ClovieEditions and deployment
What is Clovie?
Clovie is a governance platform for AI agents. It shows which agents your organization runs, decides what they are allowed to do, and attributes what they cost. Your agents send their model requests to the Clovie gateway with their own Clovie key. The gateway checks each request against your policies and budgets, forwards allowed requests to the model provider with your provider key, and records what happened.
Which editions are available?
You can use Clovie as a service we host (SaaS), or run it yourself: in your own AWS, Google Cloud or Azure account (VPC), in your own data center on Kubernetes or a single host (on-premises), or in a network with no internet access (air-gapped). Every edition has the same features. What changes is who operates it and where your data lives, and you can move from one to another later.
Does a self-hosted install need to call back to Clovie?
No. Agent traffic, audit records, cost data and provider credentials stay in your network. The licence is checked inside your install, with no connection to Clovie. In an air-gapped network, releases arrive as images you import into your own registry, and usage reports leave as files you carry out.
Where is our data stored on Clovie SaaS?
Clovie SaaS stores data in the United States, and every organization's data is kept separate from every other organization's. If your rules require another location or your own network, run Clovie yourself in your cloud account or data center.
Model providers and keys
Do we use our own model provider keys?
Yes, always. Your agents' requests are sent to model providers with your own provider keys, so you keep your existing accounts, contracts and data agreements, and each provider bills you directly. Clovie never uses its own keys for your agents' traffic. If an agent calls a provider you have not added a key for, the request is refused with a message that names the provider.
Do our agents ever see the provider keys?
No. Each agent calls the gateway with its own agent key, and only Clovie holds your provider keys. A leaked agent key cannot be used against your provider account directly, you can revoke one agent's key without touching any other, and you rotate a provider key in one place.
Which providers and models can our agents use?
The gateway accepts the Chat Completions, Messages and Responses request formats used by the major model providers, with streaming. Agents can also call models you host yourself, or models on Azure OpenAI, Amazon Bedrock and Google Vertex AI. Agents can only use the models your organization has enabled.
Data and privacy
Does Clovie store our prompts and responses?
Yes, so you can inspect each run. Sensitive data such as email addresses, phone numbers, payment card numbers, national identity numbers, cloud access keys and IP addresses is masked before request content is stored. Organization administrators decide how long data is kept. Audit records are always kept for at least one year.
Can we stop Clovie capturing what coding tools run?
Yes. Capture of coding-tool inputs and outputs is an organization-wide setting, on by default. When it is off, each tool check still records the tool, its status, its timing and the decision, but not the commands, file contents or paths.
Can policies remove sensitive data from agent traffic?
Yes. A policy can redact sensitive content from a request as it passes through the gateway, before it reaches the model.
Is our data used to train AI models?
Clovie asks the providers behind its own AI features not to train on your data, and this is on by default. Your agents' requests go to the providers you choose, under your own agreements with them.
Policies and control
Will a new policy break our agents?
Not by itself. Every policy starts in monitor mode: it runs on real traffic and records what it would have decided, but never changes a request. In the audit log, a request a monitored policy would have blocked shows as "Would deny", and it still went through. You cannot create a policy directly in enforce mode.
How does a policy move from monitor to enforce?
First you simulate it: Clovie replays at least the last 7 days of your own traffic through the policy and shows how many requests it would have allowed, denied or held for approval. Then an administrator or policy administrator promotes it. The simulation must be for the current version of the policy and less than 24 hours old. If someone later changes what an enforced policy checks or who it applies to, it goes back to monitor mode until it is simulated and promoted again.
What can a policy decide?
For each request it applies to, a policy can allow it, deny it, redact sensitive content from it, or hold it until a person approves or denies it. A policy can apply to the whole organization, to specific teams, to specific agents, or to agents that match conditions such as their environment.
What happens if Clovie cannot check a request?
Your organization decides whether traffic continues with an alert or is refused until checks are back. Kill switches apply either way.
Can we stop an agent immediately?
Yes. Turn on the agent's kill switch and its requests are refused at once, with no redeployment. There is also an organization-wide kill switch.
Agents and coding tools
Which agent frameworks work with Clovie?
Any OpenAI- or Anthropic-compatible SDK, and most agent frameworks built on them, such as LangChain and the OpenAI Agents SDK. You change two things: the base URL the agent sends requests to, and the key it sends. Prompts, tools and streaming stay the same. An optional Python SDK adds conveniences such as waiting for approvals.
