What a control plane is, and why agents need one
In networking, the control plane decides where traffic goes while the data plane moves it. Applied to AI, the control plane is where you decide which model runs a task, whether an action is allowed, what the agent remembers and how it is logged. Without one, every team makes those decisions again on its own and governance drifts.
What Daena centralizes
Four things in one layer. Routing across 10 model providers. Governance through the ten-stage request pipeline and an approval queue for high-risk actions. Memory in 5 memory tiers. Audit through a record written for each request. Agent code stays simple because policy, memory and routing live in the plane.
- Anthropic
- OpenAI
- Google Gemini
- Perplexity
- Groq
- OpenRouter
- Together
- Qwen Cloud
- Ollama (local)
- vLLM (local)
Separating model choice from policy
In governed mode, policy is enforced before the model is chosen, so switching the model does not skip it. The same governance check applies whether a task ends up on a hosted model or on a local one. That is what makes a multi-vendor setup manageable for a business.
Questions
What is an AI control plane?
The term comes from networking: the control plane decides and enforces policy while the work runs underneath. An AI control plane centralizes model routing, governance, memory and audit for a company’s agents, and keeps those concerns out of each agent’s own code.
Why do businesses need one?
Once more than one team ships AI agents, you get inconsistent policy, scattered logs and vendor lock-in. A control plane gives one place to enforce governance, route across models and produce an audit trail, so oversight does not depend on every team remembering to add it.
How is a control plane different from an agent framework?
A framework helps developers build agent behavior. A control plane governs and operates those agents in production. They fit together: build with a framework, run and audit through the control plane.
Does a control plane lock me into one model?
Not Daena. It separates model choice from policy and routes across 10 model providers, including local runtimes, so you can change providers without rewriting workflows.
Is Daena ready for production use?
Daena is in beta. An earlier build runs on Google Cloud Run, its product page is daena.mas-ai.co, and the source is public under BSL 1.1. Treat it as a reference architecture and a working beta, and ask us how it would fit your setup.