# Pia — Your Private AI Assistant > A privacy-first, local-first AI assistant. Multiple providers, offline speech, end-to-end encryption. Your intelligence, your rules. This file contains the complete English content of https://pia-ai.de as structured Markdown, intended for large language models and other automated readers. The website itself is available in English, German and French; this file is English only. A short index of the same site is at https://pia-ai.de/llms.txt. - Product: Pia, a Windows desktop AI assistant with an optional self-hostable server - Vendor: neo42 GmbH, Germany - Licence and code: open source, https://github.com/Pia-Ai-dev/Pia.Wpf - Documentation: https://docs.pia-ai.de - Contact: kontakt@neo42.de - Last updated: 2026-08-27 ## Contents 1. [Home](#home) 2. [Features](#features) 3. [Pricing](#pricing) 4. [Download](#download) 5. [Pia for business](#pia-for-business) 6. [Security and compliance](#security-and-compliance) 7. [Self-hosting](#self-hosting) 8. [About](#about) 9. [Site-wide](#site-wide) --- ## Home Source: https://pia-ai.de/ ### Hero **Your AI, your device, your rules** # Your private AI assistant. *Only one hotkey away.* Use any AI model you want. Dictate with your voice — even offline. Your data stays on your device — or syncs to a server, ours or one you run yourself — end-to-end encrypted if you want it. Calls to action: "Download for Windows" (https://pia-ai.de/download.html), "Pia for business" (https://pia-ai.de/business/). Hero badges: Multi-provider · Offline speech · E2E encryption · Open source ### The three mechanics (animated hero) 1. **Voice input — created on your device.** Your voice *never leaves* your machine. 2. **Conversations — stored on your device.** Your chats *stay private* on your device. 3. **Encrypted sync — end-to-end encrypted.** With encryption on, *not readable* on the server. ### Trust markers Open source · Made in Germany · 100% offline capable · No app telemetry ### Beyond the hotkey: the heavy lifting Knowledge, meetings, agentic work, governance — the four pillars that make Pia more than a chat window. Each one runs where you decide: on your device, or on your server. **Knowledge over your own documents.** Drop files in, ask questions, get answers with citations from your own material. A private vault on your device — or shared knowledge bases on your server, fed by a REST API. **Meetings & live transcription.** Pia joins your Teams meetings, transcribes live, tells speakers apart, and writes the summary. Or transcribe the room around you — entirely on-device, behind a spoken consent gate. **Agentic work, under your control.** Every tool Pia can use is a permission you grant — and can revoke at any time. Put recurring questions on a timer and the answers arrive as chats. Or hand Pia a whole goal: it plans, you approve the plan, it executes step by step while you steer. **Governance & security.** Guardrails route risky requests to protected models. Groups carry policy. Audit logs notice anomalies. SSO, encryption at rest — and it all runs on a server you control. Call to action: "Explore all features" (https://pia-ai.de/features.html). ### The problem **Your AI assistant shouldn't spy on you.** Most AI tools send every conversation to servers you don't control, lock you into a single provider, and require a constant internet connection. Your data becomes their business model. **Your conversations stay on your device.** No telemetry in the app, no tracking, no profiling. Data leaves your device on two occasions: when you sync, and when you send a request to a cloud model. With end-to-end encryption on, what you sync is unreadable to us. **Switch AI providers without losing your history.** Connect OpenAI, Anthropic, Mistral, Azure, OpenRouter, or your own model. Your conversations, memories, and templates remain intact — no matter which provider you use. **Talk to Pia even without internet.** Voice input and text-to-speech run entirely on your device — Whisper or Parakeet for speech recognition, Piper for the voice. No cloud connection needed for speech. ### How it works — up and running in three steps 1. **Download & install.** Free, no account required. Installs in under a minute. 2. **Pick your AI.** Either use Pia Cloud models with no key at all — we run those on European providers — or connect your own account: OpenAI, Mistral, Azure, or any compatible provider. 3. **Press your hotkey.** Pia sits in your system tray. One hotkey and she is ready for use — optimize text or chat. ### Common questions — your intelligence, your rules **Is my data really private?