runvault 0.1.1

Financial identity infrastructure for AI agents — Python SDK

RunVault Python SDK

Financial identity infrastructure for AI agents. The RunVault SDK is the standard integration layer between AI agents and the RunVault platform — it handles agent registration, JWT exchange, LLM client routing through the RunVault proxy, and run-level context propagation.

Install

The base package is small. LLM provider support is opt-in via extras:

pip install runvault[openai]                          # OpenAI raw client
pip install runvault[anthropic]                       # Anthropic raw client
pip install runvault[langgraph,langchain-openai]      # LangGraph + LangChain OpenAI
pip install runvault[all]                             # everything

Quick start

from runvault import RunVault, ChatOpenAI

rv = RunVault(
    api_key="rv_live_...",
    be_url="https://your-runvault-backend",
)

llm = ChatOpenAI(model="gpt-4o-mini")
graph = build_graph(llm)

agent = rv.init(
    framework="langgraph",
    app=graph,
    agent_id="research-v1",
    name="Research Agent",
    budget=1.0,
)

result = agent.invoke({"input": "What is..."})

rv.init() registers the agent with the backend (idempotent on agent_id), exchanges the API key for a short-lived JWT, and wraps your compiled LangGraph app so every LLM call inside it is routed through the RunVault proxy with the correct authentication and budget context.

Supported frameworks

• LangGraph — pass a compiled graph to rv.init(framework="langgraph", app=graph, …).

Supported LLM clients

Drop-in factories that return real upstream instances configured to route through the RunVault proxy:

• ChatOpenAI, OpenAI, AsyncOpenAI
• ChatAnthropic, Anthropic, AsyncAnthropic
• ChatGoogleGenerativeAI

Python version

Requires Python 3.10 or newer.

License

Apache 2.0. See LICENSE.