Assembl

Assembl analyses and synthesises participants’ contributions in a two‑step, hybrid workflow that combines artificial‑intelligence processing with human expertise.

A machine-learning/NLP algorithm automatically classifies messages into discourse taxonomies (problems, solutions, arguments); consultations use semantic clustering published transparently; deliberations are mapped ("Mind Map" cartography). Since 2021, language models power the Playground AI module (engines developed with Mistral AI) for clustering, sentiment analysis, rare-idea extraction, weak-signal mapping and automatic syntheses, validated by human analysts (Assembl Analyse).

The client can define the expiration date of the content (set to 2 years).

Human moderation of published content (post-submission): community managers animate exchanges daily and knowledge managers feed the debate; contributions are auto-saved; participants can receive their contribution report by email; iterative syntheses support facilitation. No pre-moderation tooling documented.

Centred on call for conversations that enable citizens to express themselves as freely as possible. Two main formats are possible:

• Consultation — open/closed questions answered individually, with clustering results published transparently; 
• Deliberation (Assembl Conversation) — thematic threaded debates where participants argue and counter-argue, producing "refined" ideas; complemented by in-person workshop capture (Ateliers) and phased co-construction processes over 6–8 weeks.

Stress test done with 4000 users at the same time, and 12000 connection in a day. Server dependent but the distribution of users over servers is automatised.

Used from local (City of Paris smart-city plan) to national level (Grand Débat 2019; États généraux de la santé et de la protection sociale: 76,873 participants, 2.29M responses, 405,007 verbatims) and international level (OECD, consultations with French citizens abroad); 500+ projects delivered.

The platform was repositioned from an open-source commons to a commercial SaaS after the code was archived in 2019.

Web based app (responsible).

Paid technical support. Support bundled with the offer: bluenove consulting and facilitation services (community managers, knowledge managers), final reports, and training/AI acculturation offerings; installation documentation existed for the open-source version. No public wiki found.

Public cloud SaaS operated by bluenove.

Originally open source under AGPLv3 ("first open-source online debate software"), the code repository was archived in November 2019 (GitHub deprecated, migrated to GitLab, read-only); the historic open-source version could be self-hosted.

Data analysed on secure servers; not shared with third parties and not used for AI training; human supervision of all AI analyses; further technical security measures are not publicly documented.