Skip to content

sdk-python.md

Latest commit

 

History

History
86 lines (66 loc) · 2.77 KB

sdk-python.md

File metadata and controls

86 lines (66 loc) · 2.77 KB
Raw

How to Call Grob from Python

Send requests to a local Grob proxy using the official anthropic and openai Python SDKs. The proxy listens on http://[::1]:13456 (IPv6 localhost) by default and accepts both Anthropic /v1/messages and OpenAI /v1/chat/completions traffic.

Install

Use a Python environment where you can install packages. Configure Grob first with the getting-started tutorial.

python3 -m pip install anthropic openai

Run the script

Save one of the examples below as script.py. grob exec starts the proxy when needed, sets ANTHROPIC_BASE_URL and OPENAI_BASE_URL, and stops an instance it started when the script exits. An already-running instance stays running.

grob exec -- python3 script.py

The examples read those base URLs, so a custom host or port also works. For a separately managed proxy, set the matching base-URL variable yourself.

api_key authenticates your script to Grob; the provider credentials are managed separately by Grob. On an explicitly unauthenticated local proxy, the grob-local placeholder satisfies the SDK. If Grob requires authentication, set GROB_API_KEY to a Grob virtual key (or another credential accepted by your configured auth mode). grob exec does not set this key. See Authentication.

Both SDKs can request the same Grob model: choose one listed by grob model and set GROB_MODEL if it differs from the example's claude-sonnet-4-6. The request uses real provider capacity and may incur charges.

Anthropic SDK, non-streaming

import os
from anthropic import Anthropic

client = Anthropic(
    base_url=os.environ.get("ANTHROPIC_BASE_URL", "http://[::1]:13456"),
    api_key=os.environ.get("GROB_API_KEY", "grob-local"),
)

message = client.messages.create(
    model=os.environ.get("GROB_MODEL", "claude-sonnet-4-6"),
    max_tokens=256,
    messages=[{"role": "user", "content": "Summarize the Rust borrow checker in one sentence."}],
)
print(message.content[0].text)

OpenAI SDK, streaming

import os
from openai import OpenAI

client = OpenAI(
    base_url=os.environ.get("OPENAI_BASE_URL", "http://[::1]:13456/v1"),
    api_key=os.environ.get("GROB_API_KEY", "grob-local"),
)

stream = client.chat.completions.create(
    model=os.environ.get("GROB_MODEL", "claude-sonnet-4-6"),
    max_tokens=256,
    messages=[{"role": "user", "content": "Stream a haiku about caching."}],
    stream=True,
)
for chunk in stream:
    if not chunk.choices:
        continue
    delta = chunk.choices[0].delta.content
    if delta:
        print(delta, end="", flush=True)

See also