learns.

  • Everything in Starter
  • Self-learning: halfly learn tunes routing on your own checked traffic
  • Cost per feature of your app, with quality per feature
  • Weekly Slack-style savings report and budget alerts
  • Prompt compression for long prompts
Buy Pro

Cancel any time. If you stop paying, your app keeps working: Halfly simply goes back to free observe mode.

Halfly runs on your servers, so your prompts never reach us. Prices in pounds, per month. Bigger team? Talk to us.