** Yes. Pia is open source — you can audit every line of code. All data is stored locally on your device. Cloud sync is optional, and so is end-to-end encryption: switch it on and what you sync is unreadable to us. If you use a Pia Cloud model, your prompt passes through our server on its way to the AI provider — it is processed there, not stored. Bring your own provider and it does not pass through us at all. **Is it hard to set up?** Not at all. Download, install, and a guided setup wizard walks you through provider configuration. You'll be using Pia in under two minutes. **What if I want to switch away?** Your data is yours. Everything is stored locally in standard formats. Switch AI providers anytime without losing a single conversation, memory, or template. ### Closing **Try Pia free — no account required.** No account needed. No credit card. Just download and go. --- ## Features Source: https://pia-ai.de/features.html **Capabilities** # Every feature, built on privacy. Every feature in Pia is designed with privacy as the foundation — not added later. ### Foundations **Your data stays on your device — by default.** Conversations, memories, and data are stored locally. No telemetry. No tracking. Cloud sync is optional, always under your control, and can be end-to-end encrypted — switch it on and the content is unreadable to us. **Zero-knowledge sync.** Sync conversations across devices. The server — ours or yours — never sees them in plaintext. **Use any AI model. Switch anytime.** Connect OpenAI, Anthropic, Mistral, Azure, OpenRouter — or run local models via Ollama. Switch anytime without losing a single conversation. **Voice input that works offline.** Dictate and listen without an internet connection. Your voice never leaves the machine. **Pick up where you left off, on any machine.** Sessions, memories, todos, templates, providers, personas — seamlessly synchronized across every device. Optional, always under your control, and end-to-end encryptable at your choice. **Make it yours.** Build custom templates, organize knowledge with typed schemas, and design workflows that fit your actual working method. ### Two modes: one assistant, two workflows Improve an email in seconds or brainstorm with persistent memory — all with one hotkey. **Optimize.** Turn rough notes into polished emails, articles, or messages — and auto-type the result into any app. - Text templates — emails, articles, code prompts - Local voice input — Whisper or Parakeet, on-device - Auto-type results into any app - Side-by-side comparison view **Assistant.** Your conversational AI with persistent memory, inline action cards, and smart suggestions. - Structured memory with smart search and @ references - Kanban board with drag-and-drop, priorities, and voice input - Inline action cards — accept or decline without interruption - Smart suggestions for quick actions - @-mentions to reference memories, todos, and reminders inline ### The rest of the product: everything else in the box More capabilities that ship with Pia today — for one person on a laptop or a whole team on a server. **Knowledge & documents.** Drop files into the vault and Pia compiles them into recallable knowledge pages — you see what was ingested and can remove any source. Attach files to any chat. Memory lives as plain Markdown on disk: browsable, editable, yours. **Meetings & transcription.** Pia joins Teams meetings, transcribes live speaker by speaker, and writes a summary you can keep. Live-transcribe the room around you entirely on-device, behind a spoken consent gate. Or just talk — voice conversation mode takes turns and uses tools. **Flow & reminders.** One calm rail for everything that wants your attention — notifications, reminders, and background updates arrive as actionable cards, grouped by due date. Snooze, act, or dismiss. **Personas.** Reusable identity and voice profiles. Pick one per chat, and always see which persona wrote which answer. **Coding tools.** Point Pia at a folder you choose: it reads, searches, and edits text files inside it — and shows every write as a diff for review first. Reference files inline with @Files. **Plugins, MCP & REST.** Extend Pia with server-delivered tool packs, bring your own MCP server, or wire in a REST API as a tool source — enabled and disabled per user. **Agentic work.** Every tool is a permission: grant it once, until Pia closes, or always — and revoke any of it from a full inventory. Put recurring questions on a timer; answers arrive as chats. For bigger jobs, hand Pia the goal: it plans, you approve the plan, and it executes step by step while you watch and steer. Running Pia for a team? See Pia for business (https://pia-ai.de/business/). Closing: **Start using Pia in under two minutes.