langchain-zerogpu

LangChain tools for ZeroGPU.

ZeroGPU is a compute-efficient inference provider for apps and agents. We run purpose-built small and nano language models across an edge network for the high-volume tasks you run constantly — classification, extraction, moderation, routing, summarization — at ~10x lower latency and 50%+ lower cost than frontier-model workflows. Auto-scaling, with zero GPU infrastructure to manage. Plug in and you're live.

This package exposes those models as first-class LangChain BaseTool subclasses, so any LangChain agent — including create_agent and LangGraph graphs — can offload these repeatable NLP tasks (classification, summarization, moderation, routing decisions, signal / entity / JSON extraction, PII redaction, and chat) to ZeroGPU instead of spending frontier-model tokens — plus larger open-weight models for the calls that do need reasoning headroom.

All calls go through the official zerogpu-api Python SDK.

Install

pip install langchain-zerogpu

Authenticate

Every request needs a ZeroGPU API key (starts with zgpu-api-). A project id is optional — set one to scope requests to a specific project. Provide them via environment variables:

export ZEROGPU_API_KEY="zgpu-api-..."
export ZEROGPU_PROJECT_ID="your-project-id"   # optional

…or pass them directly to any tool or the toolkit:

from langchain_zerogpu import ZeroGPUSummarizeTool

tool = ZeroGPUSummarizeTool(api_key="zgpu-api-...", project_id="your-project-id")

The API key is stored as a pydantic.SecretStr and is never logged.

The tools

Quick start

from langchain_zerogpu import ZeroGPUClassifyZeroShotTool

tool = ZeroGPUClassifyZeroShotTool()  # reads creds from the environment

print(tool.invoke({
    "text": "The new GPU smashes every benchmark we threw at it.",
    "labels": ["tech", "politics", "sports"],
}))

Tools work asynchronously too:

result = await tool.ainvoke({"text": "...", "labels": ["a", "b"]})

Bind the tools to an agent

Use the toolkit to get all 26 tools — wired to a single shared client — and bind them to an agent:

from langchain.agents import create_agent
from langchain_zerogpu import ZeroGPUToolkit

toolkit = ZeroGPUToolkit()  # reads ZEROGPU_API_KEY / ZEROGPU_PROJECT_ID
tools = toolkit.get_tools()

agent = create_agent("anthropic:claude-sonnet-4-6", tools=tools)

agent.invoke({
    "messages": [
        {"role": "user", "content": "Redact the PII in: 'Call Jane at 555-0100.'"}
    ]
})

Or bind a single tool to a chat model directly:

from langchain.chat_models import init_chat_model
from langchain_zerogpu import ZeroGPUExtractPIITool

llm = init_chat_model("anthropic:claude-sonnet-4-6")
llm_with_tools = llm.bind_tools([ZeroGPUExtractPIITool()])

Errors

Failures surface as clear, typed exceptions instead of raw stack traces:

• ZeroGPUAuthError — missing / malformed credentials, 401, or 403.
• ZeroGPUError — rate limits (429), server errors (5xx), and network failures.

Development

make install            # uv sync --all-groups
make lint               # ruff check + format --check
make mypy               # mypy (disallow_untyped_defs)
make test               # unit tests, sockets disabled
make integration_test   # integration tests (needs real ZeroGPU creds)
make bump               # bump the version; merging to main releases it

Releases are automatic — see RELEASING.md.

License

MIT — see LICENSE.