`@thinwrap/llm`

@thinwrap/llm for TypeScript — one typed Chat facade over 18 LLM providers (15 OpenAI-compatible + native Anthropic, Bedrock, Gemini), plus embeddings on 11. Stateless, zero deps, BYO fetch — the in-process complement to an LLM gateway.

npm · @thinwrap/llm@1.0.0 GitHub · thinwrap/llm-ts

TypeScript constraints

Runtime dependencies: Zero runtime dependencies
BYO HTTP client: injectable fetch parameter (Node 18+ native fetch)
Provenance: npm provenance + Sigstore
Minimum runtime: Node 18.20+

Node 18 and 20 emit a one-time ExperimentalWarning on the first fetch() invocation: "ExperimentalWarning: The Fetch API is an experimental feature. This feature could change at any time."
The warning is informational only — fetch() in Node 18.20+ and Node 20.x is functionally identical to the stable fetch in Node 21+. Suppress with the NODE_NO_WARNINGS=1 env var or the --no-warnings flag if it clutters CI logs.
Thinwrap does not patch around the warning, does not silence it on the consumer's behalf, and does not require a polyfill. It is surfaced exactly as Node's runtime surfaces it.

30-second install

npm install @thinwrap/llm

2-minute end-to-end example

One Chat facade, every provider. Switching vendor is the provider id plus the model — the complete(input) call and theChatResult shape stay identical.

import { Chat } from '@thinwrap/llm';

const chat = new Chat('openai', { apiKey: process.env.OPENAI_API_KEY! });

const res = await chat.complete({
  model: 'gpt-4o-mini',
  messages: [{ role: 'user', content: 'Say hi in one word.' }],
});

console.log(res.message.content, res.usage);

Switch vendor by changing the provider id + model — swap to 'anthropic', 'gemini', 'bedrock','groq', or any of the 15 OpenAI-compatible providers and pass the matching config. Native adapters normalize Anthropic, Bedrock, and Gemini onto the same ChatResult.

const chat = new Chat('anthropic', { apiKey: process.env.ANTHROPIC_API_KEY! });
// same .complete(input) shape, same ChatResult

Supported providers

18 providers ship with @thinwrap/llm@1.0.0.
Provider	Operations	Auth	README
`openai` OpenAI	chat, embeddings	bearer	docs
`azure-openai` Azure OpenAI	chat, embeddings	header	docs
`anthropic` Anthropic	chat	header	docs
`gemini` Google Gemini	chat	header	docs
`bedrock` AWS Bedrock	chat	aws-sig-v4	docs
`mistral` Mistral	chat, embeddings	bearer	docs
`cloudflare` Cloudflare Workers AI	chat, embeddings	bearer	docs
`together` Together AI	chat, embeddings	bearer	docs
`fireworks` Fireworks AI	chat, embeddings	bearer	docs
`deepinfra` DeepInfra	chat, embeddings	bearer	docs
`openrouter` OpenRouter	chat, embeddings	bearer	docs
`groq` Groq	chat	bearer	docs
`deepseek` DeepSeek	chat	bearer	docs
`xai` xAI (Grok)	chat	bearer	docs
`perplexity` Perplexity	chat	bearer	docs
`vllm` vLLM	chat, embeddings	bearer	docs
`ollama` Ollama	chat, embeddings	bearer	docs
`lmstudio` LM Studio	chat, embeddings	bearer	docs

Migration paths

The in-process complement to a gateway

Thinwrap LLM is a thin, in-process, zero-egress facade — it does not proxy your traffic. It normalizes the request/response shape across vendors so your code stays vendor-neutral; routing, budgets, and observability stay wherever you already run them (your own gateway or none). Provider-specific request fields ride through _passthrough; the decoded vendor body is always on result.raw.

Streaming

for await (const delta of chat.stream(input)) {
  if (delta.contentDelta) process.stdout.write(delta.contentDelta);
}

Embeddings

The OpenAI-float subset (11 of the 18 providers — see the table above) also exposes an Embeddings facade with the same provider-switching model.

Timeouts & cancellation

The wrapper holds no timeout of its own — it awaits whatever your injectedfetch does. Bound a hung provider by passing afetch that sets AbortSignal.timeout(…); an aborted request surfaces as a ConnectorError, same as every other failure.

Cross-language siblings

The PHP (/packages/llm-php) and Go (/packages/llm-go) variants ship the same facade shape and provider set — cross-language parity is a design goal, not a hard release gate.

Repo README