** Download for Windows, or view on GitHub. --- ## Pricing Source: https://pia-ai.de/pricing.html # Honest pricing. No surprises. The Windows app is always free. With your own provider key it runs with no server and no account at all — what follows is about the server behind it. ### On your server You run the server. Every edition is the same container image — the signed licence decides what it unlocks. #### Community Edition — self-hosted, 5 users, 1 admin **Free, forever.** - Local login with MFA and passkeys - AI proxy with your own provider keys - Sync and chat history across devices - End-to-end encryption, recovery and device revocation - Admin dashboard, audit log and token board Call to action: Request a licence (kontakt@neo42.de). #### Enterprise Edition — self-hosted, seats per contract **Custom, on request. Coming Soon — not available yet.** - Everything in Community, plus: - Microsoft Entra ID single sign-on - Your own groups: quotas, provider routing, plugin allowlists - Guardrails with a decision log - Knowledge bases with cited answers, managed personas, client policies ### On Pia Cloud We run the server — on European providers, in EU data centres, with no US providers. Turn on end-to-end encryption and what you sync stays unreadable to us. Pia Cloud plans are business offerings — prices are net, plus VAT. The Windows app stays free for everyone. #### Free — cloud-hosted **$0 / month.** - Cloud sync across devices — sessions, memories, todos, templates, providers, personas - Basic access to Pia Cloud models — requests run through our server - Token-metered usage with hourly, daily & weekly limits - One knowledge base with cited answers - End-to-end encryption (optional) Call to action: Download. #### Pro — power users (Recommended) **Price TBD / month. Coming Soon — not available yet.** - Everything in Free - Priority access to Pia Cloud models - Expanded token and storage quotas - More knowledge bases and larger document limits --- ## Download Source: https://pia-ai.de/download.html **Download** # Take control of your intelligence. Install Pia in only two minutes. No account. No tracking. No data leaves your device. - Download for Windows (system-wide installer): https://storage.pia-ai.de/f/wpf/Pia-Setup.msi — installs for everyone on the machine, needs admin rights. - Per-user installer: https://storage.pia-ai.de/f/wpf/Pia-Setup-PerUser.msi — installs for the current user, no admin required. - Current version, release date, size and SHA-256: https://storage.pia-ai.de/f/wpf/latest.json - Release notes for the current version (English): https://storage.pia-ai.de/f/wpf/RELEASE-NOTES.md - Audit the code on GitHub: https://github.com/Pia-Ai-dev/Pia.Wpf ### What happens next, after you install 1. **Install Pia.** Run the installer. No account creation, no sign-up forms. 2. **Pick your AI provider.** A guided wizard walks you through connecting OpenAI, Azure, or any compatible provider. 3. **Press your hotkey.** Pia sits in your system tray. One hotkey and you're ready to go. ### Requirements and terms **Windows 10+.** 64-bit Windows 10 or later. .NET 10 runtime included. **No account required.** A guided setup wizard walks you through provider configuration in minutes. Use Pia locally with your own AI keys — cloud sync is optional. **Open source.** Full source code available on GitHub. Build from source if you prefer. --- ## Pia for business Source: https://pia-ai.de/business/ **Pia for business** # Your company already uses AI. Now run it by your rules. Company data is already flowing into consumer chatbots — invisible, unpoliced, unlogged. Pia replaces shadow AI with a better assistant on infrastructure you control. Calls to action: "Talk to us" (kontakt@neo42.de), "See self-hosting" (https://pia-ai.de/business/self-hosting.html). Trust markers: Made in Germany · Open source · No app telemetry · Self-hostable ### The three mechanics (animated hero) 1. **One employee — drafted on their device.** Work starts *on the device* — not in someone else's cloud. 2. **Team sync — on your server.** It syncs to *your* server, with optional end-to-end encryption. 3. **Provider boundary — guardrails decide.** Only what policy allows *crosses the line*. ### Deployment: two ways to run Pia Both keep the same promise: the server that holds your data answers to you — because you run it, or because it can't read what it holds. **Self-host it.