AI Gateway

An OpenAI-compatible chat endpoint, a transparent Claude (Anthropic) proxy, the ccas analytics CLI, and AI Skills, a public agent-skill registry. One account, three credential types.

AI Gateway#

The Hep.gg AI Gateway is two runtimes under one account.

The first is an LLM gateway at https://ai.hep.gg. It exposes an OpenAI-compatible /v1/chat/completions endpoint (point any OpenAI SDK at it) and a transparent passthrough proxy for the Anthropic Claude API so calls made by Claude Code can be metered and analyzed. The ccas CLI installs from the same host and wires Claude Code into the proxy.

The second is AI Skills at https://skills.hep.gg, a public registry of agent skills (SKILL.md files plus supporting files) served as plain markdown for coding agents to read and import. Skills are part of the same AI account.

Three credential types#

This app uses three distinct credentials. Mixing them up is the most common source of 401s, so keep them straight.

A fourth, separate credential, the skills access key (48 hex chars), authorizes private AI Skills access and the skill import endpoint. It lives entirely on skills.hep.gg; see AI Skills.

Authentication summary#

• OpenAI-compatible endpoints (/v1/chat/completions, /v1/models): Authorization: Bearer sk-hyd-.... The key is SHA-256 hashed and matched against active keys; a disabled key or a non sk-hyd- value returns 401 {"error":"Invalid API key"}.
• Account and management endpoints (/me, /keys, /sync/*): Authorization: Bearer hyd_... (your master token).
• Claude proxy (/anthropic/...): the pt_ proxy token is the first path segment. You still send your own Anthropic credentials in the request, which are forwarded upstream unchanged.
• AI Skills reads are public. Private access and skill import use a capability URL containing your skills access key.

There is no public endpoint to create a master token or a skills access key, those come from the dashboard. API keys and proxy tokens are minted with the master token. Never commit any of these to source control or print them in logs.

Quickstart#

The smallest working call is an OpenAI-compatible chat completion. Mint an sk-hyd- key in the dashboard (or via POST /keys), then point any OpenAI client at https://ai.hep.gg/v1. The model slug is cf-gpt-oss-20b.

curl https://ai.hep.gg/v1/chat/completions \
  -H "Authorization: Bearer $HYD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cf-gpt-oss-20b",
    "messages": [{ "role": "user", "content": "Say hello in one word." }],
    "max_tokens": 256
  }'

import OpenAI from "openai";
 
const client = new OpenAI({
  baseURL: "https://ai.hep.gg/v1",
  apiKey: process.env.HYD_API_KEY, // sk-hyd-...
});
 
const res = await client.chat.completions.create({
  model: "cf-gpt-oss-20b",
  messages: [{ role: "user", content: "Say hello in one word." }],
  max_tokens: 256,
});
 
console.log(res.choices[0].message.content);

What you can do#

Available models#

The OpenAI-compatible endpoints currently serve a single model. List the minting catalog any time at GET https://ai.hep.gg/models (public, no auth).