Inference API

An OpenAI-compatible inference endpoint. Authenticate with a Bearer API key and call it like any OpenAI client.

Connecting an agent? Use your client's OpenAI / OpenAI-compatible provider, pointed at the base URL below, and put your sk-… key in the client's API-key field directly — not only OPENAI_API_KEY. Some "custom provider" integrations ignore the env var and send a placeholder, which returns 401.

Base URL

https://inference.holbrookecapital.com/v1

Authentication

Send Authorization: Bearer <your sk-… key> on all inference calls (/v1/chat/completions) and on /v1/models — both return 401 without a key. (Ollama-style discovery endpoints — /api/tags, /api/show, /api/version — respond keyless so agent clients can auto-detect the provider before a key is set.)

Models

qwen-uncensored — unfiltered assistant, agent-friendly (recommended default) · up to 256K context
gemma-4-31b — general assistant · up to 96K context
gemma4-26b-uncensored — uncensored assistant (MoE 26B-A4B) · up to 128K context

Streaming is supported ("stream": true, OpenAI SSE). The OpenAI tools / function-calling schema is accepted.

Example

curl https://inference.holbrookecapital.com/v1/chat/completions \
  -H "Authorization: Bearer $HOLBROOKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-uncensored",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Latency

First call to an idle model loads it on demand (~3–5s); it then stays warm and short follow-ups are often <1s.
Big prompts take longer to prefill — a fresh 20–30K-token agent context is ~15–25s. Expected, not a failure. Use streaming to get the first token as soon as prefill finishes.

Agents: a machine-readable usage guide lives at /skill.md.

Need a key? Create a free account.