> ## Documentation Index
> Fetch the complete documentation index at: https://docs.xenovia.io/llms.txt
> Use this file to discover all available pages before exploring further.
# Integrations Overview
> Choose the right Xenovia integration for your SDK, framework, or OpenAI-compatible runtime.
If your stack can speak the OpenAI API, Xenovia usually fits with a `base_url` change. Start there unless you need to govern actions that happen outside the model call itself.
## Recommended order
1. Start with the framework your production workload already uses.
2. If you are unsure, use the [OpenAI SDK](/integrations/openai-sdk) guide because it maps cleanly to most OpenAI-compatible clients.
3. Add the [Xenovia Python SDK](/integrations/xenovia-sdk) when you need policy checks around downstream actions.
## Proxy mode integrations
Xenovia Runtime is an OpenAI-compatible proxy. Any client that accepts a custom `base_url` works without further changes, and every call passes through Xenovia's policy and trace pipeline.
```text theme={null}
https://runtime.xenovia.io/a/{proxy_id}/openai/v1
```
Provider credentials are configured in the proxy and resolved server-side.
Python and Node.js. Best default starting point for any OpenAI-compatible client.
Python and JS chains, agents, and RAG pipelines.
RAG pipelines and agentic query engines.
Multi-agent orchestration with the Responses API.
Next.js and edge streaming.
Multi-agent frameworks that can sit on top of an OpenAI-compatible endpoint.
## SDK mode
The [Xenovia Python SDK](/integrations/xenovia-sdk) (`pip install xenovia-sdk`) gates arbitrary agent actions such as tool calls, database writes, API requests, and file operations without proxying an LLM call.
```python theme={null}
from xenovia_sdk import Xenovia
xenovia = Xenovia(api_key="xe_...", identity_id="billing-agent")
@xenovia.guard(capability="payments.transfer")
def transfer_funds(payload: dict) -> dict:
return run_transfer(payload["amount"], payload["to"])
result = transfer_funds({"amount": 500, "to": "acct_123"})
```
Full SDK reference for `execute()`, `@guard()`, session handling, and error behavior.
## Choose the right guide
| | Proxy mode | SDK mode |
| ------------------------ | ----------------------------------------------------- | ------------------------------------------------------- |
| **What is governed** | Every LLM call through the proxy | Arbitrary agent actions such as tools, writes, and APIs |
| **Code change required** | `base_url` swap only | `execute()` call or `@guard()` decorator |
| **Policy input** | Full LLM request context | Capability string plus payload |
| **Trace contains** | LLM request and response, tokens, latency, tool calls | Capability, payload, decision, and session |
| **Session tracking** | Automatic via Xenovia session resolution | Explicit `session_id` or `auto_session=True` |
Use both together: proxy mode for LLM governance, SDK mode for downstream tool governance.
## If your framework is not listed
* If it accepts an OpenAI-compatible `base_url`, start from the [OpenAI SDK](/integrations/openai-sdk) guide.
* If it only needs HTTP access, point it at `https://runtime.xenovia.io/a/{proxy_id}/openai/v1`.
* If the framework makes risky local calls after the model response, add the [Xenovia Python SDK](/integrations/xenovia-sdk) alongside it.
## Supported proxy endpoints
| Endpoint | Use case |
| --------------------------- | ------------------------------------- |
| `POST /v1/chat/completions` | Chat, agents, tool calling |
| `POST /v1/responses` | OpenAI Agents SDK and multi-step runs |
| `POST /v1/embeddings` | RAG and vector search |
| `POST /v1/completions` | Legacy text completions |