# Kev Model Deploy

Deploys the Kev decision model as a private HTTPS endpoint on Modal.

- Repository: https://github.com/jaredpalmer/kev (9751 stars)
- Category: DevOps
- License: Apache-2.0
- Last updated: 2026-10-27
- Runs scripts: yes, review before use

## Install

Claude Code: `npx skills add jaredpalmer/kev --skill kev-deploy -g -a claude-code -y`
Codex: `npx skills add jaredpalmer/kev --skill kev-deploy -g -a codex -y`
Cursor: `npx skills add jaredpalmer/kev --skill kev-deploy -g -a cursor -y`
Gemini CLI: `npx skills add jaredpalmer/kev --skill kev-deploy -g -a gemini-cli -y`

## Description

Deploy a Kev decision model (the open Jev-style System One model) as the user's own TypeSafe-compatible HTTPS endpoint on Modal with one command, wire it into their code, and take it down. Use when someone wants to host Kev, get a Kev API URL, replace Jev / TypeSafe calls with a self-hosted model, pick a Kev size or GPU, add an API key, keep an endpoint warm, or stop and remove a Kev deployment.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Deploy Kev on Modal

`scripts/kev_serve.py` is a single self-contained Modal app. `modal deploy` on it builds an image with the kev package at
a pinned commit, loads a Kev checkpoint from the Hugging Face Hub on a GPU sized for it, and serves TypeSafe's System One
protocol (`POST /v1/systemone`, `GET /v1/models`) at `https://<workspace>--kev-api.modal.run`. It scales to zero when idle.
The file is short on purpose: if the user's case does not fit, read it and edit it.

## 1. Check the prerequisites

```bash
pip install modal==1.5.5
modal profile current 2>/dev/null || modal setup      # opens a browser to sign in; the user must do this step
modal skills install -y                               # optional: Modal's own agent skill + docs (into .agents/, or -g for home)
```

If `modal setup` is needed, tell the user and wait; do not try to authenticate for them. `modal skills install` gives you
Modal's official skill and documentation, which helps with anything beyond this file: GPUs, secrets, volumes, logs, billing.

## 2. Pick the model

| `KEV_MODEL` | GPU (automatic; fallbacks in parentheses) | $/h while up | Model time, 6 questions (new / repeated state) | Cold start (cached weights) | When |
| --- | --- | --- | --- | --- | --- |
| `jaredpalmer/kev-0.8b` | L4 (L40S) | 0.80 | 23 / 16 ms | ~40 s | cheapest, prototyping |
| `jaredpalmer/kev-4b` (default) | L40S (H100) | 1.95 | 42 / 28 ms (H100: 18 / 13 ms) | ~35 s | the default: best quality per dollar |
| `jaredpalmer/kev-9b` | H100 (H200, L40S) | 3.95 | 24 / 17 ms | ~55 s | accuracy on smaller GPUs |
| `jaredpalmer/kev-27b` | B200 (H200, H100) | 6.25 | 47 / 32 ms (H200: 67 / 50 ms) | ~50 s | best released accuracy; 51 GB of full weights |

The rest: https://github.com/jaredpalmer/kev/blob/HEAD/skills/kev-deploy/SKILL.md