Product
Fenja is not a single product. It is a platform for running advanced AI inside the customer's own infrastructure. It solves many use cases because it provides the building blocks every regulated AI deployment needs: a local model, agents that do work, a layer that teaches the AI the organisation's language, and the tools those agents use to act.
One deployment, fully under your control, with no data ever touching external infrastructure.
Locally hosted, open-source foundation model (Llama 3, Mistral). No data ever leaves the customer's infrastructure.
Specialised AI agents that handle distinct tasks — retrieval, summarisation, analysis, drafting — and collaborate to solve complex queries.
The business vocabulary through which the AI understands the organisation — its structure, terminology, decisions, and the people behind them. Grows automatically as the organisation uses the platform.
The capabilities agents use to act: read files, query systems, search the corpus, produce structured output, trigger processes.
The same sovereign infrastructure supports fundamentally different shapes of work. One deployment, one security model, one audit trail — but the use cases it enables expand as you add tiers.
{tier.desc}
The MVP in Project Bifrost is Fenja Basic: a locally hosted LLM with agent flows, plus Fenja Knowledge — the semantic layer that lets the AI understand not just your documents, but the context behind them.
This is the foundation. It solves real work today. And every tier that follows — Code, Analyze, Agentic — is built on it without requiring a new deployment, new infrastructure, or new compliance review.