Sign in Publish

Open source

Open source

Contexte Tech is built entirely with open-source software. Thanks to their authors and communities.

The code of Contexte Tech's first version (v1) is free software, under the AGPL-3.0 licence: github.com/contexte-tech/contextetech

Mirrors : GitLab · Codeberg

★ Star on GitHub

For transparency, traffic statistics are public: stats.contextetech.com

Open data

Public open data sources

Contexte Tech's datasets on municipalities, regions and places are built from open data published by public bodies. The answers cite their source. Here are the data used, their licence and the dataset built from them.

Interface

Fast pages that search engines can read

Every resource page is rendered on the server by Astro and reaches the browser as plain HTML, with no JavaScript to wait for. That is what lets a resource be indexed properly and appear instantly, even on mobile. Only interactive parts, such as the publishing form, likes or the sandbox, are small Svelte islands loaded on demand.

  • AstroServer-rendered pages, plain HTML, fast and indexable.MIT
  • SvelteInteractive islands: forms, likes, sandbox.MIT
  • Instrument SansMain typeface.OFL-1.1
  • JetBrains MonoCode typeface.OFL-1.1
  • FontsourceFonts hosted by the site itself.MIT

API

A simple, documented API

Behind the site, a Python API built with FastAPI handles resources, accounts, comments and moderation. Pydantic validates every incoming piece of data, SQLAlchemy and Alembic evolve the database safely, and the API documentation is generated automatically, so any developer can plug the catalog into their own tools.

  • PythonAPI language.PSF
  • FastAPIAsync API and its documentation.MIT
  • UvicornASGI server for the API.BSD-3
  • PydanticData validation.MIT
  • SQLAlchemyDatabase access.MIT
  • AlembicSchema migrations.MIT
  • asyncpgAsync PostgreSQL driver.Apache-2.0
  • HTTPXHTTP calls to models.BSD-3

Data

Solid data and a fast cache

Resources are stored in PostgreSQL, a proven database that also handles search and tags. Redis keeps sessions, caches the most viewed lists and limits abuse such as bursts of posts or comments. Large datasets go to S3-compatible object storage and are downloaded through temporary links.

  • PostgreSQLDatabase.PostgreSQL
  • RedisSessions, cache and rate limiting.RSALv2 / SSPL
  • boto3Object storage for large datasets.Apache-2.0

Server

Hosted in France

The site runs in Docker containers behind Caddy, which handles HTTPS and HTTP/3 automatically, on a Linux server. Server and database are hosted in France by Scaleway, which makes GDPR compliance simpler for members and for companies using the catalog.

  • CaddyAutomatic HTTPS, HTTP/3, compression.Apache-2.0
  • DockerContainers for the API, site and cache.Apache-2.0
  • Oracle LinuxServer OS.GPL
  • BunnyCDN and DNS: files served close to visitors.

AI

Built for AI models

The sandbox works with Anthropic's Claude models or with open models run locally through Ollama or vLLM. Tools follow the Model Context Protocol, the open standard that lets one tool server plug into Claude, Cursor and other assistants. Exports target open fine-tuning formats: Axolotl, PEFT and JSONL datasets.

  • Anthropic SDKSandbox with Claude models.MIT
  • Model Context ProtocolStandard for MCP tool servers.MIT
  • OllamaLocal models for the sandbox.MIT

Security and privacy

Security and privacy

Passwords are hashed, sessions use protected cookies and every write checks that it really comes from the site. Fonts are served by the site itself: no browsing data is sent to any external service to display a page.

Hosting

Hosting

Server and database hosted in France by Scaleway, CDN and DNS by Bunny. Fonts are served by the site itself, with no external service.

Python

The contexte package

Load any resource from the catalogue in one line of Python: a dataset ready for training, a prompt to fill, a few-shot context ready for a chat API. No dependencies, MIT license.

Read the introductionView on PyPI

$ pip install contexte
import contexte

ds = contexte.load("contextetech/anonymisation-rgpd-fr")
p = contexte.load("jr6t0n")   # code court : fin de l'adresse de la fiche
$ contexte get jr6t0n