axio-transport-anthropic

Anthropic Claude transport for axio.

Streams Claude responses over the Anthropic Messages API using aiohttp and SSE parsing. Supports direct API, Vertex AI, prompt caching, extended thinking, and automatic retry on rate-limit and overload errors.

Features

• All Claude models — Opus, Sonnet, Haiku; configurable via ANTHROPIC_MODELS
• Vertex AI — use Claude via Google Cloud with ADC authentication
• Prompt caching — cache_control: ephemeral applied automatically to the system prompt and the last tool definition
• Extended thinking — ReasoningDelta events emitted for thinking blocks
• Retry logic — automatic backoff on 429 / 529; honours Retry-After header
• TUI integration — settings screen for API key and model selection

Installation

pip install axio-transport-anthropic

With the TUI settings screen:

pip install "axio-transport-anthropic[tui]"

With Vertex AI support:

pip install "axio-transport-anthropic[vertexai]"

Usage

import asyncio
import aiohttp
from axio import Agent
from axio.context import MemoryContextStore
from axio_transport_anthropic import AnthropicTransport, ANTHROPIC_MODELS


async def main() -> None:
    async with aiohttp.ClientSession() as session:
        transport = AnthropicTransport(
            api_key="sk-ant-...",
            model=ANTHROPIC_MODELS["claude-sonnet-4-6"],
            session=session,
        )
        agent = Agent(system="You are helpful.", tools=[], transport=transport)
        ctx = MemoryContextStore()
        print(await agent.run("Hello!", ctx))


asyncio.run(main())

Set the API key via environment variable instead of passing it directly:

export ANTHROPIC_API_KEY="sk-ant-..."

Vertex AI

transport = AnthropicTransport(
    vertexai=True,
    project="my-gcp-project",
    location="us-east5",
    session=session,
)
# Uses Application Default Credentials (gcloud auth application-default login)

Models

Configuration

License

MIT

Vertex entry point

Install axio-transport-anthropic[vertexai] to use Claude through Vertex without the Google transport package. Select --transport anthropic-vertex in REPL, or Anthropic Vertex AI in TUI. Set GOOGLE_CLOUD_PROJECT and optionally GOOGLE_CLOUD_LOCATION, or enter project and location in TUI settings. axio_transport_anthropic.VertexAITransport also accepts these fields directly. This entry point requires explicit selection and uses Google ADC.

When migrating from Google Vertex, remove anthropic/ from the model ID and copy project and location. Existing Google settings remain unchanged.