Quickstart
Run your first request against the gateway.
Hyde Inference uses the OpenAI API format. The only changes from OpenAI are the base URL and the key.
Prerequisites
- A Hyde API key. Create one in the dashboard if you don't have one yet.
- The OpenAI SDK for your language, or any HTTP client.
1. Get an API key
Sign in at inference.hyde.tech with Google, or with your email and a one-time code. Open the API keys page and click Generate API key. Name it after the thing that will use it (for example ada-cursor); the name shows up in usage and traces. Copy the sk-… value now, because it is shown once.
2. First completion
Install the OpenAI SDK:
pip install openaiPoint the client at the gateway with base_url:
from openai import OpenAI
client = OpenAI(
base_url="https://inference.hyde.tech/v1",
api_key="sk-PASTE-YOUR-KEY",
)Then make a request. Swap model to use any Hyde model. The Models page lists the catalog:
completion = client.chat.completions.create(
model="hyde/glm-5.3-flash",
messages=[
{"role": "user", "content": "Hello!"}
],
)
print(completion.choices[0].message.content)Streaming. The same endpoint supports SSE streaming. Pass stream: true exactly as you would with OpenAI.
Embeddings
The same client turns text into vectors through /v1/embeddings. Use hyde/qwen3-embedding-8b or hyde/nemotron-3.5-embed-8b. Each returns one 4,096-dimension vector per input, for up to 8,192 tokens per input:
result = client.embeddings.create(
model="hyde/qwen3-embedding-8b",
input=["The cat sat on the mat.", "A kitten rested on the rug."],
)
print(len(result.data[0].embedding))Embedding models have no fallback. If the embedding capacity is briefly unavailable, the request fails instead of being served elsewhere. Retry after a short wait.
3. Wire up your agent
Coding agents take a couple of env vars or a config file rather than the SDK. Each guide covers the exact config for that tool:
- Claude Code uses a settings file with model discovery
- Cursor takes a custom OpenAI base URL
- OpenCode works with one opencode.json
- GitHub Copilot (VS Code) connects with a Custom Endpoint
What's next
- See Models for the model IDs and what each one is good at.
- See Claude Code, Cursor, or OpenCode for agent setup.