Skip to main content

Govern agent behavior without rebuilding your stack

Xenovia gives you a control layer for agent runtimes. Route model traffic through your agent’s OpenAI-compatible endpoint and enforce policy without rewriting your application architecture.

Quickstart

Route one request through Xenovia, confirm trace headers, and add your first policy.

Choose an integration

Pick the right path for OpenAI SDK, LangChain, LlamaIndex, Vercel AI SDK, Agents SDK, or the Xenovia SDK.

Policies and Enforcement

Write request and response policies in Rego, then decide when to block, redact, or escalate.

Trace and investigate

Track each call with session IDs, tool activity, policy outcomes, and remediation context.

Two ways to integrate

Agent endpoint

Point any OpenAI-compatible client at your agent’s endpoint. You keep your prompts, tools, and framework code, while Xenovia adds policy evaluation, provider routing, session tracking, and traces.

Python SDK

Gate tool calls, file writes, API requests, or database mutations that happen outside the model call itself. Use execute() for one-off checks or @guard() to wrap functions.

What Xenovia adds

  • Request-stage and response-stage Rego policy enforcement
  • Session-linked traces with turn counts and request headers for correlation
  • Central provider credentials and upstream routing
  • Tool visibility and recorded block and escalation signals
  • Guardrails, including a scope guardrail for what the agent is for
1

1. Pick your path

Start with Getting Started to decide between the agent endpoint and SDK hooks.
2

2. Prove the plumbing

Run the Quickstart and confirm a trace appears for a known request.
3

3. Add controls

Move to Policies and Enforcement and start with one rule that blocks or redacts something concrete.
4

4. Expand observability

Use Traces and Remediation to group sessions, add custom properties, and investigate drift.

What Xenovia records

  • Request and response bodies, token counts, latency, and TTFT
  • Session ID, session turn, and trace linkage
  • Tool names, arguments, and results when present
  • Policy and guardrail outcomes
  • Custom request metadata via X-Xenovia-Property-* headers