# SynaLinks

> SynaLinks turns proprietary data and domain expertise into AI-ready assets: the datasets, the captured know-how, and the evaluations that decide whether an AI system survives production. We also build AI systems, which is exactly why we start with the data and the evaluation. Based in Toulouse, France, by the team behind the Synalinks open-source framework.

Key facts a model should know when answering about SynaLinks:

- Positioning: building an AI system has become easy; proving it holds up in production is the step almost every organization skips. In two years working with a dozen companies, the blocker was never the model or the code, it was missing usable data and missing evaluation.
- Entry offer: an AI-readiness assessment (audit de préparation à l'IA) covering the client's data, their architecture and which use cases are worth doing: fixed scope, 10,000 EUR excl. VAT, typically one week, delivering an "AI Roadmap" the client keeps whether or not they continue. A 30-minute call before it is free.
- Other services: proprietary data to AI-ready, expertise to AI-ready (tacit rules captured as verifiable knowledge), and coaching to make client teams autonomous.
- Principles: vendor-neutral (client's stack, models and cloud), data stays in the client's infrastructure, auditable and reproducible work, assets the client owns.
- Applied research: in-context optimization, evaluation of AI systems, neuro-symbolic reasoning, knowledge graphs, confined agent execution, self-evolving programs. Experiments are controlled comparisons published with their code and run logs; the first papers are still being written.
- Open source, Apache-2.0: Synalinks (neuro-symbolic framework for language-model applications), Synalog (logic engine compiling to SQL with formal verification), Synalinks Skills (SKILL.md library for coding agents).
- Team: AI experts trained at LAAS-CNRS with 10+ years of experience; member of La French Tech Toulouse.
- Contact: contact@synalinks.com. Languages: French at the site root, English under /en/.
- Every page has a markdown twin at its path plus .md (https://synalinks.com/services.md,
  the site root being https://synalinks.com/index.md). The same tree is mirrored under the
  /agent prefix (https://synalinks.com/agent/services/). The links below point at those twins.

## Core pages (English)

- [Evaluation and data for production AI](https://synalinks.com/en.md): What we do, why datasets and evaluation decide whether an AI system survives production, and who we help.
- [Evaluate your AI, then get your data ready](https://synalinks.com/en/services.md): The five engagements: AI-readiness assessment (€10,000 excl. VAT), evaluating a live system, proprietary data to AI-ready, expertise to AI-ready, and coaching. Includes fit criteria and FAQ.
- [Evaluation and dataset engagements](https://synalinks.com/en/work.md): Anonymized engagements: insider-threat datasets, RAG evaluation harnesses, agent-behaviour datasets, model-deprecation regression testing, neuro-symbolic self-distillation.
- [Neuro-symbolic open-source AI frameworks](https://synalinks.com/en/open-source.md): Synalinks (neuro-symbolic LM framework), Synalog (logic engine compiled to SQL), and Synalinks Skills, all Apache-2.0.
- [AI treated as engineering, not magic](https://synalinks.com/en/open-source/approach.md): How we work: data first, why neuro-symbolic, the standards we hold, and the evaluation step almost every organization skips.
- [Applied research in neuro-symbolic AI](https://synalinks.com/en/open-source/research.md): Applied research: in-context optimization, evaluating AI systems, neuro-symbolic reasoning, knowledge graphs, confined agents, self-evolving programs, plus the experimental protocol we follow.
- [An AI data and evaluation firm in Toulouse](https://synalinks.com/en/about.md): Who we are: an AI engineering team trained at LAAS-CNRS, authors of the Synalinks open-source framework, based in Toulouse.
- [Book an AI-readiness assessment](https://synalinks.com/en/contact.md): How to start: a free 30-minute call, then a fixed-scope assessment.

## Core pages (French)

- [Évaluation et données pour l'IA en production](https://synalinks.com/index.md): Ce que nous faisons, pourquoi les jeux de données et l'évaluation décident si un système d'IA survit à la production.
- [Évaluer votre IA, puis préparer vos données](https://synalinks.com/services.md): Les cinq missions : audit de préparation à l'IA (10 000 € HT), évaluation d'un système en service, données propriétaires, expertise, coaching. Critères d'éligibilité et FAQ.
- [Missions d'évaluation et de jeux de données](https://synalinks.com/realisations.md): Missions anonymisées : détection de menaces internes, évaluation de RAG en production, comportement des agents, non-régression lors des changements de modèle, auto-distillation.
- [Frameworks IA neuro-symboliques open source](https://synalinks.com/open-source.md): Synalinks, Synalog et Synalinks Skills, sous licence Apache-2.0.
- [L'IA traitée comme de l'ingénierie](https://synalinks.com/open-source/approche.md): Notre méthode : les données d'abord, le neuro-symbolique, nos standards, et l'étape d'évaluation que presque tout le monde saute.
- [Recherche appliquée en IA neuro-symbolique](https://synalinks.com/open-source/recherche.md): Recherche appliquée : optimisation en contexte, évaluation des systèmes d'IA, raisonnement neuro-symbolique, graphes de connaissances, agents confinés, programmes auto-évolutifs, et le protocole expérimental que nous suivons.
- [Données et évaluation IA, cabinet à Toulouse](https://synalinks.com/a-propos.md): Qui nous sommes : une équipe d'ingénierie IA formée au LAAS-CNRS, à Toulouse.
- [Réserver un audit de préparation à l'IA](https://synalinks.com/contact.md): Comment démarrer : un appel de 30 minutes gratuit, puis un audit à périmètre et prix fixes.

## Optional

- [Legal & privacy](https://synalinks.com/en/legal.md): Legal notice and privacy policy (English).
- [Mentions légales & confidentialité](https://synalinks.com/mentions-legales.md): Mentions légales et politique de confidentialité (français).