What an AI governance platform does
It controls what AI agents may see, decide and do before they execute, not only what they did afterward. When a business moves from one chatbot to a fleet of agents that call tools, spend money and touch production systems, the open question is: who decided this agent could take this action, and can you prove it later?
Daena answers that by putting governance inside the execution path. The check happens before the action, and the decision is recorded.
The ten-stage request pipeline
In governed mode every request passes these stages, in order:
- Security gate
- Session
- Intent and risk
- Governance check
- Cost preflight
- Model routing
- Memory recall
- Request build
- Streaming
- Persist and audit
The security gate screens the request. Intent and risk classify it. The governance check applies policy. Only then does routing pick a model and memory recall add context. The last stage persists the result and writes the audit record.
Risk, approval and modes
Each action gets a risk level. High-risk actions do not run on their own: they go to an approval queue and wait for a person. Action mode is plan-only by default.
Daena has three governance modes. Governed runs the full pipeline with all 9 hard laws. Balanced runs a lighter path and asks for approval only on dangerous operations. Unleashed skips the governance pipeline and keeps only data protection. The governed mode is the one meant for business use.
Governance that outlives your model choice
Policy and model choice are separate. The same governance check applies whichever provider handles the request.
- Anthropic
- OpenAI
- Google Gemini
- Perplexity
- Groq
- OpenRouter
- Together
- Qwen Cloud
- Ollama (local)
- vLLM (local)
Context persists in 5 memory tiers, where unverified content expires and permanent tiers need approval. Klyntar, Daena’s security layer, runs scans with evidence checkpoints and rejects serious findings that have no evidence chain.
Questions
What is the difference between AI governance and AI observability?
Observability tells you what an agent did after the fact. Governance decides what an agent may do before it acts. Daena is governance-first: in governed mode a request is checked and classified by risk before any action runs, and the decision is written to an audit trail.
How do you govern AI agents before they use tools?
Put the check in the execution path. In Daena a request goes through a security gate, intent and risk classification and a governance check before the model is routed and anything is built or executed. Plan-only is the default action mode, and execution goes through the security gate and an approval queue for high-risk actions.
Does an AI governance platform require human approval?
Only where it matters. Daena classifies each action by risk. High-risk actions wait in an approval queue for a person. Lower-risk actions do not need a human on every step.
Which governance platform supports multi-LLM routing?
Daena routes across 10 model providers: Anthropic, OpenAI, Google Gemini, Perplexity, Groq, OpenRouter, Together, Qwen Cloud, Ollama (local), vLLM (local). The routing stage sits after the governance check, so changing the model does not skip the check.
How does Daena compare with LangChain, AutoGen and CrewAI?
Those are frameworks for building agents. Daena is a governed platform for running them, with governance, memory and audit built into the request path. The two layers can be combined: build behavior with a framework, and run it under a control plane.
Is Daena tied to one model vendor?
No. Routing is separate from policy, and Daena supports 10 model providers, including local runtimes.
Is Daena a real product or a concept?
It is a working system in beta, and an earlier build runs on Google Cloud Run. The source is public under BSL 1.1 on GitHub, so you can read what it does.