Index of every page on this site: llms.txt
All skills
jaredpalmer avatar

/kev-deploy

@2074cc4 #16 trending · Oct 2026
by Jared Palmerjaredpalmer/kev9.8k stars
670
Run it yourself

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.

Use this Skill: https://skilld.dev/gh/jaredpalmer/kev/kev-deploy

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

README.md

≈681 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Deploy Kev on Modal

Your own Kev endpoint, speaking TypeSafe's System One protocol, in three commands. It scales to zero when idle, so an unused endpoint costs nothing.

pip install modal==1.5.5 && modal setup            # once: sign in to Modal in the browser
curl -LO https://raw.githubusercontent.com/jaredpalmer/kev/main/skills/kev-deploy/scripts/kev_serve.py
KEV_API_KEY=$(openssl rand -hex 24) modal deploy kev_serve.py

The deploy prints https://<your-workspace>--kev-api.modal.run. Keep the key: requests need Authorization: Bearer <key>. Point any TypeSafe client at the URL:

client = TypeSafeClient(api_key=KEV_API_KEY, base_url="https://<your-workspace>--kev-api.modal.run", model="kev-latest")
Model Set GPU ($/h while up) Warm model time, 6 questions (new / repeated state) First request after idle
Kev-0.8B KEV_MODEL=jaredpalmer/kev-0.8b L4 (0.80) 23 / 16 ms ~40 s
Kev-4B (default) nothing L40S (1.95) 42 / 28 ms ~35 s
Kev-9B KEV_MODEL=jaredpalmer/kev-9b H100 (3.95) 24 / 17 ms ~55 s
Kev-27B KEV_MODEL=jaredpalmer/kev-27b B200 (6.25) 47 / 32 ms ~50 s

Concurrent requests are batched in each container (Kev-4B: about 100 requests/s on an H100 in-process). Modal's ordinary web endpoint retains its 64-request limit and 32-request autoscaling target: raising its limit or scaling target reduced throughput in the HTTP pilot. KEV_FLASH=1 (with KEV_REGION=us-east, or another supported routing region) uses a Modal Server to bypass the Function input plane, targeting 64 concurrent requests per container (formerly 32). It remains opt-in and keeps at least one paid GPU warm; it does not scale to zero. Bearer authentication and the System One API are unchanged. See the paired HTTP measurements and deployment details. Increasing a concurrency limit alone is not a latency guarantee.

The round trip adds the network and Modal's proxy (about 80-100 ms on the ordinary web endpoint from a US client with a kept-alive connection; direct mode can be lower); KEV_REGION=us keeps the ordinary web endpoint near US callers. The very first deploy also downloads the weights and compiles kernels (1-2 minutes; several minutes for Kev-27B's 51 GB); later cold starts reuse the cache.

KEV_MIN_CONTAINERS=1 keeps it warm; modal app stop kev takes it down. With an agent, npx skills add jaredpalmer/kev@kev-deploy and ask it to deploy Kev; it follows SKILL.md. modal skills install adds Modal's own agent skill and docs alongside it, for anything beyond this file (GPUs, secrets, logs, billing).

Source: SKILL.md on GitHub

Skill behaviors

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

Third-party checks

1 warning15d3 checks · Risk SAFE
  • Gen Agent Trust Hub15d

    The skill provides a legitimate way to deploy the 'Kev' decision model on the Modal cloud platform. It uses standard deployment patterns, handles secrets securely via Modal's secret management, and downloads resources from the author's official GitHub repository. No malicious patterns or security risks were detected.

  • Socket15d

    No alerts

  • Snyk15d

    Risk: MEDIUM · 1 issue

Provenance

Signed by skilld at 2074cc4. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 weeks ago.

Activeupdated 4 weeks ago

Capability

Other metadata
compatibility
Requires Python 3.10+ and a Modal account (`pip install modal==1.5.5 && modal setup`, free tier works for the small models). No GPU, no clone of the Kev repo, no Hugging Face account for the public checkpoints.
metadata
{
  "author": "jaredpalmer",
  "version": "1.1",
  "repository": "https://github.com/jaredpalmer/kev"
}

README badge

README badge for jaredpalmer/kev/kev-deploy