Watch
The open-source building blocks we track, assess — and, for some, run in production. License explained, maturity, trend: what a decision-maker needs to know before betting on them.
High-performance LLM inference server (continuous batching, PagedAttention).
Apache 2.0: free for commercial use in production, no strings attached.
Sovereign alternative to: OpenAI API (inférence)
Run open-weight LLMs locally, in a single command.
MIT: one of the most permissive licenses, fully free for commercial use.
LLM inference in C/C++, from server to laptop, no GPU required.
MIT: fully free for commercial use.
A unified gateway to 100+ LLM providers behind a single API.
MIT: free for commercial use.
European open-weight language models — the model layer of our sovereign deployments.
“Open-weight”: weights are downloadable (often Apache 2.0), but not always open source in the OSI sense. Check model by model.
Sovereign alternative to: OpenAI / Anthropic (modèles)
Multilingual speech-to-text, runnable locally — used in our healthcare deployments.
MIT: fully free for commercial use, including locally on sensitive data.
Sovereign alternative to: API de transcription cloud
Vector database for semantic search and RAG.
Apache 2.0: free for commercial use.
Sovereign alternative to: Pinecone (base vectorielle SaaS)
Vector search directly inside PostgreSQL — our default RAG foundation.
PostgreSQL License (BSD/MIT-style): free for commercial use.
Sovereign alternative to: Pinecone (si vous avez déjà PostgreSQL)
Self-hostable workflow and integration automation — our automation engine.
“Fair-code” (Sustainable Use License): free for internal use, but NOT OSI open source — reselling it as a SaaS is restricted.
Sovereign alternative to: Zapier / Make
Observability and tracing for LLM applications (cost, latency, quality).
MIT core; some enterprise features are under a separate commercial license.
Personal AI agent wired into your messaging apps, running on your own machine or server.
MIT: free for commercial use. GitHub reports “Other” because the file points to third-party notices (THIRD_PARTY_NOTICES.md) — the licence text itself is standard MIT.
Sovereign alternative to: ChatGPT, Copilot (assistant personnel)
Our position
Deployable in a company under conditions: an agent wired into messaging reads content written by third parties and acts on your behalf. Governed by an independent foundation since February 2026.
See the hardening frameworkContinuous-memory agent runtime from Nous Research: the agent learns from how you use it over time.
MIT: free for commercial use in production, no strings attached.
Sovereign alternative to: ChatGPT (agent personnel et mémoire)
Our position
Deployable in a company under conditions, stricter than most: persistent memory is the whole point of the product and where the GDPR risk concentrates — a memory you cannot search is a memory you cannot purge.
See the hardening frameworkSelf-service, self-hostable business intelligence.
AGPL: free for internal use; but embedding it in a distributed product forces you to open-source your own code. Commercial enterprise edition available.
Sovereign alternative to: Tableau / Power BI (BI)
Versioned, tested SQL data transformation.
Apache 2.0: free for commercial use (the core; the Cloud offering is separate).
Data ingestion connectors (ELT), several hundred sources.
Mixed licensing (including Elastic License v2): free to use, but you may not offer it as a competing managed service.
Sovereign alternative to: Fivetran
The reference relational database — the foundation of our whole stack.
PostgreSQL License (BSD-style): among the most permissive there is.
Self-hosted deployment platform (sovereign PaaS) — what deploys our apps.
Apache 2.0: free for commercial use.
Sovereign alternative to: Vercel / Heroku (PaaS)
Identity and access management (SSO, OIDC, SAML).
Apache 2.0: free for commercial use.
Sovereign alternative to: Auth0 / Okta
Stars and trends updated weekly by our automated tracking.