> ## 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.

# AI platforms

> Govern the Claude and OpenAI platforms your teams already use, and see who uses them.

Govern the AI platforms your teams already use, without changing how they use them. Each integration lists what the platform holds in [Discovery](https://docs.xenovia.io/discovery/overview.md), and Claude's two integrations also check prompts and tool calls before Claude acts.

For example, connect Claude Enterprise and add its inference hook. When someone asks Claude to do something a policy forbids, Claude is told no, with a reason, and the decision shows in **Sessions** under the **Claude chat** agent.

| Integration | Discovery | Checks |
| - | - | - |
| [Claude Platform](#claude-platform) | The Managed Agents in a Claude Console workspace | Each tool call that waits for approval |
| [Claude Enterprise](#claude-enterprise) | Who uses Claude chat, Claude Code and Cowork | Each prompt and tool call, through Claude's inference hook |
| [OpenAI Platform](#openai-platform) | Every API key in your OpenAI organization | None: it shows which keys skip Xenovia |

Start one from first-run setup with **Protect your agent**, or in **Settings › Integrations** under **AI apps and platforms**. You need to be an admin of your Xenovia organization. Each integration keeps its keys and signing secrets encrypted and never shows them again. Choose **Sync now** on the integration page to sync right away.

## Claude Platform

![Claude Platform](https://mintcdn.com/xenovia/RO5UF1Tk8L9B7yPt/images/logos/claude.svg?fit=max\&auto=format\&n=RO5UF1Tk8L9B7yPt\&q=85\&s=5bfb22ba63d5baa0ed7b5aa8434fae5a)

The Claude Platform integration connects one Claude Console workspace. It does two things:

* **Discovery.** Xenovia lists every Managed Agent in the workspace, with its model, MCP servers and each risky tool's permission policy. Each one shows up in [Discovery](https://docs.xenovia.io/discovery/overview.md).
* **Checks.** A Managed Agent tool set to `always_ask` pauses the session before it runs. Claude Console's webhook tells Xenovia, and Xenovia answers each paused tool call with allow, or deny with the reason.

Xenovia only checks the tool calls that ask first. A tool that runs without asking is never checked, and Discovery reports it as a finding.

### Before you start

* In Claude Console, create an API key in the workspace your Managed Agents run in.

### Connect Claude Platform

**1. Open the integration**

In **Settings › Integrations**, find **Claude Platform** under **AI apps and platforms** and choose **Connect**.

**2. Paste the key**

Paste the key into **Claude API key** and choose **Connect**. Xenovia checks that the key can list the workspace's Managed Agents. If the check fails, Xenovia shows why and saves nothing.

**3. Add Xenovia's webhook**

The **Send prompts to Xenovia** step shows the webhook address for this connection. In Claude Console, open the workspace's webhooks and add an endpoint with that address for the `session.status_idled` event. Claude Console shows a signing secret that starts with `whsec_`. Paste it into **Webhook signing secret**: in the dialog, or later with **Settings** on the integration page.

**4. Ask before risky tools run**

In each Managed Agent, set the tools Xenovia should check to `always_ask`.

### The agents Xenovia creates

Connecting creates a Xenovia agent called **Claude Managed Agents**. After each listing, every Managed Agent also gets a Xenovia agent of its own, named after it, which its paused tool calls are checked as. An entry you linked to another agent keeps that agent.

The agents stay drafts until the signing secret is set, and go live once it is. Set each agent's [Neon](https://docs.xenovia.io/control/neon.md) protections and policies on its page, as for any agent. Each checked tool call is recorded in [Sessions](https://docs.xenovia.io/observability/sessions.md).

### What Xenovia reads

| What | How often |
| - | - |
| The workspace's Managed Agents: model, MCP servers and each tool's permission policy | Every 15 minutes |
| Sessions waiting for approval, so a call whose webhook was missed is still answered | Every 5 minutes |

Xenovia only reads listings. It never changes an agent or a session; it only answers the paused tool calls.

A Managed Agent's page in Discovery shows how many of its tools Xenovia checks, its MCP servers, and each tool's policy: **asks first**, **runs without asking**, or **Anthropic decides**. Discovery reports a finding when:

* risky tools run without asking, so Xenovia never checks them,
* MCP tools of a server run without asking,
* the agent can run in an environment that can reach any website.

To fix the first two, open the entry and choose **Plan a fix**, or set those tools to `always_ask` in Claude Console.

Choose **Sync now** on the integration page to sync right away. **Recent syncs** shows how the last ones went.

### How a paused call is answered

When a session pauses, Xenovia reads its paused tool calls with the workspace key, checks each as the Managed Agent's Xenovia agent, and answers each one. The webhook delivery is acknowledged at once; the checks run after.