Can we govern coding agents such as Claude Code, Codex and Cursor?
Yes, with the clovie-gov command-line tool, which also supports Claude Desktop, on macOS, Linux and Windows. Developers sign their coding agent in with your company's identity provider instead of pasting a key, and get a short-lived credential tied to them and bound to their machine. Before the coding agent runs a tool, such as a shell command or a file edit, the tool is checked against your policies, and what ran is recorded.
How do we find agents nobody registered?
Clovie shows AI traffic that does not come from a registered agent as shadow agents, so you can review it and bring it under governance. Discovery also finds AI agents and identities that already exist in your cloud accounts, AI platforms and SaaS tools. Discovery only reads: it never changes anything in your accounts unless you choose an action that says so.
Identity and access
Do you support single sign-on, SCIM and multi-factor authentication?
Yes. People sign in with your identity provider over SAML or OIDC, for example Okta, Microsoft Entra ID, Google Workspace or Ping, and you can require single sign-on for everyone. SCIM 2.0 provisioning creates people and groups automatically, and suspends people as soon as your identity provider deactivates them. Administrators can require multi-factor authentication for every user.
Can our identity provider groups decide roles?
Yes. You map identity provider groups to roles, and each person's role is updated at every sign-in and every SCIM change. When someone is in several mapped groups, the highest role wins. Clovie shows who would change role before anything is saved, always keeps at least one organization administrator, and records every change in the security audit log.
Which roles are there?
Org admin, Admin, Policy admin, Editor, Agent operator, Member and Viewer. The roles separate who can view, who can change agents and policies, and who can turn enforcement on, so the people who build agents are not the people who decide what is enforced.
How do we invite people?
Invite up to 50 people at a time with one role, and optionally a project and a team. Clovie emails each invitation, or gives you a single-use link to share when email is not set up. Invitations expire after 7 days.
Audit and compliance
What does Clovie record?
Every decision the gateway makes about an agent's request, with the agent, the policy version, the decision and the reason. Sign-ins, role changes and administrative changes go to a separate security audit log. Audit records cannot be edited or deleted from the product.
Can we send audit records to our SIEM?
Yes. Send AI requests, policy decisions and security audit events continuously to Splunk, Datadog, an OpenTelemetry collector, a webhook, or syslog in self-hosted installs. Prompts and responses are left out unless you choose to include them for a destination, and personal data is masked even then.
What evidence can we give our auditors?
A signed evidence pack for the period under audit, labelled with SOC 2 Type II, NIST AI RMF, the EU AI Act, OSFI E-23 or ISO/IEC 42001. It contains the policy decisions, approvals and policy versions in force, with checksums and signatures so an auditor can verify it has not been changed.
Pricing and usage
How much does Clovie cost?
Pricing depends on the edition and the scale of your deployment, so we do not publish a price list. Talk to sales for a quote, or compare the four editions.
How is self-hosted usage measured?
In governed transactions: agent requests Clovie governed through the gateway. Reports also count governed tool calls, registered agents with an owner that were active, developer seats, connectors and installed policy packs. How each unit is counted is set out in your licence agreement.
How do usage reports work without a network connection?
Each quarter, your install produces a signed report file that you review and then send to Clovie. It contains counts only, never prompts, responses, agent names or user names. Reports are numbered and each one refers to the one before it, so a missing or altered report is easy to spot.
What happens when a self-hosted licence expires?
A 14-day grace period starts, during which everything keeps working. After it, the product becomes read-only and the gateway refuses agent requests, but sign-in and the licence page keep working and nothing is deleted. Install a renewed licence and everything works again straight away.
Support and security
Are self-hosted images signed?
Yes. Every Clovie container image is signed when it is built and comes with a signed software bill of materials (SBOM) in CycloneDX format and signed build provenance. You can verify them with cosign before you run or mirror a release, including offline in an air-gapped network.
Can Clovie access our self-hosted install?
No. Clovie has no access to your install. When you ask for support we may ask you to run commands or share logs, but we never need your secrets or your data.
Can we get detailed security documentation?
Yes. Your account team can share detailed security documentation under NDA. Email hello@clovie.io.
How do we report a vulnerability?
Email hello@clovie.io. Please do not disclose it publicly until we have had a chance to fix it.