VaultKit Python SDK

Secure, policy-driven data access for AI agents and applications.

VaultKit is a control plane for governed data access. This SDK allows Python applications and AI agents to safely query data with built-in policy enforcement, approval workflows, and auditability.

---

Features

• Policy-enforced data access (masking, approval, deny)
• First-class support for AI agents (OpenAI, Anthropic)
• Built-in approval workflows
• Automatic polling and retries
• Schema-aware dataset discovery
• Simple, high-level API via execute()

---

Installation

pip install vaultkit

---

Quick Start

from vaultkit import VaultKitClient

client = VaultKitClient(
    base_url="http://localhost:3000",
    token="YOUR_TOKEN",
    org="YOUR_ORG",
)

result = client.execute(
    dataset="users",
    fields=["id", "email"],
    limit=10,
    purpose="Analyze user activity",
)

print(result.rows)

---

AI Agent Usage

VaultKit provides built-in tools for LLM agents.

from vaultkit.tools import ToolBuilder, ToolExecutor, ToolProvider

builder = ToolBuilder(client)

tools = builder.build(
    provider=ToolProvider.OPENAI,
    include_check_approval=True,
)

executor = ToolExecutor(client)

result = executor.execute(
    "vaultkit_query",
    {
        "dataset": "users",
        "limit": 5,
        "purpose": "Analyze user trends",
    },
)

See full example: examples/agent_openai_demo.py

---

Approval Flow

Some queries require human approval before data is returned.

from vaultkit.errors.exceptions import ApprovalRequiredError

try:
    client.execute(dataset="sensitive_data", purpose="Analysis")
except ApprovalRequiredError as e:
    print(f"Approval required. Request ID: {e.request_id}")

Once approved, resume with:

result = client.poll_request(request_id="req_123")

---

API Overview

High-Level

Method	Description
client.execute(...)	Full lifecycle: query → poll → fetch. Recommended for most use cases.

Low-Level

Method	Description
client.query(...)	Submit an intent request, get a QueryResult
client.poll(result)	Block until a queued result reaches a terminal state
client.fetch(grant_ref=...)	Redeem a grant for data
client.poll_request(request_id=...)	Poll by request ID (used in approval flows)

Discovery

Method	Description
client.datasets()	List authorized datasets from the registry
client.schema("users")	Get field-level schema for a dataset

---

How It Works

Client → VaultKit → Policy Engine → Data Source
                         ↓
                  Enforced Policies

1. Queries are evaluated against policy bundles at runtime
2. Sensitive fields may be masked based on requester context
3. Some datasets require human approval before access is granted
4. All access is logged and auditable

---

Why VaultKit?

Traditional access control is static — permissions are set upfront and rarely change. VaultKit enables:

• Runtime, policy-driven access — decisions made at query time based on context
• AI-safe data access — purpose and clearance are first-class query parameters
• Auditability and compliance — every request is tracked with correlation IDs

---

Environment Variables

export VAULTKIT_URL=http://localhost:3000
export VAULTKIT_TOKEN=your_token
export VAULTKIT_ORG=your_org

Or use a .env file (see .env.example).

---

Local Development

Start VaultKit locally with Docker:

docker compose up

Run the test suite:

pytest

---

License

MIT