A call is denied when a check finds a problem and the agent enforces. If Xenovia can't check a call, it denies it when the agent enforces, because the session would otherwise wait for ever, and allows it when the agent only monitors.

### Disconnect

Choose **Disconnect** on the integration page. Xenovia stops syncing, deletes the stored key and signing secret, and turns the connection's agents off. Open findings from Claude Platform are resolved, and Managed Agents only it reported leave Discovery. Your reviews are kept.

Remove Xenovia's webhook from Claude Console too, and revoke the API key. Paused sessions are no longer answered by Xenovia.

> [accordion] Troubleshooting
>
> | Message | What to do |
> | - | - |
> | Claude rejected the API key | Create a new key in the workspace your Managed Agents run in |
> | Could not reach the Claude API | Try again with **Check again** |
> | Claude is rate limiting the API; try again shortly | Wait, then choose **Check again** |
> | The signing secret starts with whsec\_ | Paste the webhook's signing secret, not the webhook address or the API key |
> | No signing secret yet: tool calls are not checked | Add Xenovia's webhook and paste its signing secret |

## Claude Enterprise

![Claude Enterprise](https://mintcdn.com/xenovia/RO5UF1Tk8L9B7yPt/images/logos/claude.svg?fit=max\&auto=format\&n=RO5UF1Tk8L9B7yPt\&q=85\&s=5bfb22ba63d5baa0ed7b5aa8434fae5a)

The Claude Enterprise integration does two things:

* **Discovery.** Through Claude's Compliance API, Xenovia lists the people in your Claude organization and which Claude products each uses: Claude chat, Claude Code and Cowork. Each product shows up in [Discovery](https://docs.xenovia.io/discovery/overview.md) with the people who use it.
* **Checks.** Through Claude's inference hook, Claude asks Xenovia about each prompt and each tool call before it acts. Xenovia allows it, or denies it with a reason.

You can connect for discovery only, and add the inference hook later.

### Before you start

* In Claude Enterprise settings, create a **Compliance access key** with read access to the Compliance API. Xenovia never asks for delete access.

### Connect Claude Enterprise

**1. Open the integration**

In **Settings › Integrations**, find **Claude Enterprise** under **AI apps and platforms** and choose **Connect**.

**2. Paste the key**

Paste the key into **Compliance access key** and choose **Connect**. Xenovia checks the key against the Compliance API. If the check fails, Xenovia shows why and saves nothing.

**3. Add the inference hook**

The **Send prompts to Xenovia** step shows the hook address for this connection. In Claude's admin settings, add an inference hook with that address. Claude shows a signing secret that starts with `whsec_`. Paste it into **Inference hook signing secret**: in the dialog, or later with **Settings** on the integration page.

### The agents Xenovia creates

Connecting creates Xenovia agents that the hook's prompts and tool calls are checked as:

| Agent | Checks |
| - | - |
| **Claude chat** | Claude chat |
| **Claude Code (Enterprise)** | Claude Code |
| **Claude Cowork** | Cowork |
| **Claude Enterprise** | Anything else Claude sends through the hook |

Each product's Discovery entry is governed by its own agent. The agents stay drafts until the signing secret is set, and go live once it is. Set each agent's [Neon](https://docs.xenovia.io/control/neon.md) protections and policies on its page, as for any agent. Each checked prompt and tool call is recorded in [Sessions](https://docs.xenovia.io/observability/sessions.md) under its agent.

If you disconnect and connect again, the **Claude Enterprise** agent comes back with its history.

### What Xenovia reads

| What | How often |
| - | - |
| Your Claude organizations and their users | Every 6 hours |
| Sessions and chats, as which person used which product | Every 30 minutes |

The first sync reads the last 30 days. Later syncs continue from where the last one stopped. Xenovia only calls listing endpoints. It never reads the contents of messages or transcripts, and never deletes anything.

Choose **Sync now** on the integration page to sync right away.

### How the hook answers

* For a **prompt**, Xenovia checks the latest user message, with its attachments' text.
* For a **tool call**, Xenovia checks each tool call the model wants to make.

A prompt or tool call is denied only when a check finds a problem and the agent enforces. In monitor mode it is recorded and allowed. A denial carries a reason, and the trace id as its reference so you can find it in Sessions.

Xenovia answers within four seconds. The hook fails open: if Xenovia can't decide in time, can't find the agent, or hits any other error, the answer is allow, and the error is recorded where there is a session to record it on. Until the signing secret is set, nothing is checked.

### Disconnect

Choose **Disconnect** on the integration page. Xenovia stops syncing, deletes the stored key and signing secret, and turns the connection's agents off. Open findings from Claude Enterprise are resolved, and entries only it reported leave Discovery. Your reviews are kept.

