Data leakage
Keep sensitive data inside your network.
Prompts and tool results can carry API keys, credentials or customer records. Cardea detects them and blocks the call before it reaches the model provider.
coming soon Cardea is in early access. Details may change before launch.
Open source · Hosted in the EU
Cardea sits between your agents and the models they call. It identifies every agent, records every tool call, and stops risky calls before they leave your network.
Loading…
The Cardea platform, shown with sample data.
How it works
Agents send their model calls to Cardea Gateway instead of the provider. The gateway authorizes and identifies the agent, checks the call against your OPA policy rules, redacts personal data from the prompt, records a trace in Cardea Cloud, and forwards it only if it is allowed.
01 · Authorize
Only known callers get in
Agents point their base URL at the gateway. Calls without a valid credential are refused, and agent code stays the same.
02 · Identify
Know who is calling
Each call is tied to an agent, a session and a host, so every decision has a name on it.
03 · Policy rules
Decide with OPA
Your Rego rules decide which agent may call which model and tool, within which limits. A denied call never leaves your network.
04 · PII redaction
Strip personal data
Names, emails, phone numbers and IBANs are replaced with placeholders before the prompt reaches the provider.
05 · Traces
Record, then forward
Every decision is traced to Cardea Cloud, hosted in the EU, for audit. Allowed calls go on to Mistral AI, Anthropic, OpenRouter or another provider.
Shadow AI
Cardea builds the map of your agents automatically, from the calls it sees. You know which agents run, where, and for which team.
Use cases
Agents call models, run tools and handle data on their own. Cardea gives you visibility and control over each of these actions.
Data leakage
Prompts and tool results can carry API keys, credentials or customer records. Cardea detects them and blocks the call before it reaches the model provider.
Usage and cost
Cardea measures tokens and calls by agent, session and host. You can tell which agent, team or workflow drives your model spend.
Policies
Define which models and tools each agent may use, and set usage limits. Cardea enforces your policies on every call and refuses unidentified agents.
Audit
Every call is recorded with its agent, session, tools and decision. You have the evidence that audits and regulators ask for.
Session traces
Model calls and tool calls appear in order, on one timeline per session. When something goes wrong, you see which step caused it. Prompt and result content is never stored.
Adoption
Point your agents at Cardea instead of the model provider. There is no SDK to add and no agent code to change.
On your own infrastructure, or use our service hosted in EU data centres.
Each agent gets its own credential. Your provider keys stay with Cardea.
Claude Code, opencode and any Anthropic or OpenAI SDK work without further changes.
Open and sovereign
Governance should not mean handing your traffic to another third party. With Cardea, you choose where your data lives, and it stays there.
Pilot
One agent, one workflow, on your infrastructure. We are working with a small number of design partners in healthcare, legal, finance and software.
We also help with integration, policy design and rollout.