up-to-date graphs for
accurate answers
> irm https://contaixt.ai/setup | iexConfigures Claude Code, Cursor and VS Code if they are installed. Client by client
| Name | Scope | References | Notions | Sourced links | Topics | Last update |
|---|---|---|---|---|---|---|
| AI compliance: the AI Act, GDPR applied to AI, Swiss law, standards and audits, copyright | 2 368 | 32 429 | 86 147 | 16 | 15 Oct 2026 | |
Main sourcesReference texts (AI Act, GDPR, Swiss FADP, Council of Europe convention, NIST AI RMF), article by article (927) Example question (as asked, in French)The graph's answer (13 Oct 2026, verbatim) | ||||||
| AI ethics: fairness, governance, safety, regulation, privacy | 697 | 41 632 | 97 146 | 18 | 8 Oct 2026 | |
Main sourcesJust Security (256) Example question (as asked, in French)The graph's answer (13 Oct 2026, verbatim) | ||||||
| Systems security and privacy: adversarial attacks and defences, extraction and poisoning, reverse engineering, anonymity and counter-surveillance | 1 174 | 41 084 | 90 264 | 14 | 8 Oct 2026 | |
Main sourcesSchneier on Security (572) Example question (as asked, in French)The graph's answer (benchmark of 7 Oct 2026, verbatim) | ||||||
| AI engineering: agents, memory, knowledge graphs, models, evaluation | 6 718 | 60 889 | 183 912 | 24 | 8 Oct 2026 | |
Main sourcesarXiv — papers cited by the corpus and topic watch (5 491) Example question (as asked, in French)The graph's answer (benchmark of 7 Oct 2026, verbatim) | ||||||
| Image, rendering and 3D visualisation: image generation and analysis, render engines, techniques, architecture and design | 1 297 | 30 619 | 73 997 | 20 | 8 Oct 2026 | |
Main sourcesarXiv — papers cited by the corpus and topic watch (784) Example question (as asked, in French)The graph's answer (benchmark of 8 Oct 2026, verbatim) | ||||||
| Audio and AI: music and voice, audio representations, generation, speech recognition, similarity, analysis | 382 | 9 186 | 23 505 | 9 | 7 Oct 2026 | |
Main sourcesMusic Ally (London) (919) Example question (as asked, in French)The graph's answer (benchmark of 7 Oct 2026, verbatim) | ||||||
| A critical look at trading bots and crypto investment strategies | 674 | 16 939 | 36 499 | 18 | 6 Oct 2026 | |
Main sourcesQuantpedia (297) Example question (as asked, in French)The graph's answer (13 Oct 2026, verbatim) | ||||||
Click a graph for its main sources and an example question.
We see the graph as a medium of a new kind, in step with today's practices and tools: immaterial and fluid, it adapts to its context and gives access to specialised, sourced, current knowledge — a kind of “liquid knowledge” that plugs into the LLM you already use.
The graphs are queried from the assistant you already use. You need not be an expert to ask the question, nor to check where the answer comes from: every sentence points to its source.
No source enters without being measured, no change ships without being benchmarked against real questions, and a change that lowers the score is corrected before it stays. The processes evolve and criticise themselves; the graphs are maintained regularly, so that what they say is what the field says today.
Its purpose is to spread knowledge and raise the general level of understanding. What experts know need not stay among experts.
Two questions from our benchmarks, put to the same model without and then with the graph. The answers are reproduced verbatim, in French, as they were given.
Quel montant atteint l'amende prononcée en août 2026 contre Uber pour ses décisions individuelles automatisées concernant les chauffeurs, et par quelle autorité ?
No answer.
824 990 000 € (environ 825 M€), amende infligée par l'autorité néerlandaise de protection des données (Autoriteit Persoonsgegevens) le 21 août 2026, en lien avec la CNIL.
Source: EDPB, « Dutch DPA fines Uber EUR 824 990 000 » ; sota.io (2026-08-25)
Dans un balayage d'environ 100 millions de backtests sur BTC/USDT, combien de stratégies franchissent toutes les étapes de filtrage (nombre de trades, Deflated Sharpe au-dessus du seuil de chance, walk-forward) alors que 336 818 passaient le seul test PSR > 0,95 ?
No answer.
3 130 stratégies (0,003 %) sur 99 878 688 backtests passent le filtrage complet : au moins 30 trades, au-dessus du seuil de chance et positives hors échantillon.
Source: we ran 100 million bitcoin backtests only 3 130 beat the luck bar (11.07.2026)
A graph can weigh tens of megabytes and cannot be handed to an agent as is. contaixt gives the agent a condensed version of the graph: a map, and leads to follow.
The vocabulary, the central notions, the communities and the texts that carry authority, in a few pages.
Look up a notion, see its neighbourhood, connect two ideas. Deterministic calls, with no model in between.
Every link carries its provenance. The agent names its source instead of guessing, and says when it does not know.
> irm https://contaixt.ai/setup | iexIn PowerShell on Windows, in a terminal on macOS and Linux. The script configures Claude Code, Cursor and VS Code if present, nothing else. Read the script
claude mcp add --scope user --transport http contaixt https://contaixt.ai/mcp
Then, in a session: “ask the ai-quant graph about grid bots”.
{
"mcpServers": {
"contaixt": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://contaixt.ai/mcp"
]
}
}
}In claude_desktop_config.json (Settings → Developer → Edit Config), then restart Claude Desktop. Needs Node.js: mcp-remote relays the server.
{
"mcpServers": {
"contaixt": {
"url": "https://contaixt.ai/mcp"
}
}
}In ~/.cursor/mcp.json.
{
"servers": {
"contaixt": {
"type": "http",
"url": "https://contaixt.ai/mcp"
}
}
}In .vscode/mcp.json for a project, or in the user configuration (“MCP: Open User Configuration”).
Address https://contaixt.ai/mcp Transport HTTP (Streamable HTTP)
For any MCP client that accepts an HTTP server: Windsurf, Gemini CLI, Codex, Zed, JetBrains…
All take the graph's name. Start with graphes, then carte.
grapheswhat each graph covers: topics, volume, periodrepondrecall it first: sourced syntheses, sentence by sentence, and the best excerpts, in one calldescendrethe paragraph of the source file cited by a reference in repondrecartethe map of a graph: vocabulary, central notions, communities, authoritieschercherthe nodes whose label matches, from the most to the least connectedexpliquera node's record: type, relations, neighbours, sources that attest itvoisinagethe typed relations around a node, with file and linecheminhow two notions connect, relation by relationsourcesthe source files that attest a notionfichewhat one source file brings to the graphsommetsthe most connected nodes, by typerelationsthe graph's relation vocabulary and countslire_fichethe text of a cited source file, numbered by line