Relay

Connect

The decision engine (Jev, Kev or your AI)

Updated 26 Oct 20266 min read

Before anyone drafts a reply, Relay sorts each email: what kind it is, how urgent, whether it needs a reply, whether it looks like someone trying to trick an assistant, and which agent should take it. That's the decision engine. Choose it under Settings > AI > Decision engine:

  • Automatic (the default): Jev when its key is set, otherwise your main AI. You don't need a Jev key to start.
  • Jev (TypeSafe AI, cloud): a service built for typed decisions with confidence scores.
  • Kev (on your computer, open source): a free model that answers the same questions as Jev, on your own hardware.
  • Use my main AI: the AI you chose under Settings > AI. With a local AI, nothing leaves your PC.

Whichever you choose, nothing falls back from your PC to the cloud. If a local engine can't answer, the email goes to a person and the case says why.

Getting a Jev key

Jev is made by TypeSafe AI. These steps were checked on 26 October 2026.

  1. Go to typesafe.ai and choose Sign in, or go straight to console.typesafe.ai. You sign in with Google or an emailed code.
  2. Open API Keys in the console and create a key. Copy it straight away: it's shown only once.
  3. In Relay, paste it into Settings > AI > Jev (TypeSafe AI) API key.

Note: TypeSafe paused new sign-ups in late September 2026, and access has at times been by waitlist. If you can't get an account, use Automatic or Use my main AI: Relay works fully without Jev. Jev is priced per token by TypeSafe, and output tokens have been free; check typesafe.ai for current terms.

Setting up Kev

Kev is an open-source family of decision models (Apache-2.0) that work like Jev. It runs as a small server that Relay talks to.

What you need: Kev-4B, the standard model, needs an NVIDIA graphics card with 24 GB or an Apple Silicon Mac with 32 GB. The smaller Kev-0.8B runs on a 4 GB card or any Apple Silicon Mac. On Windows, Kev usually runs inside WSL with an NVIDIA card. It can also run on another computer on your network.

  1. Install Python 3.13 and uv.
  2. Download Kev: git clone https://github.com/jaredpalmer/kev.git, then cd kev and uv sync --extra serve.
  3. Start it: uv run --extra serve python -m kev.serve --run jaredpalmer/kev-4b --port 8009. The first start downloads the model. For Kev-0.8B, use its name from the Kev page instead.
  4. In Relay, open Settings > AI, set Decision engine to Kev, leave Kev address empty (or enter the other computer's address, such as http://192.168.1.20:8009), and press Test Kev.

Kev settings in Relay

If you started Kev with KEV_API_KEY, enter the same key under Kev API key. Keep Kev running while Relay works. If it stops, Relay says so on each email it couldn't sort.

Kev on a computer outside your network counts as the cloud, so Relay hides private details before sending, as it does for Jev.