** Run the Pia server on your own hardware, in your own network. It comes in two editions on the same image: the free Community Edition — open source, local authentication built in, five users — and the licensed Enterprise Edition, which adds SSO, your own groups, knowledge bases and negotiated seats. See what each edition includes: https://pia-ai.de/business/self-hosting.html#editions - Your hardware, your network, your rules - You choose the AI providers — including local models via Ollama - Free with the Community Edition — licence on request, no payment details **Or use Pia Cloud.** The same server, operated by us, on European providers in EU data centres — no US providers. Turn on end-to-end encryption and synced content is encrypted before it leaves the device — then what you sync stays unreadable to us. Prompts sent to a Pia Cloud model are processed on the server rather than stored; with your own provider they bypass it entirely. - No infrastructure to run - End-to-end encryption (optional) - Start free, upgrade when you need to ### For the people using it: an assistant that doesn't need a mandate Pia wins on its own merits — the governance comes built in. Here's what a rollout actually gives your team. **Company knowledge with citations.** Server-side knowledge bases scoped to groups: your documents in, cited answers out. Feed them from your own pipelines over a REST ingestion API. **Meetings & transcription.** Teams meetings transcribed live with speakers told apart, plus on-device transcription of the room — behind a spoken consent gate. Summaries land where your people can keep them. **Agentic work.** Tools are permissions, schedules deliver answers as chats, and every capability can be revoked. Employees can hand bigger jobs to your Pia server to run in the background — behind an explicit consent screen that states exactly what leaves the device, with caps and a job list. **Managed personas.** Author a persona once — its voice, its knowledge bases, its tools — and publish it to a group. Everyone gets the same reviewed assistant, and every answer shows which persona wrote it. ### Where to go next **Security & compliance.** Guardrails, SSO, audit with anomaly detection, encryption at rest — and precisely what an admin cannot see. Security in detail: https://pia-ai.de/business/security.html **Self-hosting & administration.** Groups as the unit of policy, token usage boards, live-editable runtime policies, devices, licensing — the whole admin plane. Self-hosting in detail: https://pia-ai.de/business/self-hosting.html ### Closing **Talk to us about your rollout.** Deployment, licensing, security reviews — email us and a person answers, in German or English. Email kontakt@neo42.de --- ## Security and compliance Source: https://pia-ai.de/business/security.html **Security & compliance** # Security you can verify. Everything on this page is shipped, documented, and open source. Where a protection has limits, we state them — that's what makes the rest believable. **Guardrails that act before the model answers.** A catalog of guard types, configured per group. Requests are classified on the way in; a risky one is rerouted to a protected model instead of being answered normally. Verdicts are cached so the check stays fast — and every decision lands in a log you can review. - Classified per group - Risky requests rerouted to a protected model - Every verdict logged - Request path: Requests → Guardrail → Requested model or Protected model → Decision log **Sign in with what you already run.** Microsoft Entra ID single sign-on with a fully documented OAuth registration and login flow. SSO comes with the Enterprise Edition; the free Community Edition signs people in locally, with MFA and passkeys. **An audit trail that flags anomalies.** Structured events with filtering and retention — including login-anomaly detection that flags unusual sign-in patterns. Guardrail decisions sit alongside, so one place answers what happened and what the AI layer did about it. **Encryption at rest, with stated limits.** A master key, an AES-GCM wire format, layered disk protection — and documentation that states the threat model, including what it does not protect against. We'd rather you know. **Pia Cloud runs on European providers.** We select the providers behind Pia Cloud to be European companies processing in EU data centres, with a data-processing agreement in place for each. No US providers. That is our procurement policy for Pia Cloud — self-hosting or your own provider key keeps the question in your hands entirely. **What an admin cannot see.