Daena in one paragraph
Daena coordinates AI agents the way an operating system coordinates programs. It routes each task to a model, checks policy before an action runs, remembers context in tiers, and records what happened and why. The result is one control plane for a company’s AI, instead of a scatter of scripts and chatbots that nobody can audit.
What every request goes through
In governed mode a request passes ten stages in order. Daena also has lighter governance modes for cases where less oversight is acceptable. The governed path is the one built for business use.
- Security gate
- Session
- Intent and risk
- Governance check
- Cost preflight
- Model routing
- Memory recall
- Request build
- Streaming
- Persist and audit
What Daena includes
10 department agents, 6 capabilities each. Each of the 10 departments maps to a business function. The six capabilities are mind, eyes, hands, voice, shield and memory.
- Engineering
- Product
- Marketing
- Sales
- Finance
- Operations
- Research
- Legal and Compliance
- Skill Governance
- Security Operations
Three reasoning modes. Standard picks the best single model. Council sends the task to three or more models and synthesizes one answer. Quintessence pairs experts with models in a matrix for the hardest calls.
5 memory tiers. Unverified content expires. Permanent tiers need approval.
9 hard laws enforced in code, and 116 connectors in the connector catalog. Klyntar, the security layer, runs scans with evidence checkpoints and rejects serious findings that come without an evidence chain.
Where Daena stands today
Daena is in beta, live at daena.mas-ai.co. Source-available under BSL 1.1, with the code on GitHub. It is the reference architecture we draw on when we build governed agent systems for other companies. Its memory design is covered by US provisional patent application 64/020,421.
Read the full write-up in the Daena case study.
Why it exists
One chatbot does not need governance. A fleet of agents that call tools, spend money and touch production systems does. Daena is built for the moment a business goes from one assistant to many agents and has to answer two questions: who authorized this action, and can we prove it later.
Questions
What does governed mean in Daena?
In governed mode every request passes the same ten-stage pipeline: a security gate, intent and risk classification, a governance check, cost preflight, model routing, memory recall and, at the end, persistence and an audit record. High-risk actions wait in an approval queue for a person. Governance sits in the execution path, not beside it.
Which models does Daena use?
Daena routes across 10 model providers: Anthropic, OpenAI, Google Gemini, Perplexity, Groq, OpenRouter, Together, Qwen Cloud, Ollama (local), vLLM (local). Local models run on your own hardware through Ollama or llama.cpp. Whether a whole deployment stays local also depends on its connectors, logs and backups, which we check per deployment.
Is Daena hosted or self-hosted?
Both are possible. Daena is in beta, and an earlier build of it runs on Google Cloud Run. The source is public on GitHub under BSL 1.1, and it can use local models through Ollama or llama.cpp.
Who builds Daena?
MAS-AI Technologies Inc., a founder-led company in the Greater Toronto Area, Canada, founded by Masoud Masoori. Klyntar, the security-scan layer, is part of Daena. KYA Mission Control, a separate research project on agent identity, lives at kya.mas-ai.co.
How do I try Daena?
Open daena.mas-ai.co for the live beta. If you want a governed agent system built for your own business, use the Bring us a bottleneck form on mas-ai.co and describe what you want to fix.