[Feature Request] Add @llamaindex/openrouter LLM provider package

[alex129]

[Feature Request] Add @llamaindex/openrouter LLM provider package

Feature Description

LlamaIndex Python has a dedicated llama-index-llms-openrouter package that provides first-class OpenRouter integration, allowing users to easily access 300+ LLM models through a single, unified API. However, LlamaIndex.TS currently has no equivalent package.

The TypeScript repo supports providers like OpenAI, Anthropic, Groq, Mistral, Fireworks, DeepSeek, Replicate, Together, HuggingFace, DeepInfra, and Gemini — but OpenRouter is missing.

Why OpenRouter?

OpenRouter acts as a unified gateway to hundreds of models (OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, and many more) through a single API key and a standardized OpenAI-compatible API. This is especially valuable for:

• Developers who want model flexibility without managing multiple API keys and provider-specific packages.
• Cost optimization — OpenRouter routes to the cheapest provider for a given model automatically.
• Fallback/redundancy — automatic failover between providers if one is down.
• Access to open-source models that aren't directly available through other LlamaIndex.TS providers.

Python parity

In Python, the integration is straightforward:

from llama_index.llms.openrouter import OpenRouter

llm = OpenRouter(
    api_key="<your-api-key>",
    max_tokens=256,
    context_window=4096,
    model="anthropic/claude-sonnet-4",
)
response = llm.complete("Hello World!")

There is no equivalent in LlamaIndex.TS today.

Proposed Solution

Create a new @llamaindex/openrouter package that:

1. Wraps the OpenRouter API (which is OpenAI-compatible, so it could potentially extend @llamaindex/openai with a custom baseURL).
2. Supports chat, complete, and their streaming variants.
3. Exposes OpenRouter-specific features such as:
   • provider ordering/filtering (see [OpenRouter provider routing docs](https://openrouter.ai/docs/features/provider-routing)).
   • Model-specific context_window and max_tokens defaults pulled from OpenRouter's model metadata.
   • Cost tracking via response headers (x-openrouter-cost).

Desired API (example)

import { OpenRouter } from "@llamaindex/openrouter";

const llm = new OpenRouter({
  apiKey: "<your-api-key>",
  model: "anthropic/claude-sonnet-4",
  maxTokens: 256,
});

const response = await llm.complete({ prompt: "Hello World!" });

Workaround

Currently, users can hack around this by using @llamaindex/openai with a custom baseURL pointed at https://openrouter.ai/api/v1, but this is undocumented, fragile, and doesn't expose OpenRouter-specific features like provider routing or cost headers.

Additional Context

• OpenRouter's own docs list LlamaIndex as a [supported framework](https://openrouter.ai/docs/guides/community/frameworks-and-integrations-overview) for both Python and TypeScript, but the TypeScript integration doesn't actually exist as a dedicated package.
• The Python package source can serve as reference: llama-index-llms-openrouter on PyPI.
• Since OpenRouter is OpenAI-compatible, the implementation effort should be relatively low — likely extending the existing OpenAI provider with a custom base URL and additional config options.