agentvisa

AgentVisa Python SDK

The official Python SDK for the AgentVisa API.

Quick Start

uv add agentvisa

import agentvisa

# Initialize the SDK
agentvisa.init(api_key="your-api-key")

# Create a delegation (type defaults to "ephemeral")
delegation = agentvisa.create_delegation(
    end_user_identifier="user123",
    scopes=["read", "write"],
)
print(f"Credential: {delegation['credential'][:12]}... (agent_id={delegation['agent_id']})")

Installation

uv add agentvisa

Authentication

Get your API key from the AgentVisa Dashboard.

import agentvisa

agentvisa.init(api_key="av_1234567890abcdef")

Usage

Simple Interface (Recommended)

For most use cases, use the simple global interface:

import agentvisa

# Initialize once
agentvisa.init(api_key="your-api-key")

# Create delegations
delegation = agentvisa.create_delegation(
    end_user_identifier="user123",
    scopes=["read", "write", "admin"],
    expires_in=7200,  # 2 hours
    metadata={"description": "Test agent"},
)

Client Interface

For advanced use cases or multiple configurations:

from agentvisa import AgentVisaClient

# Create client instance
client = AgentVisaClient(api_key="your-api-key")

# Use the client
delegation = client.delegations.create(
    end_user_identifier="user123",
    scopes=["read", "write"],
    expires_in=3600
)

Integrating with LangChain

You can easily secure LangChain agents by creating a custom tool that wraps AgentVisa. This ensures that every time an agent needs to perform a privileged action, it acquires a fresh, scoped, short-lived credential tied to the end-user.

This example shows how to create a secure "email sending" tool.

1. Create the AgentVisa-powered Tool

The tool's _run method will call agentvisa.create_delegation to get a secure token just before it performs its action.

from langchain.tools import BaseTool
import agentvisa

# Initialize AgentVisa once in your application
agentvisa.init(api_key="your-api-key")

class SecureEmailTool(BaseTool):
    name = "send_email"
    description = "Use this tool to send an email."

    def _run(self, to: str, subject: str, body: str, user_id: str):
        """Sends an email securely using an AgentVisa token."""
        
        # 1. Get a short-lived, scoped credential from AgentVisa
        try:
            delegation = agentvisa.create_delegation(
                end_user_identifier=user_id,
                scopes=["send:email"]
            )
            token = delegation.get("credential")
            print(f"Successfully acquired AgentVisa for user '{user_id}' with scope 'send:email'")
        except Exception as e:
            return f"Error: Could not acquire AgentVisa. {e}"

        # 2. Use the token to call your internal, secure email API
        # Your internal API would verify this token before sending the email.
        print(f"Calling internal email service with token: {token[:15]}...")
        # response = requests.post(
        #     "https://internal-api.yourcompany.com/send-email",
        #     headers={"Authorization": f"Bearer {token}"},
        #     json={"to": to, "subject": subject, "body": body}
        # )
        
        # For this example, we'll simulate a successful call
        return "Email sent successfully."

    async def _arun(self, to: str, subject: str, body: str, user_id: str):
        # LangChain requires an async implementation for tools
        raise NotImplementedError("SecureEmailTool does not support async")

2. Add the Tool to your Agent

Now, you can provide this tool to your LangChain agent. When the agent decides to use the send_email tool, it will automatically generate a new, auditable AgentVisa.

from langchain.agents import initialize_agent, AgentType
from langchain.chat_models import ChatOpenAI

# Initialize your LLM
llm = ChatOpenAI(model_name="gpt-5-2025-08-07", temperature=0)

# Create an instance of your secure tool
tools = [SecureEmailTool()]

# Initialize the agent
agent = initialize_agent(
    tools,
    llm,
    agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
    verbose=True
)

# Run the agent
# The agent will correctly parse the user_id and pass it to the tool
agent.run("Please send an email to team@example.com with the subject 'Project Update' and body 'The new feature is live.' for user_id 'user_anne'.")

This pattern provides the perfect combination of LangChain's powerful reasoning capabilities with AgentVisa's secure, auditable credentialing system.

API Reference

Delegations

create_delegation(end_user_identifier, scopes, expires_in=3600, *, delegation_type="ephemeral", metadata=None, timeout=15)

Create a new agent delegation.

Parameters:

• end_user_identifier (str): Unique identifier for the end user
• scopes (List[str]): List of permission scopes
• expires_in (int): Expiration time in seconds (default: 3600)

Returns: Dict containing delegation details including agent_id, credential, and expires_at. For backward compatibility the dict also includes aliases: id mirrors agent_id, and token mirrors credential.

Example:

delegation = agentvisa.create_delegation(
    end_user_identifier="user123",
    scopes=["read", "write"],
    expires_in=7200
)

Error Handling

The SDK raises standard Python exceptions:

import agentvisa
from requests.exceptions import HTTPError

try:
    agentvisa.init(api_key="invalid-key")
    delegation = agentvisa.create_delegation("user123", ["read"])
except ValueError as e:
    print(f"Invalid input: {e}")
except HTTPError as e:
    print(f"API error: {e}")
except Exception as e:
    print(f"SDK not initialized: {e}")

Development

Setup

git clone https://github.com/your-org/agentvisa-python.git
cd agentvisa-python
make install

Check the code

Run all checks (format, lint, typecheck, test)

make check

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Make your changes
4. Run tests (make test)
5. Commit your changes (git commit -m 'Add amazing feature')
6. Push to the branch (git push origin feature/amazing-feature)
7. Open a Pull Request

Support

• Documentation
• Issues
• Discord Community

License

This project is licensed under the MIT License - see the LICENSE file for details.