Remove the inference hook from Claude's admin settings too, and revoke the Compliance access key in Claude Enterprise.

> [accordion] Troubleshooting
>
> | Message | What to do |
> | - | - |
> | Claude rejected the Compliance access key | Create a new key with read access to the Compliance API |
> | Could not reach the Claude Compliance API | Try again with **Check again** |
> | Claude is rate limiting the Compliance API; try again shortly | Wait, then choose **Check again** |
> | The signing secret starts with whsec\_ | Paste the signing secret Claude shows for the hook, not the hook address or the key |
> | No signing secret yet: prompts and tool calls are not judged | Add the inference hook and paste its signing secret |
> | This integration has no agent: nothing is judged until one is set | The connection's agent was deleted. Disconnect and connect again to create it |

## OpenAI Platform

![OpenAI Platform](https://mintcdn.com/xenovia/RO5UF1Tk8L9B7yPt/images/logos/openai.svg?fit=max\&auto=format\&n=RO5UF1Tk8L9B7yPt\&q=85\&s=640765b3114b827bd22faf57c1422f70)

![OpenAI Platform](https://mintcdn.com/xenovia/RO5UF1Tk8L9B7yPt/images/logos/openai-dark.svg?fit=max\&auto=format\&n=RO5UF1Tk8L9B7yPt\&q=85\&s=61aca9a05c624f40270a7bdbba35ee3e)

The OpenAI Platform integration reads your OpenAI API organization with an admin key. Xenovia lists every project's API keys, with each key's owner, when it was last used and the past week's usage. Each key shows up in [Discovery](https://docs.xenovia.io/discovery/overview.md), and Xenovia tells you whether its calls go through Xenovia or straight to OpenAI.

Xenovia only reads. It never changes or deletes a key.

### Before you start

* In OpenAI's organization settings, open **Admin keys** and create one. Only organization owners can.

### Connect OpenAI Platform

**1. Open the integration**

In **Settings › Integrations**, find **OpenAI Platform** under **AI apps and platforms** and choose **Connect**.

**2. Paste the key**

Paste the admin key into **OpenAI admin key** and choose **Connect**. Xenovia checks that the key can list your projects. If the check fails, Xenovia shows why and saves nothing.

Xenovia keeps the key encrypted and never shows it again.

### What Xenovia reads

Every hour, Xenovia reads each active project's API keys: the key's name, its owner, when it was last used, and its requests, tokens and models over the past 7 days. It only calls listing endpoints.

Choose **Sync now** on the integration page to sync right away. **Recent syncs** shows how the last ones went.

### Through Xenovia, or direct

After each sync, Xenovia compares each key with the OpenAI [providers](https://docs.xenovia.io/providers/overview.md) in your Xenovia organization:

| The key | In Discovery |
| - | - |
| Is one of your OpenAI providers in Xenovia | **Through Xenovia: Yes**. The entry is governed by the agent that uses that provider, unless you linked it to another |
| Was used in the past 7 days and isn't one of your providers | **Through Xenovia: No, calls OpenAI directly**, and a warning: nothing it sends is checked |
| Hasn't been used in over 90 days | A low-severity finding: consider deleting it |

To bring a direct key under Xenovia, add it as an OpenAI provider in Xenovia and point the app that uses it at a Xenovia agent. See [Supported providers](https://docs.xenovia.io/providers/overview.md) and [Connect your app](https://docs.xenovia.io/agents/connect-your-app.md). After the next sync, the key shows **Through Xenovia: Yes**.

Each key's page links to the project's keys in OpenAI with **Open the project's keys**.

### Disconnect

Choose **Disconnect** on the integration page. Xenovia stops syncing and deletes the stored admin key. Open findings from OpenAI Platform are resolved, and keys only it reported leave Discovery. Your reviews are kept. Delete the admin key in OpenAI's organization settings too.

To use a different admin key, choose **Settings** on the integration page. Xenovia checks the new key before saving it.

> [accordion] Troubleshooting
>
> | Message | What to do |
> | - | - |
> | OpenAI rejected the admin key | Check that you pasted an admin key from **Admin keys**, or create a new one |
> | Could not reach the OpenAI API | Try again with **Check again** |
> | OpenAI is rate limiting the API; try again shortly | Wait, then choose **Check again** |

## What next?

> [card] [Neon](https://docs.xenovia.io/control/neon.md)
>
> Set what each platform agent enforces.

> [card] [Coding agents](https://docs.xenovia.io/agents/coding-agents.md)
>
> Govern Claude Code, Codex and Cursor through their hooks.

> [card] [Discovery](https://docs.xenovia.io/discovery/overview.md)
>
> Review what the platforms reported.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.