** With end-to-end encryption on, the admin console shows what exists — devices, item counts, sync state — while synced content stays unreadable, even to the operator. Prompts sent to a Pia Cloud model are processed on the server rather than stored, and with a self-hosted server or your own provider they do not reach ours at all. That separation is documented and demonstrable: exactly what a works council or data-protection officer asks for. Server-side data — managed personas, device and plugin metadata, trusted certificates — is generated on the server and therefore not end-to-end encrypted. **A lost device is not lost data.** A recovery code lets encrypted data survive a lost or reset device. And the honest flip side: if the code is gone too, the data is unrecoverable — by design, because a backdoor for you would be a backdoor for everyone. **One policy file for the whole fleet.** An enterprise policy file enforces defaults across managed installations — including how the assistant itself behaves — with rollout guidance for IT. **Revoke the laptop, not the account.** Every registered device is visible with its pairing status and sync cursors. A lost machine gets revoked — cut off from sync — without touching the account or the other devices. Delete it entirely when it's truly gone. **The pod uplink trust model.** Pods connect over a single inbound route with a narrow contract: what a pod can and cannot see is documented, and the operator plane is unreachable from chat. ### Closing **Send us your security questions.** Security questionnaires, DPO questions, threat-model deep dives — we answer them ourselves. Talk to us about security: kontakt@neo42.de --- ## Self-hosting Source: https://pia-ai.de/business/self-hosting.html **Self-hosting** # Your server. Your rules. No fine print. The whole admin plane, on hardware you control. Here's what running Pia yourself looks like day to day. **Start free with the Community Edition.** The Community Edition is the real server, not a demo: open source, local authentication built in, and a documented walkthrough from zero to first login. If you can run a standard .NET service and a database, you can run Pia. Read the walkthrough: https://docs.pia-ai.de/wpf/guides/community-edition Every Pia server runs on a signed licence file, including this one — mail us and you get the Community licence for free. On first start the server prints a setup token; paste it together with the licence on the setup page and it activates without a restart. ### Two editions, one server Community and Enterprise are the same container image. The signed licence decides what it unlocks, so moving up later means replacing one file — same server, same data. | Capability | Community | Enterprise | |---|---|---| | Local login with MFA and passkeys | Included | Included | | AI proxy with your own provider keys | Included | Included | | Sync and chat history across devices | Included | Included | | Admin dashboard, token board, runtime policies | Included | Included | | Audit log and login-anomaly detection | Included | Included | | Encryption at rest | Included | Included | | End-to-end encryption, recovery, device revocation | Included | Included | | Plugins, MCP and the public REST API | Included | Included | | Groups as the unit of policy | One default group | Groups you define | | Guardrails with a decision log | Not included | Included | | Microsoft Entra ID single sign-on | Not included | Included | | Managed personas and client policies | Not included | Included | | Knowledge bases with cited answers | Not included | Included | | Users / admins | 5 / 1 | Per contract | Guardrails, quotas, provider routing and plugin allowlists are settings on a group — that is why they sit on the Enterprise side, not because the feature itself is switched off. Everything above the group row works on a Community licence. The Enterprise Edition is not on sale yet. The Community licence is free: mail us and we send you the file. **The group is the unit of policy.** (Enterprise Edition.) Set it once on the group, and every member inherits it: token limits and per-user quotas, knowledge-base quotas, AI provider routing — overall and per mode — guardrail configuration, and the plugin allowlist. Policy stops being tribal knowledge. (A group carries: Quota, Providers, Plugins, Knowledge.) **Account lifecycle without loose ends.** Roles, identity sources, disabling, session revocation, soft delete with recovery. Offboarding is a checklist the server already knows. **See where the tokens go.