LlamaIndex LLM Integration: BlockRun

BlockRun is one OpenAI-compatible gateway to models from OpenAI, Anthropic, Google, DeepSeek, xAI, NVIDIA and others, where every call pays for itself in USDC over x402. There is no API key and no account: you pick the chain the wallet pays from, Base or Solana.

Why not OpenAILike?

OpenAILike(api_base="https://blockrun.ai/api/v1", api_key="fake") reaches the gateway and stops at the first paid call. The gateway answers an unpaid request with HTTP 402 and a price, and nothing in the OpenAI client knows how to sign a payment for it. This package hands that step to the blockrun-llm SDK, which signs the payment locally and retries the request with the signature attached.

Install

pip install llama-index-llms-blockrun            # pay on Base
pip install "llama-index-llms-blockrun[solana]"  # pay on Solana

Pick a chain

The key never leaves your machine; it only signs the payment for each call. llm.wallet_address prints the address to fund.

from llama_index.llms.blockrun import BlockRun

llm = BlockRun(model="openai/gpt-5.5")  # Base
llm = BlockRun(model="openai/gpt-5.5", chain="solana")  # Solana

print(llm.wallet_address)

Usage

from llama_index.core.llms import ChatMessage
from llama_index.llms.blockrun import BlockRun

llm = BlockRun(model="anthropic/claude-sonnet-4.6")

print(llm.complete("Paris is the capital of"))

response = llm.chat(
    [
        ChatMessage(role="system", content="Answer in one sentence."),
        ChatMessage(role="user", content="What is x402?"),
    ]
)
print(response.message.content)
print(response.additional_kwargs.get("cost_usd"))  # USD charged, where the SDK reports it (Base)

for chunk in llm.stream_complete("Write a haiku about USDC."):
    print(chunk.delta, end="", flush=True)

Async works the same way: achat, acomplete, astream_chat, astream_complete.

As the default LLM

from llama_index.core import Settings

Settings.llm = BlockRun(model="openai/gpt-5.5")

Tools and agents

BlockRun is a FunctionCallingLLM, so tool calling, predict_and_call, structured outputs and FunctionAgent work as they do with OpenAI:

from llama_index.core.agent.workflow import FunctionAgent


def multiply(a: float, b: float) -> float:
    """Multiply two numbers."""
    return a * b


agent = FunctionAgent(tools=[multiply], llm=BlockRun(model="openai/gpt-5.5"))
print(await agent.run("What is 1234 * 4567?"))

Spending

Each call is quoted by the gateway before anything is signed. Cap what a single call may cost, and a quote above it is refused before it is paid:

llm = BlockRun(model="openai/gpt-5.5", max_cost_per_call=0.05)

A refused quote raises blockrun_llm.SpendLimitError; an empty wallet raises blockrun_llm.PaymentError. Prices per model are listed at https://blockrun.ai/api/v1/models.

Options

Development

pip install -e "../..[dev]" -e ".[dev]"
pytest