Clovie Get started
Quickstart
Connect your first agent, add a policy in monitor mode and find the audit record, in about fifteen minutes.
In this quickstart you connect one agent to Clovie, add a policy that only watches, and then find the record of what the policy decided. Nothing you do here blocks any traffic.
Before you begin
You need:
- A Clovie account with the Org admin or Admin role. Adding provider keys and installing policy packs needs one of these roles.
- An API key from a model provider you already use, such as OpenAI or Anthropic.
- An agent, script or notebook that calls that provider, or a terminal with
curl.
1. Add your provider key
Clovie calls the model provider with your own key, so add it first.
Go to AI Governance → Models → Provider credentials and select Add credential.
Choose the Provider, keep Stored encrypted, paste the key into API key and, if you like, add a Name such as "Production".
Save, then select Test. Continue once its status is Verified.
2. Enable a model
Agents can only use models your organization has enabled.
Go to AI Governance → Models → Catalog, find the model you want, for example gpt-4o, and select Enable.
3. Register your agent
Go to AI Governance → Agents → Registry and select Register agent.
Enter a Name, for example support-bot. This is the only required field. Choose the Environment and, if you have one, the Owner team.
Select Register agent. The Agent registered. Copy its key now dialog shows your agent's key.
Copy the key and store it where your agent reads its secrets. Clovie shows it only once.
Treat the agent key like a password
Anyone with the key can send requests as this agent. If a key leaks, open the agent and rotate or revoke the key.
4. Send a request through the gateway
Point your agent at the Clovie gateway and use the agent key instead of the provider key. If you use Clovie as a hosted service, the gateway URL is https://gateway.clovie.io. If your organization runs Clovie itself, ask your administrator for your gateway URL.
With the OpenAI Python SDK:
from openai import OpenAI
client = OpenAI(
base_url="https://gateway.clovie.io/v1",
api_key="YOUR_CLOVIE_AGENT_KEY",
)
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Say hello in five words."}],
)
print(response.choices[0].message.content)Or with curl:
curl https://gateway.clovie.io/v1/chat/completions \
-H "Authorization: Bearer YOUR_CLOVIE_AGENT_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Say hello in five words."}]}'You get a normal response from the model. In AI Governance → Agents → Registry, your agent's status changes from Waiting for first call to Active.
5. Add a policy in monitor mode
The quickest way to get a useful policy is to install one from the Policy Hub.
Go to AI Governance → Policies → Policy hub and open a pack, for example Data Privacy & PII Protection.
Select Install pack. On the first step, keep All policies in Monitor mode.
Choose All teams, review the summary and select Install.
The policies now appear in AI Governance → Policies → My policies with the mode Monitoring. In monitor mode a policy records what it would have decided, but never changes a request.
6. Find the audit record
Send another request from your agent, then go to AI Governance → Activity → Audit log.
Each row is one decision: the time, the agent, the policy and its version, the stage that made the decision, the decision itself and the reason. A request that a policy in monitor mode would have blocked shows as Would deny, so you can see what would have happened without anything being blocked.
Select a row to open the full run in AI Governance → Activity → Agent runs, where you can see the request, the response and every policy that looked at it.
Next steps
- Policies explains monitor and enforce, and how to promote a policy once you trust it.
- Budgets and price books shows how to cap what an agent can spend.
- Connect an agent covers the other request formats the gateway accepts, streaming and the Clovie SDK.