# André Wolke > Four decades on one question: how do you make a system you don't fully control behave in a way you can trust? Sovereign agents, LLM governance, confidential compute. Zürich, Switzerland. Writing code since 1986, in business since 1996. ## Identity - https://github.com/CandooLabs - https://github.com/48Nauts-Operator - https://www.linkedin.com/in/andrewolke - https://x.com/andrewolke ## Building - [xNAUT](https://andrewolke.me/building/xnaut/) — A native terminal for running a fleet of coding agents — and a record of what they actually did. Hash-linked, arguments sealed under an HSM-held key, batches signed on the HSM, and an offline verifier a customer runs without you. 20 MB, macOS and Windows. - [Engram](https://andrewolke.me/building/engram/) — The memory layer for AI agents. Self-hosted and tool-agnostic: one brain that outlives whichever CLI is hot this quarter. Open-source core, with an enterprise edition for organisations. - [NautGate](https://andrewolke.me/building/nautgate/) — A memory-aware LLM gateway. Route Claude Code, Codex, Pi or your own app to any model — and get a provable record of which model really served each call, what it cost, and what left your machine. - [NautLoop](https://andrewolke.me/building/nautloop/) — Sovereign agents in sandboxed execution. Your data never leaves the boundary you drew. Swiss-hosted, privacy by architecture rather than by promise. - [The Council](https://andrewolke.me/building/council/) — A multi-agent deliberation engine. Point it at a decision and a panel of models argues it to consensus — with the dissent kept on the record — before you commit. - [NautPay](https://andrewolke.me/building/nautpay/) — Conversational stablecoin payments on Solana — type 'send 50 USDC to alice.sol' and compliance happens invisibly. Every transaction screened by NautLense in real-time. No forms, no friction, no middleman. Banking-grade compliance with crypto-native UX. - [NautLense](https://andrewolke.me/building/nautlense/) — The compliance engine for digital assets. Banking-grade sanctions screening, multi-jurisdiction Travel Rule, cryptographic wallet verification and emergency lockdown, with no middleman. Live and in production use. - [Vynl](https://andrewolke.me/building/vynl/) — Your own private Spotify — built from scratch. Self-hosted music streaming with Beets at the core, AI-powered playlists, karaoke with synced lyrics, and an experimental AI DJ that mixes sets for your party. After years of trying every library system out there, Beets came closest to perfect — so I built everything else around it. ## Writing - 2026-10-05 [AI in the kitchen: from the first token to the bill](https://andrewolke.me/blog/ai-kitchen/) — What happens after you press send? Follow a request through an animated kitchen, with tokens, memory, processors and an illustrative bill. Narrated in German with my AI voice. - 2026-07-16 ["We use a compliant model" means nothing](https://andrewolke.me/blog/ai-compliance-for-kmu/) — There is no such thing as a compliant model — only compliant use. For a Swiss or German SME putting AI into real work, the question is never 'which model is allowed' but 'what are we doing with which data, and can we prove it.' Here's the shape that actually holds up. - 2026-06-25 [I built a terminal for the agent era](https://andrewolke.me/blog/xnaut-terminal-agent-era/) — My workflow stopped being 'me in a terminal' and became 'me directing a crew of agents.' None of my tools were built for that. So I built one: xNAUT — a 20 MB native app that runs your agents, isolates their work, keeps the notes, and never phones home. Free, MIT, local-first. - 2026-06-04 [Sovereign sandboxes for AI agents](https://andrewolke.me/blog/nautloop-sovereign-agents/) — An agent that can run code needs somewhere to run it. Most platforms give you 'somewhere' and stay quiet about where. NautLoop makes where the whole point: disposable microVMs you can pin to the EU or Switzerland, running local models, on a control plane that holds none of your data. - 2026-05-14 [One gateway for every LLM call](https://andrewolke.me/blog/nautgate-llm-gateway/) — Every project I run points its LLM calls at the same OpenAI-compatible URL: a gateway I host. One place to see the cost, swap the model, fall back to local, and keep the receipts. This is NautGate, and why you probably want one too. - 2026-04-22 [Take your private search anywhere with Tailscale](https://andrewolke.me/blog/private-search-anywhere-tailscale/) — Your SearXNG box lives on your LAN. Put it on a Tailscale tailnet and it — and your agent's private search — follows you to any café, hotel, or airport, without opening a single port. - 2026-03-25 [When Your AI Agent Team Builds a Platform in 48 Hours](https://andrewolke.me/blog/ai-agent-platform-48h/) — The honest story of shipping SecondBrain's knowledge graph from concept to production. What the agents got right, what broke at 3am, and why we'd do it again. - 2026-03-18 [Give your AI agent a search engine you own](https://andrewolke.me/blog/searxng-ai-agents/) — The follow-up how-to: self-host SearXNG on your NAS, bridge it to Claude Code and Codex over MCP, and switch the built-in search off. Ten minutes to take the query stream back. - 2026-02-12 [Your AI agent's search isn't private](https://andrewolke.me/blog/private-search-for-ai-agents/) — Every time your coding agent researches something, the query leaves your machine and lands in two logs you don't control. Here's exactly what happens behind the scenes — and how to take search back. - 2025-10-16 [Nautilus: Confidential AI-as-a-Service for Regulated Industries](https://andrewolke.me/blog/nautilus-bluepaper/) — Technical bluepaper for Nautilus — the world's first AI-native confidential computing platform. Hardware-guaranteed privacy for healthcare, finance, legal, and government AI workloads. $140B TAM. ## Interactive learning - [AI in the kitchen](https://andrewolke.me/blog/ai-kitchen/) — Free animated explanation of tokens, inference, memory and illustrative token costs. German AI narration by André, with a transcript. ## Jargon Definitions with a stated point of view. Quote freely, cite the anchor. - [Confidential computing](https://andrewolke.me/jargon/#confidential-computing) — Running computation inside a hardware-enforced enclave so that even the operator of the machine cannot read the data being processed. Attestation lets a third party verify which code is running inside it. - [Agent](https://andrewolke.me/jargon/#agent) — A model given tools, a goal and the authority to act in a loop without a human approving each step, deciding for itself when it is finished. - [High-risk system](https://andrewolke.me/jargon/#high-risk-system) — An EU AI Act classification covering systems used in areas such as employment, credit scoring, education, critical infrastructure and law enforcement. It triggers obligations on risk management, data governance, logging, human oversight and conformity assessment. - [Sovereign AI](https://andrewolke.me/jargon/#sovereign-ai) — AI systems where the operator retains control over where the model runs, where the data goes, and who can inspect either. Often used for national or sector-level infrastructure, but it applies to any single organisation. - [Egress](https://andrewolke.me/jargon/#egress) — Network traffic leaving a system boundary. In a sandboxed agent context, controlling egress means deciding which outbound destinations the agent may reach at all. - [Attestation](https://andrewolke.me/jargon/#attestation) — A cryptographic statement, signed by hardware, describing exactly what code is running in an enclave. It lets a remote party verify an environment they do not control. - [Human in the loop](https://andrewolke.me/jargon/#human-in-the-loop) — A design where a person reviews or approves an automated system's output before it takes effect. Named explicitly in EU AI Act obligations around human oversight for high-risk systems. - [Audit trail](https://andrewolke.me/jargon/#audit-trail) — An append-only record of what a system did, on what input, under whose authority, and when. For AI systems this includes the prompt, the model, the response and the cost. ## Attribution If you quote or summarise this work, cite https://andrewolke.me.