Technology and AI

Fren combines several AI models depending on the task:

• Reasoning models (such as GPT-5, Claude) for complex questions that require cross-referenced analysis.
• More efficient models: depending on the type of internal task being performed, smaller models are used to maximise efficiency and reduce response time.
• Semantic search models (encoders or embedding models) to find relevant information in large corpora. These are models adapted to the specific language of parliamentary or legislative texts.

Fren offers two answer modes depending on the complexity of the question:

• Pro: deep analysis with intermediate reasoning (~1 min)
• Think: exhaustive research with verification across multiple sources (several minutes)

Choose the mode based on the type of question and how urgent it is.

RAG is the technique Fren uses to ensure reliable answers. Instead of generating answers solely from the model's trained "knowledge", Fren first searches for relevant information in its official databases and then generates the answer based on those documents. This avoids hallucinations and makes it possible to cite exact sources. In addition, Fren uses subagents, so if it does not find an answer to the user's question, it can adjust its search strategy.

Fren updates its data depending on the source:

• Daily: press releases, agendas, BOE, social media
• After each session: votes and interventions (same day)
• After each official change: the procedural status of initiatives

Fren automatically generates visualisations depending on the type of answer:

• SVG chamber view with the individual vote of each member
• Bar charts showing the breakdown by parliamentary group
• Timelines of how initiatives evolve
• Stakeholder networks with relationships and influence
• Comparison tables across parties, regions or periods

It depends on the mode and the complexity:

• Pro: 30-90 seconds
• Think: 2-5 minutes

Fren shows progress in real time with streaming, so you can read while it generates.

Yes, Fren uses response streaming: the text appears progressively as it is generated, without waiting for it to finish. In addition, in Pro and Think modes you can see the intermediate reasoning steps ("thoughts") to understand how it reaches the answer.

Fren works exclusively with official, real data obtained from public sources: the portals of the Congreso, Senado, BOE, ministries, regulatory bodies, etc. It does not use synthetic or made-up data.