QUICK START
Bring your own agent — Claude Code, OpenClaw Hermes, Cline, Aider, Continue, or anything that speaks OpenAI. Point it at one base URL, drop in your key, and code against open models on shared GPUs.
your agent → api.openalchemy.io → open-weight models on the Grid
THREE STEPS
Sign in to the console, create a project, and mint an API key. The free tier needs no credit card.
Set the base URL and API key in your agent's settings — one OpenAI-compatible URL, or the Anthropic-compatible one for Claude Code.
Choose a model id such as llama-3.1-8b-instruct and start sending requests. Switch models any time without changing code.
TWO WAYS TO CONNECT
Most agents speak the OpenAI API. Claude Code speaks the Anthropic Messages API. OpenAlchemy exposes both, so whichever your tool uses, the same key works.
Drop this base_url and key into the OpenAI SDK or any OpenAI-compatible agent — Hermes, Cline, Aider, Continue, Codex CLI.
Set this as ANTHROPIC_BASE_URL so Claude Code routes its Messages API calls to OpenAlchemy with no proxy in between.
AGENT SETUP
Each snippet uses the OPENALCHEMY_API_KEY you created and the example model llama-3.1-8b-instruct. Swap in any model from the catalog.
Anthropic's terminal-native coding agent. Route it through the Anthropic-compatible endpoint.
# Claude Code speaks the Anthropic Messages API — point it at
# OpenAlchemy's Anthropic-compatible endpoint and run as usual.
export ANTHROPIC_BASE_URL="https://api.openalchemy.io"
export ANTHROPIC_AUTH_TOKEN="$OPENALCHEMY_API_KEY"
export ANTHROPIC_MODEL="llama-3.1-8b-instruct"
claudePoint Hermes' OpenAI-compatible provider at OpenAlchemy and choose a model.
# Hermes uses an OpenAI-compatible provider. Export the base URL and
# key, then select the model inside Hermes.
export OPENAI_BASE_URL="https://api.openalchemy.io/v1"
export OPENAI_API_KEY="$OPENALCHEMY_API_KEY"
# Provider: OpenAI-compatible
# Base URL: https://api.openalchemy.io/v1
# Model: llama-3.1-8b-instruct
hermesThe VS Code autonomous coder (and the Roo Code fork). Use the 'OpenAI Compatible' provider.
# VS Code → Cline (or Roo Code) settings → API Provider:
API Provider = OpenAI Compatible
Base URL = https://api.openalchemy.io/v1
API Key = $OPENALCHEMY_API_KEY
Model ID = llama-3.1-8b-instructTerminal pair-programmer. Aider routes openai/llama-3.1-8b-instruct through your OPENAI_API_BASE.
export OPENAI_API_BASE="https://api.openalchemy.io/v1"
export OPENAI_API_KEY="$OPENALCHEMY_API_KEY"
aider --model openai/llama-3.1-8b-instructThe open-source IDE assistant. Add an openai provider with a custom apiBase.
# ~/.continue/config.yaml
models:
- name: OpenAlchemy
provider: openai
apiBase: https://api.openalchemy.io/v1
apiKey: ${OPENALCHEMY_API_KEY}
model: llama-3.1-8b-instructAny tool built on the OpenAI SDK — including Codex CLI — connects with the same base_url + key.
from openai import OpenAI
client = OpenAI(
base_url="https://api.openalchemy.io/v1",
api_key="$OPENALCHEMY_API_KEY",
)
resp = client.chat.completions.create(
model="llama-3.1-8b-instruct",
messages=[{"role": "user", "content": "Write a haiku about distributed GPUs."}],
)
print(resp.choices[0].message.content)
# Codex CLI: add a custom provider with the same base_url + key.FAQ
Yes. The /v1 surface mirrors the OpenAI REST API — chat completions, embeddings, audio, and more — so existing OpenAI SDKs and tools work by changing only the base_url and key.
Yes. OpenAlchemy also exposes an Anthropic-compatible endpoint. Set ANTHROPIC_BASE_URL to https://api.openalchemy.io and ANTHROPIC_AUTH_TOKEN to your key — no proxy needed.
Open-weight models across modalities — Llama, Qwen, DeepSeek, Gemma for chat, plus Whisper, embeddings, rerankers and more. List them at GET /v1/models and pass the id in your request.
No. The free tier lets you create one project and run your first 1,000 requests without a card.
Usage, quotas, and request logs are scoped per project; rate limits and access control apply per domain. Your key carries that scope automatically.
Streaming responses (SSE) are rolling out — some endpoints currently return 501 for stream:true. Non-streaming requests are fully supported today.
START BUILDING
Create a free project, grab a key, and point your coding agent at open models running on the Grid.