Point at any endpoint that implements OpenAI's `/v1/chat/completions` with `tool_calls`. Covers Groq's sub-second inference, Together's open-weight catalog, self-hosted Ollama on your LAN, and vLLM clusters. Our streaming adapter handles the incremental `tool_calls` chunk format.

Install

1 · Config

Append to ENABLED_PLUGINS in wrangler.toml:

openai-compatible

2 · Secrets

Run each from your Worker project:

• wrangler secret put OPENAI_COMPATIBLE_URL Base URL, e.g. https://api.groq.com/openai/v1
• wrangler secret put OPENAI_COMPATIBLE_KEY optional — Bearer token for the endpoint. Optional for local Ollama.

OPENAI_COMPATIBLE_URL=Base URL, e.g. https://api.groq.com/openai/v1
OPENAI_COMPATIBLE_KEY=# optional — Bearer token for the endpoint. Optional for local Ollama.

3 · Steps

1. Add the endpoint's hostname to ALLOWED_HOSTS before your first call.

Tags

Related

• Anthropic (direct) — Claude Messages API with native tool-use streaming.
• Cloudflare AI Gateway — Many providers behind one plugin with BYOK via Secrets Store.
• Codex / ChatGPT subscription — Use your ChatGPT Plus/Pro subscription instead of a separate API bill.