Use the envclaw Python package to connect AI agents to services through Keystore without exposing secrets.

Python SDK

The envclaw Python package lets your AI agents access provider APIs through the Keystore proxy. Agents receive a ks_ token instead of real credentials -- the vault resolves real keys at request time.

Installation

bash

pip install envclaw

Quick Start

python

from envclaw import Keystore

ks = Keystore(agent_token="ks_abc123...")

The Keystore instance is the entry point for all SDK functionality. Pass your agent token (starts with ks_) to the constructor.

Wrapping Provider Clients

OpenAI

Use wrap_openai to route OpenAI requests through the Keystore proxy. The agent token replaces the API key, and the base URL is rewritten to point at the proxy.

python

from openai import OpenAI
from envclaw import Keystore, wrap_openai

ks = Keystore(agent_token="ks_abc123...")
client = wrap_openai(OpenAI(), ks)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Anthropic

Use wrap_anthropic for Anthropic's Claude SDK.

python

from anthropic import Anthropic
from envclaw import Keystore, wrap_anthropic

ks = Keystore(agent_token="ks_abc123...")
client = wrap_anthropic(Anthropic(), ks)

message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)
print(message.content[0].text)

Intercepting All Requests

intercept_all patches Python's HTTP layer to route all matching provider requests through the Keystore proxy. This works with any SDK or HTTP client that hits a supported provider domain.

python

from envclaw import Keystore, intercept_all

ks = Keystore(agent_token="ks_abc123...")
intercept_all(ks)

# All requests to OpenAI, Anthropic, etc. now go through the proxy.
from openai import OpenAI
client = OpenAI(api_key="unused")
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
)

You can limit interception to specific providers:

python

intercept_all(ks, providers=["openai", "anthropic"])

Setting Up Environment Variables

setup_env writes provider-specific environment variables so that frameworks and SDKs that read from os.environ work automatically.

python

from envclaw import Keystore, setup_env

ks = Keystore(agent_token="ks_abc123...")
setup_env(ks, providers=["openai", "anthropic"])

# Environment variables are now set:
#   OPENAI_BASE_URL     -> proxy URL
#   OPENAI_API_KEY      -> ks_abc123...
#   ANTHROPIC_BASE_URL  -> proxy URL
#   ANTHROPIC_API_KEY   -> ks_abc123...

This is particularly useful for frameworks like LangChain or CrewAI that initialize clients from environment variables.

Supported Providers

Provider	Wrapper Function	Env Vars Set by `setup_env`
OpenAI	`wrap_openai`	`OPENAI_BASE_URL`, `OPENAI_API_KEY`
Anthropic	`wrap_anthropic`	`ANTHROPIC_BASE_URL`, `ANTHROPIC_API_KEY`
Neon	--	`DATABASE_URL`
Resend	--	`RESEND_BASE_URL`, `RESEND_API_KEY`
Vercel	--	`VERCEL_API_URL`, `VERCEL_TOKEN`
S3	--	`AWS_ENDPOINT_URL_S3`

Error Handling

If an invalid or expired agent token is provided, the proxy returns a 401 error. Handle this in your application:

python

from openai import OpenAI, AuthenticationError
from envclaw import Keystore, wrap_openai

ks = Keystore(agent_token="ks_abc123...")
client = wrap_openai(OpenAI(), ks)

try:
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": "Hello"}],
    )
except AuthenticationError:
    print("Agent token is invalid or expired")