#Composite Keys

A composite key binds several groups at once: each group gets a prefix, and the prefix on the model ID picks the group at call time. No per-group keys, no config juggling in the client.

#Creating one

1. Click Create Key on the API Keys page.
2. Name it and switch on Composite Key.
3. In the mapping editor add rows, choosing a group and typing a prefix for each.
4. Add as many mappings as needed and save.

Editing an existing key offers the same switch and flow.

SCREENSHOT

#Prefix rules

• 1–32 characters; letters, digits, underscores and hyphens only;
• Case-insensitive — GPT and gpt are the same prefix and cannot both exist;
• Each group may appear only once per key;
• Mapping count per key is capped — the creation dialog shows the limit.

#Calling

Write the model ID as prefix/modelID. Suppose fast → a high-speed group and cheap → a budget group:

{ "model": "fast/claude-sonnet-4", "messages": [{ "role": "user", "content": "Hello" }] }

{ "model": "cheap/deepseek-v3", "messages": [{ "role": "user", "content": "Hello" }] }

#Model lists are auto-prefixed

Fetching the model list with a composite key aggregates models from all bound groups, each already carrying its prefix. Clients can pick straight from the list — no manual prefixing.

#Billing

Identical to regular keys: billed at the multiplier of whichever group the prefix hits; quotas, rate limits and expiry apply the same way. See Billing.

#Unsupported interfaces

Composite keys do not support WebSocket / Realtime interfaces — long-lived connections may switch models mid-session, so a prefix cannot pin the group. Use regular keys there. Ordinary chat, completion, image and usage endpoints all work.

#Common errors

Prefix problems all return HTTP 400:

#Related

• Create an API Key — the regular key flow
• Billing — how group multipliers are applied
• Model plaza — models available per group