** Usage over time, broken down by user, group, and model. The heaviest requests, per-user request logs, and which templates drive the spend — before the invoice tells you. **Policies you can edit live.** Rate limits, quotas, identity and lockout policy, AI payload caps, provider retry behaviour — all editable from the admin UI while the server runs. No config-file archaeology, no restart window. Test a provider connection live from the settings page — before your users find out the hard way. **Every client, accounted for.** Registered devices with pairing status and sync cursors. Revoke a lost machine to cut it off from sync, or delete it outright — the difference is deliberate, and documented. **Licence status at a glance.** Edition, expiry, and seat counts at a glance; licence events on a timeline; install or replace a licence from the UI, with an automatic rollback if the new file fails to validate. Expiry is a hard stop, not a slow fade: the server returns to setup mode and waits for a valid licence. Your data is untouched, but nothing else runs — so watch the warnings, which start 30 days out. **Extend it your way.** (Enterprise Edition for the per-group allowlist.) Plugins come in three kinds — server-delivered tool packs, MCP servers you bring, and REST APIs wired in as tools. Which of them a group may use is an Enterprise setting; the catalogue itself works on every edition. Third-party plugins are covered by signing certificates, with the trust list managed by you. **Pia Mesh — included, and off until you switch it on.** Connectors put pods on the tool plane: register a pod, issue its token, pick a transport, grant it to a group, and watch its presence. Operators put them on the task plane: serve skills, supervise running assignments, control retention. Nothing participates until an admin turns it on. **A public API, documented.** Auth, sync, AI proxy, assignments, and the E2EE endpoints are a documented contract — build your own integrations against the same API the client uses. ### Closing **Start on your own server today.** Request the free Community licence (kontakt@neo42.de) or read the walkthrough (https://docs.pia-ai.de/wpf/guides/community-edition). --- ## About Source: https://pia-ai.de/about.html **About Us** # Built by a team that uses AI every day. We're a small team in Germany building an AI assistant we actually want to use ourselves. **Our Story.** We got tired of AI assistants that uses your data as their business model. So we built one that doesn't. Pia started as an internal tool we used every day — now we decided to open-source it. Based in Germany, we believe the best products come from teams that practice what they preach. ### Philosophy: AI-first. In everything we do. We don't just build AI — we work alongside it. Every task starts with one question: can AI do this? **AI evaluates first.** Before assigning a task to a human, we ask: can AI do this? Not to replace people, but to amplify what the team can achieve. **AI as teammates.** Every feature in Pia gets tested in our own daily workflows first. From code review to content creation, AI colleagues are part of our daily routine. **Humans set direction.** People define the vision, values, and strategy. AI handles execution, research, and scaling. Together, we move faster. ### Values: what we stand for **Privacy by default.** We build what we believe in. Privacy isn't a feature — it's a principle. **Transparency.** Open source, honest pricing, no hidden agendas. **Lean & fast.** Small team, big ambitions. We ship fast because AI lets us punch above our weight. Closing: **Try Pia or contribute on GitHub.** --- ## Site-wide ### Navigation Features (/features.html) · Business (/business/) · Pricing (/pricing.html) · About (/about.html) · Download (/download.html) Business sub-navigation: Overview (/business/) · Security & compliance (/business/security.html) · Self-hosting (/business/self-hosting.html) Footer: Business · GitHub (https://github.com/Pia-Ai-dev/Pia.Wpf) · Docs (https://docs.pia-ai.de) · Imprint (/impressum.html, German) · Privacy (/datenschutz.html, German) © 2026 Pia — a product by neo42 GmbH. All rights reserved. ### Privacy of this website This website sets no cookies. Fonts, styles, and analytics are self-hosted — no third-party CDNs. Our analytics (analytics.pia-ai.de) processes your IP address on our own infrastructure. The download page fetches the latest release version from GitHub. ### Languages The site is published in English, German and French, selectable in the header. This Markdown file and https://pia-ai.de/llms.txt are English only.