vLLM (self-hosted)

vLLM serves OpenAI-compatible chat completions with logprobs, so Verdict's LLM link gets a real distribution. Set LLM_VLLM_BASE to your server.

LLM_CHAIN=vllm,anthropic   # failover to a second endpoint is one more name
LLM_VLLM_BASE=http://localhost:8000/v1
LLM_VLLM_MODEL=meta-llama/Llama-3.3-70B-Instruct
# LLM_VLLM_API_KEY only if you started vLLM with --api-key

Every endpoint takes the same knobs: LLM_<NAME>_TEMPERATURE (none omits it), LLM_<NAME>_BODY (JSON merged into every request), LLM_<NAME>_PRICE (input,cached,output per million, for the cost ledger), LLM_<NAME>_NOTES (a per-model notes file). Two models of one provider: LLM_FAST_PRESET=vllm under another name. Environment has the full list.

Check it: npm run gateway, then curl localhost:8787/api/health — the llm.chain entry shows the endpoint, whether its key is set, and its state.

Provider docs: https://docs.vllm.ai

---

Source: docs/providers/vllm.md — edit it there; the site mirrors the repository.